-
[
Elife,
2022]
Analyses across imaging modalities allow the integration of complementary spatiotemporal information about brain development, structure and function. However, systematic atlasing across modalities is limited by challenges to effective image alignment. We combine highly spatially resolved electron microscopy (EM) and highly temporally resolved time-lapse fluorescence microscopy (FM) to examine the emergence of a complex nervous system in C. elegans embryogenesis. We generate an EM time series at four classic developmental stages and create a landmark-based co-optimization algorithm for cross-modality image alignment, which handles developmental heterochrony among datasets to achieve accurate single-cell level alignment. Synthesis based on the EM series and time-lapse FM series carrying different cell-specific markers reveals critical dynamic behaviors across scales of identifiable individual cells in the emergence of the primary neuropil, the nerve ring, as well as a major sensory organ, the amphid. Our study paves the way for systematic cross-modality data synthesis in C. elegans and demonstrates a powerful approach that may be applied broadly.
-
Wu, Yicong, Shroff, Hari, Catena, Raul, Kovacevic, Ismar, Christensen, Ryan, Marquina-Solis, Javier, Santella, Anthony, Mohler, William A., Kumar, Abhishek, Bao, Zhirong, Moyle, Mark, Colon-Ramos, Daniel
[
International Worm Meeting,
2015]
WormGUIDES is a 4D interactive atlas of C. elegans embryogenesis. Its purpose is to support the exploration and analysis of embryogenesis at the molecular, cellular, tissue and organism levels. The current WormGUIDES release contains a minute-by-minute record of nuclear positions for all cells until twitching. Current efforts focus on adding records of 3D cell morphology to reconstruct neural morphogenesis and the dynamics of neurite outgrowth. Our strategy consists of acquiring 3D time-lapse images of embryogenesis using promoter-driven fluorescent markers to sparsely label subsets of neurons. Ubiquitous histone markers enable automated lineaging to identify cells and align datasets acquired from different embryos into a digital composite record.Data from our initial set of markers reveals dynamic processes that contribute to the architecture of the major neural organs such as the nerve ring, ventral nerve cord and the head nerve bundles. Much of this morphogenesis occurs before twitching. Surprisingly, some morphogenetic modules are formed in neural progenitors and maintained through two cell cycles. Perturbations suggest novel collective cell behaviors and unexpected roles for surrounding tissue in shaping the nervous system.Lineage-based cell identification has yielded a list of approximately 50 neurons that are currently being tracked and segmented. These cells account for about a quarter of embryonic neurons and include critical components of the major neural structures. To support the throughput needed for systematic reconstruction, we have developed computational pipelines for semi-automated segmentation and alignment of cell shapes from multiple embryos and markers. In addition, we have developed computational tools to untwist the elongating embryo and trace post-twitching events. We are making additional markers, and developing a desktop application extending the current mobile app. Ultimately, the goal is to produce a complete, dynamic record of the assembly of the C. elegans connectome.
-
[
BMC Bioinformatics,
2014]
BACKGROUND: Advances in fluorescence labeling and imaging have made it possible to acquire in vivo records of complex biological processes. Analysis has lagged behind acquisition in part because of the difficulty and computational expense of accurate cell tracking. In vivo analysis requires, at minimum, tracking hundreds of cells over hundreds of time points in complex three dimensional environments. We address this challenge with a computational framework capable of efficiently and accurately tracing entire cell lineages. RESULTS: The bulk of the tracking problem-tracking cells during interphase-is straightforward and can be executed with simple and fast methods. Difficult cases originate from detection errors and relatively rare large motions. Therefore, our method focuses computational effort on difficult cases identified by local increases in cell number. We force these cases into tentative cell track bifurcations, which define natural semi-local neighborhoods that permit Bayesian judgment about the underlying cell behavior. The bifurcation judgment process not only correctly tracks through cell divisions and large movements, but also offers corrections to detection errors. We demonstrate that this method enables large scale analysis of Caenorhabditis elegans development, an ideal validation platform because of an invariant cell lineage. CONCLUSION: The high accuracy achieved by our method suggests that a bifurcation-based semi-local neighborhood provides sufficient information to recognize dependencies between nearby tracking choices, and to interpret difficult tracking cases without reverting to global optimization. Our method makes large amounts of lineage data accessible and opens the door to new types of statistical analysis of complex in vivo processes.
-
Bao, Zhirong, Santella, Anthony, Yu, Zidong, Shroff, Hari, Wu, Yicong, Du, Zhuo
[
International Worm Meeting,
2013]
The invariant lineage has been a cornerstone of C. elegans biology. John Sulston's initial lineage 30 years ago used multiple embryos to assemble the invariant pattern. Complete continuous lineaging has never been performed for a single animal. While image analysis software has facilitated lineage creation during early embryogenesis, embryo movement at later stages has hampered the analysis of later development. We have made progress towards lineage tracing during this developmental period through a combination of innovative computational and imaging methods. Our efforts on image analysis are focused on the challenge of reliably following small and crowded nuclei over long periods of time in under sampled images. Our nuclear detection method uses per slice segmentation and a learned shape model to robustly detect and segment nuclei in crowded configurations. The detection results are merged into a cell lineage using multiple linking steps, which select from a set of possible causes and actions, including cell movements, divisions, or detection errors. This is based on probabilistic models of nuclear appearance and local spatial configuration. These computational improvements are bolstered by a qualitative improvement in image quality through the dual-view inverted Selective Plane Illumination Microscope (diSPIM), which captures isotropically sampled volumes at speeds that largely eliminate motion artifacts even through the final stage of embryogenesis. We anticipate that our combined effort will allow not just lineage tracing, but that the detailed dynamics of development (including cell positions and expression patterns) can be followed at high spatiotemporal resolution to build a quantitative model of development. A long term application is the production of WormGUIDES, a 4D atlas tracking both cell positions and neuronal outgrowth.
-
[
International Worm Meeting,
2019]
Though automation makes EM data more accessible, navigation and identification of structures remains a bottleneck for interpretation and analysis. To assist with this problem we present a robust, general method for assigning single cell identities from a template to a sample in the presence of both differences in the cells present and changes in spatial configuration, validating this in the context of the C. elegans embryo. We introduce relative neighbor graph constraints to model the invariant spatial structure of a labeled samples and use this to assign a global quality score to an answer based on its internal consistency with expected cell-cell contacts. This score is used in a novel gradient descent optimization of the template sample which removes cells whose presence cause neighbor constraint violations and are therefore hypothesized to be missing in the unlabeled sample. Our final answer is produced by an instance-based learning like approach where the sample is independently matched against each example in the ensemble of reference data sets and a consensus identity is assigned. We apply this method to identifying all cells in Electron Micrographs of two C. elegans embryos at ~300min and 350min p.f.c. Imaging with Focused Ion Beam SEM and a serial array method provides an undistorted image of the worm simplifying the problem of alignment. Time lapse fluorescence microscopy provides the reference atlas data set of cell positions. Three embryos, underlying data for the WormGUIDES atlas, have identities established by lineaging between the four-cell stage and twitching. This data and our method were used to assign cell identities to all nuclei in the two unknown data sets. To validate automated results cell identities were independently and manually assigned to a subset of cells in the two EM data sets based on position and cell morphology. Predicted identities were 88% and 91% accurate in the early and late data sets on the cases confirmed by morphology suggesting cell identities can be reliably assigned based on position alone. *Irina Kolotueva co-first author and co-PI
-
[
BMC Bioinformatics,
2010]
BACKGROUND: To exploit the flood of data from advances in high throughput imaging of optically sectioned nuclei, image analysis methods need to correctly detect thousands of nuclei, ideally in real time. Variability in nuclear appearance and undersampled volumetric data make this a challenge. RESULTS: We present a novel 3D nuclear identification method, which subdivides the problem, first segmenting nuclear slices within each 2D image plane, then using a shape model to assemble these slices into 3D nuclei. This hybrid 2D/3D approach allows accurate accounting for nuclear shape but exploits the clear 2D nuclear boundaries that are present in sectional slices to avoid the computational burden of fitting a complex shape model to volume data. When tested over C. elegans, Drosophila, zebrafish and mouse data, our method yielded 0 to 3.7% error, up to six times more accurate as well as being 30 times faster than published performances. We demonstrate our method's potential by reconstructing the morphogenesis of the C. elegans pharynx. This is an important and much studied developmental process that could not previously be followed at this single cell level of detail. CONCLUSIONS: Because our approach is specialized for the characteristics of optically sectioned nuclear images, it can achieve superior accuracy in significantly less time than other approaches. Both of these characteristics are necessary for practical analysis of overwhelmingly large data sets where processing must be scalable to hundreds of thousands of cells and where the time cost of manual error correction makes it impossible to use data with high error rates. Our approach is fast, accurate, available as open source software and its learned shape model is easy to retrain. As our pharynx development example shows, these characteristics make single cell analysis relatively easy and will enable novel experimental methods utilizing complex data sets.
-
[
International Worm Meeting,
2021]
Advances in Electron Microscopy (EM) bring about the possibility of temporal analysis using staged samples over time as well as offering opportunities to gain insights by synthesizing EM with other imaging modalities. We present a pseudo time series of C. elegans embryonic development, four volumes covering around 2 hours particularly rich in development between 320 and 475 minutes post first cleavage. This period encompasses neurulation, neural organogenesis and neuropil formation key events that build most of the major structures of the nervous system as well as critical events for many other organ systems. We correlate this EM data with florescence data spanning the first eight hours of embryogenesis. Every cell in the florescence data is identifiable via lineaging. To correlate these data sets we develop a novel computational method for alignment of identities between data sets in the challenging presence of spatial and temporal variation. This approach involves co-optimization of spatial alignment and the structure of labeled data based on a model of dynamic anatomy in the form of an adjacency graph with expected variation. This model captures both variable elements and consistent spatial proximity relationships. We identify every cell in three of the four time points. Identity results are accurate, ranging from 72 to 78 percent correct when assessed against a large set of manual annotations based on position and morphology. This is better than any previously reported results for identifying all cells in an organism based on position alone. The resulting single cell level annotation allows efficient navigation of this large EM data set. We use the sequence to probe the interactions over time between different components within the nerve ring elucidating the relationship between the temporal and spatial location of initial outgrowths into the ring and ultimate structure. We also examine the interaction between cells during the formation of the amphid dendrite structure providing insight into the timing and succession of events at the inter and intra cellular level. Our observations only scratch the surface of the details available in the data set. We will provide the EM data with single-cell level annotation as a resource for the community.
-
[
International Worm Meeting,
2011]
The study of embryonic phenotypes at high detail and large scale requires accurate and fast assembly of detected nuclei into a cell lineage. Accuracy is essential. Every error limits analysis, requiring either manual correction or a retreat from the goal of single cell tracking. In addition, high throughput work limits the effort that can be spent on correction and quality control of results. The time it would take to confirm, unguided, the correctness of even a perfectly accurate cell lineage may be unsupportable. As such, a measure of the local reliability of results is just as important as low error. Existing methods do not to provide this combination of features. Common, simple methods such as nearest neighbor association across time are error prone and lack a model that can provide confidence in results. More sophisticated general tracking methods such as particle filters lack an explicit model of cell division, a key source of ambiguity and error. These requirements lead to an approach that scores lineage configurations against a learned model. The key design decision is what aspects of the lineage, as revealed via imaging and cell detection, to include in the model used to judge alternative tree configurations. Our desire is for a general method applicable to any embryo capable of normal cell division. As such, we model key local behaviors of nuclei, firstly their individual change over time (in appearance and position), and secondly the correlation in these measures expected between daughter cells at division. We also model two key aspects of the detection method that provides the raw nuclear positions for lineage assembly, firstly the relationship between nuclear appearance and the probability of a detection being a false positive and secondly the probability of a hypothesized series of detection failures. This model, which can be learned from a set of corrected lineages, is relatively simple but sufficient to produce accurate results and to highlight ambiguous areas for human inspection.
-
Mohler, William, Shroff, Hari, Kumar, Abhishek, Katzman, Braden, Nguyen, Nhan, Barnes, Kris, Sengupta, Titas, Bao, Zhirong, Christensen, Ryan, Duncan, Leighton, Santella, Anthony, Duncan, William, Bosque, Gabriela, Moyle, Mark, Fan, Li, Shah, Pavak, Harvey, Brandon, Ikegami, Richard, Colon-Ramos, Daniel, Tang, Doris
[
International Worm Meeting,
2017]
WormGUIDES is an interactive 4D atlas of C. elegans embryogenesis. Its goals are to (1) provide a model of neural development based on detailed time lapse measurements of nuclear positions and neurite outgrowth; (2) cross reference worm community data with the 4D model and (3) provide an easy to use visualization platform for exploring, understanding and annotating the model and sharing insights. The major tracts of the adult nervous system are laid out early and added to over time. By the 1.5 fold stage many major structures are established. The nerve ring (NR) forms a complete loop that includes dorsal and ventral cells. Sensory nerves have extended to the dent, and the amphid commissure is established. Motor neurons in the VNC have intercalated and others have extended into the VNC and toward the NR. The early emergence of tracts motivates a hierarchal approach to modeling and measuring neural development. Our model contains three levels of structure: (1) Tracts, major nerve tracts; (2) Multi-cellular structures, small groups of co-labeled neurons; and (3) Individual cells. The latter two include cell bodies as well as fascicules or individual neurites. Neurites are threaded through the tracts based on measured lengths and tip positions to minimize noise in alignment and maximize legibility. The current model contains 9 tracts representing the amphid sensory nerves, amphid commissures, NR, VNC and connections, 21 neurons (5 single cell) and nuclear positions up to twitching. We have mapped 6 groups of neurites in the NR with stereotypical positions involving 30 neurons, and sorted the 38 ventral-going amphid commissure axons into 4 temporal groups. 182 markers slated for analysis cover almost all neurons. Stochastic labeling by heat shock, mosaicism of reporter arrays and single cell photo conversion are being pursued to distinguish intertwined neurons. Additional imaging and analysis tools are being developed to push our model to hatching. The wormguides.org website provides comprehensive information on the project, access to our reagents and image data, and a download of the latest atlas. We accept nomination of markers and cells as priorities, and promote sharing of user-driven annotations of developmental processes.
-
[
International Worm Meeting,
2009]
Reliable, automated cell detection is a critical component of automated biological image analysis. Accuracy is particularly important for efficient, high-throughput cell lineaging. When tracing a C. elegans lineage overall detection error rates as low as one or two percent result in lineages that can require hours of hand editing to correct. To achieve more reliable automation we are investigating image processing approaches tailored to the appearance of nuclei in confocal fluorescence images. Our approach uses a Difference of Gaussians blob detector to guide an efficient extraction of the nuclear boundary as a set of disks. Preliminary results show a significant improvement in detection rates over Starrynite, our existing detection and lineaging system. Total detection error during the 9th stage of cell division drops from 4.5 to 1.1 %, overall error through the 9th round is reduced from 1.5 to .78%. The method is both accurate and fast enough to be used in real time imaging applications. Results suggest that with further development on extraction and tracking, editing will cease to be a bottleneck. This will enable automated lineaging through the previously daunting 10th and final round of cell divisions and make possible uniquely detailed, large-scale studies of embryonic development.