Understanding how the brain and nervous system develops from a few cells into complex, interconnected networks is a key goal for neuroscientists. Although researchers have identified many of the genes involved in this process, how these work together to form an entire brain remains unknown.

A simple worm called Caenorhabiditis elegans is commonly used to study brain development because it has only about 300 neurons, simplifying the study of its nervous system. The worms are easy to grow in the laboratory and are transparent, allowing scientists to observe how living worms develop using a microscope. Researchers have learned a great deal about the initial growth of the nervous system in C. elegans embryos. However, it has been difficult to study the embryos once their muscles have formed because they constantly twist, fold, and move, making it hard to track the cells.

Now, Christensen, Bokinsky, Santella, Wu et al. have developed a computer program that allows scientists to virtually untwist the embryos and follow the development of the nervous system from its beginning to when the embryo hatches. First, images are taken of worm embryos that produce fluorescent proteins marking certain body parts. The program, with user input, labels the fluorescent cells in the images, which indicates how the embryo is bending and allows the program to straighten the worm. The program can also track how cells move around the embryo during development and show the positional relationships between different cells at different stages of development.

Christensen et al. have made the program freely available for other researchers to use. The next step is to increase automation, making the software faster and more straightforward for users. Ultimately, the software could help in the challenge to comprehensively examine the development of each neuron in the worm.

Understanding how complex neural circuits and entire nervous systems form is one of the fundamental goals of neuroscience. While substantial progress has been made in identifying guidance factors in neurodevelopment (Kolodkin and Tessier-Lavigne, 2011; Dudanova and Klein, 2013; Chilton, 2006; O'Donnell et al., 2009), how known factors interact to direct the formation of complex neural circuits remains mysterious (Dudanova and Klein, 2013). Examining the entirety of neurodevelopment in intact, living samples would be useful in understanding larger scale principles that orchestrate nervous system formation. Unfortunately, technological limitations and inherent nervous system complexity have hindered our ability to capture a 'systems-level' view of the developing brain.

One model organism well-suited to systems-level neuroscience research is Caenorhabditis elegans, which possesses a simple nervous system comprising 302 neurons (White et al., 1986), 222 of which form during embryogenesis (Sulston et al., 1983). The adult connectome has been reconstructed, and the morphology of all adult neurons has been mapped at electron-microscopy resolution (White et al., 1986); the genome sequenced (C. elegans Sequence Consortium, 1998); and the organism is genetically tractable and transparent at all life stages, enabling investigation with light microscopy. The simplicity of the C. elegans nervous system, its experimental accessibility, and the extensive knowledge base make it a promising candidate for following the development of all neurons in the embryo, and eventually understanding associated molecular mechanisms. The resulting 'neurodevelopmental atlas' would represent the first view of how an entire nervous system forms.

Despite the potential of the nematode as a model, imaging neurodevelopment (Wu et al., 2013a) throughout embryogenesis is challenging due to embryonic sensitivity to photodamage and photobleaching, limiting imaging to several hours on most systems; the need for subcellular spatial resolution due to the small size of the embryo; and motion blur caused by rapid embryo movement after muscular twitching begins. Once images are captured, data analysis poses new problems: while it would be easy to assemble an atlas of neuronal positions and morphology if all cells were easily identifiable in one animal, techniques that allow imaging with single-cell contrast (such as Brainbow [Livet et al., 2007]) are unavailable in the nematode. Currently, any attempt to build a neurodevelopmental atlas would require imaging small numbers of non-overlapping, easily distinguishable neurons, and finding methods to combine the data from multiple embryos into a composite whole. To our knowledge, comprehensive solutions to these problems do not yet exist.

Recent advances in light-sheet fluorescence microscopy (LSFM [Santi, 2011]) have solved many of the imaging problems outlined above. LSFM sweeps a thin sheet of light through the sample, relying on perpendicular detection of fluorescence. This geometry allows far more rapid imaging and reduced phototoxicity relative to confocal microscopy (Huisken et al., 2004; Holekamp et al., 2008), enabling the use of LSFM in a variety of transformative applications. These include recording whole-brain calcium signaling in larval zebrafish (Ahrens et al., 2013; Ito, 2013), and imaging (Wu et al., 2011; Keller et al., 2008) and tracking (Amat et al., 2014; Bao, 2006; Santella et al., 2010) large numbers of cells in developing embryos. Multiple LSFM implementations now obviate the problems of motion blur and photo damage in worm embryos (Wu et al., 2011; Wu et al., 2013; Kumar et al., 2014; 2015), and also offer sufficient spatiotemporal resolution (sub-μm in all three spatial dimensions, sub-second volumetric imaging [Wu et al., 2013; Kumar et al., 2014]) that subcellular morphology may be observed over the entire 14-hour period of embryogenesis. Despite these advances, morphological changes still pose problems when trying to follow individual cells, or when combining data from multiple embryos.

To address these problems, we have generated a nematode strain that expresses fluorescent markers within specific cells, and designed software that uses these markers to computationally 'untwist' the embryo, resulting in straightened volumes that significantly ease the tracking of developmental events in later embryonic stages (described briefly in a preliminary conference proceeding [Christensen, 2015]). Our open-source software is based on the NIH’s Medical Image Processing, Analysis, and Visualization (MIPAV [McAuliffe, 2001; Haak et al., 2015]) platform, implemented as a standalone plugin (http://mipav.cit.nih.gov/plugin_jws/mipav_worm_plugin.php). Computational untwisting algorithms have previously been used to straighten images of L1 larval worms for use in tracking nuclear position (Peng et al., 2008; Long et al., 2009) in both two and three dimensions, but to our knowledge, these algorithms are not suitable for the nematode embryo. In addition to the untwisting capability, our plugin includes the ability to annotate and track 3D positions over time, allowing semi-automated quantification of cell and neurite positions in twisted (and untwisted) embryos. The positional data so derived also facilitate comparison and combination of information from multiple embryos, allowing us to create a composite model of development.

We demonstrate the capabilities of our method by computationally untwisting eight nematode embryos; tracking the position of seam cell nuclei, the canal-associated neuron (CAN), ALA, and AIY neuron cell bodies, and the growing neurites of the ALA neuron in the untwisted reference frame; and combining the data from multiple embryos to model the time-evolution of all these elements within the elongating embryo. We find that seam cell nuclear positions are highly stereotyped across different embryos, while the rate of elongation varies according to position along the embryo. Of the neurons we examined, ALA and AIY move in concert with neighboring seam cell nuclei, suggesting they are passively 'dragged' with the rest of the elongating worm embryo, while the CAN neurons actively migrate at a faster rate than the surrounding seam cell nuclei. Tracking ALA neurites reveals that anterior-posterior neurite outgrowth starts toward the end of elongation and continues after cells reach their final positions. Our method is the first to track cell positions in the context of the entire embryo, from the beginning of twitching until hatching. We anticipate that our software will significantly further the ability to examine C. elegans development in the post-twitching regime and lay a foundation for understanding the formation of the C. elegans nervous system.

In order to computationally straighten an embryo, an essential first step is defining limits of the growing worm body, thus specifying how the embryo folds inside the eggshell. Nematode embryos undergo both bending and helical twisting around the nose-to-tail axis (Figure 1—figure supplement 1) posing challenges in untwisting the embryo relative to larval or adult nematodes. Our approach uses fluorescent markers driven by cell-specific promoters to define the boundaries of the worm body. We use a seam cell marker (SCM::GFP) to label the 20–22 seam cell nuclei, identifying the left and right sides of the worm; and a dlg-1::GFP fusion protein to label apical gut junctions and hypodermal junctions, revealing the locations of the anterior tip of the pharynx (hereafter referred to as the nose), tail, midline, and hypodermal cell boundaries (Figure 1A,B). This combination of markers allows automated segmentation of seam cells and manual identification of the nose, tail, and sides of the worm, thus enabling us to model the twisted, bent embryo within the eggshell, and serving as a basis for computationally untwisting the worm (Figure 1C).

Figure 1 with 4 supplements see all

Download asset Open asset

We used a dual-view selective plane illumination microscopy (diSPIM) implementation of LSFM to capture images of developing embryos (Wu et al., 2013; Kumar et al., 2014). The diSPIM was chosen due to the combination of high-imaging speed and isotropic resolution that it provides, making the identification of cells and cellular structures in a twisted-up embryo significantly easier than with lower resolution alternatives (such as single-view light-sheet microscopy, Figure 1—figure supplement 2). After images are acquired in the diSPIM, registered, and deconvolved, a user begins untwisting by downloading and running our software (http://mipav.cit.nih.gov/plugin_jws/mipav_worm_plugin.php, Supplementary file 1).

First, seam cell nuclei are automatically detected, segmented, and paired to create candidate lattices. Seam cell segmentation and lattice-building are manually verified by a user, who can also incorporate additional information derived from pharyngeal and hypodermal markers, which are difficult to automatically segment (Figure 1D,E). Several possible lattices are generated, and the five most likely to be correct are displayed to the user for selection and editing of the correct lattice. The resulting lattice is used to generate a 3D model of the worm volume (Figure 1F, Video 1). In cases where automated lattice-building fails, lattices can be built manually by marking the positions of seam cell nuclei, nose, bends in the embryo, and tail. When manually building lattices, minimally 22 +2B lattice points are recommended (22 is the number of lattice points corresponding to seam cell nuclei, plus a pair of points to mark the nose, and B is the number of bends between seam cell nuclei in the embryo). Fewer lattice points than the number of seam cell nuclei gives unphysical, short volumes, and more than ~32 points does not noticeably improve quality in the untwisted volumes (Figure 1—figure supplement 3).

Video 1

Download asset

The first step in creating the 3D model is to generate curves defining the center and sides of the worm. The centerline curve is uniformly sampled to generate a series of planes extending along and normal to the curve, while avoiding overlap within the model. This series of restricted planes comprises the worm model and is updated as new lattice points are added. To generate a straightened volume, the voxels in the original image that intersect with the sampling planes in the worm model are captured, and the sampling planes and associated voxels are concatenated in the head-to-tail direction to generate a straightened volume (Figure 1G, Video 1). The same process can be used to straighten images of older animals (such as L2 larvae, Figure 1—figure supplement 4). More details are provided in Supplementary file 1 and Appendix.

In addition to untwisting, it is also useful to obtain the 3D position of a cell or point of interest within the nematode embryo. Thus, our software also includes an annotation capability, allowing a user to define points within the embryo for which they would like to obtain 3D coordinates both before and after untwisting (Supplementary file 1). The user adds annotations similarly to lattice points, marking the volume location where the desired point should appear. The user must also add an origin point from which the relative 3D position of all other points is calculated from. As pharyngeal labeling is consistent and bright in most diSPIM volumes, we use the nose as a standard origin in all datasets described in the paper. Once the origin and annotation points have been defined, the user can untwist the worm and obtain the 3D coordinates of each annotation point in a spreadsheet file.

In order to ensure that our algorithm did not alter the distance between portions of the embryo during the untwisting process, we compared the apparent 3D distance between, or along, landmark features within twisted and untwisted embryo volumes (embryos 1–6, Figure 2). First, we determined the distances between nuclei in seam cell pairs (Figure 2A,B). If untwisting did not effect morphology, we reasoned that these distance should be conserved regardless of whether the embryo is twisted or untwisted. We measured the difference between pair distance in twisted- and untwisted datasets at every fifth or tenth time point for both the first (H0) and last (T) pairs of seam cells in six different embryos, reasoning that the difference should be close to 0. The apparent untwisted distance between seam cell pairs H0 and T closely mirrored the values in the twisted worm, with the population average difference across timepoints and embryos (<μ_{Difference, time}>_embryo ± population standard deviation <σ_{Difference, time}>_embryo) for H0 0.4 µm ± 0.3 µm, and for T 0.3 ± 0.2 µm (Figure 2C, Figure 2—figure supplement 1, Supplementary file 5, Materials and methods). The largest difference at any individual timepoint between twisted and untwisted values was 1.7 µm for H0 and 1.2 µm for T.

Figure 2 with 1 supplement see all

Download asset Open asset

Since the model of the twisted embryo is based on positional coordinates of the seam cell nuclei, we would expect these paired distances in twisted- and untwisted embryos to agree. For a more stringent control, we also assessed the apparent distance between nose and the pharynx-gut transition (effectively the pharyngeal contour length) in twisted and untwisted embryos (Figure 2A,B). Although the pharynx is not used as a landmark for defining the worm model used in untwisting, we still expect its contour length to be conserved despite untwisting. Here, too, we measured a close correspondence (typically less than 5% of the total pharyngeal length). The population <μ_{Difference, time}>_embryo ± <σ_{Difference, time}>_embryo between twisted and untwisted pharyngeal lengths was 2.5 µm ± 1.6 µm (with the maximum difference between the untwisted and twisted values for any individual timepoint being 8.8 µm, Figure 2C, Figure 2—figure supplement 1). We conclude that our untwisting procedure accurately captures distances present in the twisted embryo.

The combination of untwisting and annotation capabilities we developed allows the analysis of overall morphological changes in a developing embryo and the precise tracking of positions for individual cells or subcellular structures. We first examined overall morphological changes in the nematode embryo. Embryos lengthened (from 86 ± 5 µm at early 1.5-fold, measured from the nose to the tail, to 162 ± 19 µm within the last 30 min before hatching, measured from the nose to the last pair of seam cells, mean ± standard deviation [SD], 5 embryos) and narrowed in width (measured diameter across the widest cross-section 22 ± 1 µm at early timepoints, and 16 ± 1 µm at late timepoints, mean ± SD, 5 embryos) as they progressed from comma stage to late-3 fold (Figure 3A–H, Figure 3—figure supplement 1; Figure 3—figure supplement 2; Video 2). We used our software to manually annotate and extract the positional trajectories of seam cell nuclei during this time period, as they moved relative to the nose of the animal (Figure 3J–3L, Figure 3—figure supplement 1). We note that seam cell V5 divides late in the threefold embryo into Q and V5 daughters; in such cases, we tracked the anteriormost daughter, Q, and thus refer to V5 as Q/V5 in our paper. The motion of seam cell nuclei followed relatively simple trends that were easily evident, despite the noise present in the raw untwisted trajectories. During elongation, seam cell nuclei moved laterally (‘X’ motion, Figure 3J) towards the midline, while maintaining a relatively fixed dorso-ventral position (‘Y’ motion, Figure 3K). Along the axial, head-to-tail axis, the displacement of seam cell nuclei was biphasic, showing a fast, approximately linear dependence on time, followed by a slower plateau (‘Z’ motion, Figure 3L) (Priess and Hirsh, 1986; Chin-Sang and Chisholm, 2000; Ding et al., 2004; Norman and Moerman, 2002). While embryo elongation has been examined before (Priess and Hirsh, 1986), our method is the first that allows 3D interrogation of whole, live, untwisted nematodes at arbitrary timepoints in embryogenesis (Figure 3, Video 2).

Figure 3 with 3 supplements see all

Download asset Open asset

Video 2

Download asset

The strong qualitative similarities in seam cell nuclear trajectories among the five embryos we inspected led us to investigate whether data from different embryos could be combined to yield a composite model of development representing growth in an 'average' embryo. Initial examination of the axial (nose-to-tail) seam cell nuclear trajectories from different embryos suggested a high degree of stereotypy; except for a relative shift in time, the trajectories displayed very similar shapes (Figure 4A). We thus shifted the axial data in time until the trajectories from multiple embryos overlaid (Figure 4B, Figure 4—figure supplement 1). We determined the amount of shift by using a three parameter logistic function to fit the raw axial displacement data (Figure 4—figure supplement 2, Tables 1, 2, Materials and methods), overlaying the data from various embryos until the inflection points in each curve were identical.

Figure 4 with 4 supplements see all

Download asset Open asset

We applied the same time shift to the X- and Y- seam cell nuclear coordinates, finding that seam cell nuclear positions followed similar trajectories throughout elongation (average SD calculated across all 20 seam cell nuclei and all timepoints, <<σ_X>_time>_{seam cell} 0.8 µm, <<σ_Y>_time>_{seam cell} 0.7 µm, <<σ_Z>_time>_{seam cell} 4.6 µm, see also Figure 4—figure supplement 3, Table 1, and 'Materials and methods'). After shifting, we averaged (Figure 4C) and fitted (Figure 4D, Figure 4—figure supplement 4, Table 1, Supplementary file 2) the embryo XYZ trajectories, thus generating positions representing the noise-free time evolution of seam cell nuclei. We note that the choice of fitting functions is somewhat arbitrary. For axial positions, the growth that we and others (Priess and Hirsh, 1986) have observed leads to a sigmoidal fitting function. Amongst the various three-parameter sigmoidal functions (Table 2), we found that the three-parameter logistic function gave the best qualitative and quantitative (Figure 4—figure supplement 2) agreement with the data. We fitted lateral (‘X’) seam cell nuclei positions with a two parameter power law function, and dorso-ventral (‘Y’) positions with a linear function, as empirically these functions described our data well. Despite the ad hoc nature of these fits, we found that fitted values were within 1.5 µm of the X, Y averaged data, and within 7.5 µm of the Z averaged data (Supplementary file 3). For reference, the total length of the untwisted embryo at the final time point was 162.0 ± 18.7 µm (mean ± SD, 5 embryos), measured from the nose to the last pair of seam cells, and the corresponding diameter at the last time point 16.1 ± 1.3 µm, measured at the widest cross-section in the animal.

The averaged, fitted seam cell nuclei data allowed us to inspect the relative relationships among seam cell nuclei in an elongating embryo (Figure 5A, Videos 3,4). Since we fixed the nose as the stationary origin in our untwisting procedure, this location does not move in 4D representations of the fitted embryo. In this 'nose-centric' reference frame, points further from the origin also appear to move faster and farther than points closer to the origin. To better understand the growth rates of individual seam cell nuclei in relation to their neighbors, and the overall length changes within the elongating embryo in a frame-independent manner, we also computed the differences in position between adjacent pairs of nearest-neighbor seam cell nuclei over time (Figure 5B–D, Figure 5—figure supplement 1). In ‘X’ and ‘Y’ dimensions, seam cell nuclei exhibited similar movement patterns, remaining largely stationary in ‘Y’ (Figure 5—figure supplement 1), and moving inwards (toward the origin) in ‘X’ (Figure 5—figure supplement 1) at similar rates. In contrast, seam cell nuclei movement along the ‘Z’ direction was more heterogeneous. For example, the distance between the origin and nuclei of seam cell pair H0, measured from the fitted data, changed from 2.4 µm to 23.8 µm over elongation (Figure 5B), while the distance between seam cell nuclear pairs V6 and T remained essentially constant, at 22.5 µm (Figure 5D). Thus, the rate of increase in distance between the origin and H0 was significantly greater than the increase in distance between V6 and T, over the same period. Other adjacent nuclear pairs separated at roughly similar rates from start to end of elongation (these pairs increased in distance 6.8 ± 2.8 µm, mean ± SD from 7 adjacent pairs of seam cell nuclei, again derived from the fitted data in Figure 5C). These trends were not the results of artifacts in our fitting procedure, as they were evident also in the raw, averaged data (compare left and right graphs in Figure 5B-D). The apparent differences in X- and Z- pre- and post-elongating seam cell nuclei positions that we observe are consistent with the asymmetric morphology of the pre-elongating embryo. Since the embryo starts out in a tadpole-like shape with the head larger than the tail, the seam cell nuclei in the head must move a greater distance than the nuclei in the tail to achieve a uniform diameter in the elongated embryo.

Video 3

Download asset

Video 4

Download asset

Figure 5 with 1 supplement see all

Download asset Open asset

Embryo elongation is thought to be dependent on an actin-based contractile mechanism (Priess and Hirsh, 1986). The complex, position-dependent motion we observed is likely inconsistent with a simple, uniform contraction, as this phenomena cannot explain our finding that different regions of the embryo elongate at markedly different rates. To our knowledge, current models of embryo elongation have not taken into account the differential elongation we observed across the worm body. We expect that incorporating additional data derived from cell positions and subcellular markers (especially cytoskeletal [Priess and Hirsh, 1986; Gally et al., 2009]) in the embryo would help further refine existing models (Ciarletta et al., 2009) of embryo elongation.

Currently, building a composite model of neuronal positions and morphological development in the embryo depends on pooling distinct datasets from many independent embryos. Given our experience tracking seam cell nuclei, we next turned our attention to modeling the 4D motion of neurons and neurite outgrowth in the elongating worm embryo as a proof of concept for a neurodevelopmental atlas (Figure 6A, Videos 5,6). Four of the five embryos used in constructing our seam cell model also had neuronal cell bodies marked with a pceh-10::GFP construct; neurons included AIYL/R, CANL/R, and ALA. We manually annotated the position of these neuronal cell bodies, then temporally aligned, averaged, and fitted the positions as we did the seam cell nuclei (Figure 4). The axial motion of these neurons was qualitatively similar to the seam cell nuclei and could be well described by the three parameter logistic function. However, their XY motion appeared different than the seam cell nuclei. For example, the lateral motion of CANs could not be easily described by a simple function, so we used a 50 point smoothing of the averaged data as our 'fit'. The ALA X motion was better described by a 4^th degree polynomial than a power law, so we used the former function to fit the data (Figure 4—figure supplement 4, Table 1). As evident by their axial displacements, ALA and AIYL/R moved similarly to nearby seam cell nuclei (Figure 6B). In contrast, CANs moved faster than adjacent seam cell nuclei, suggesting a more 'active' mode of migration (Figure 6C). Finally, the motion of ALA and CANs (especially CANL) were considerably more variable between datasets than the seam cell nuclei (Figure 4—figure supplement 3, Supplementary files 2,3,4). While it is currently unclear whether this variability is strain-dependent or reflects underlying biology, this observation underscores the need to study multiple embryos and assess the degree to which cellular motion is stereotyped in elongating embryos.

Video 5

Download asset

Video 6

Download asset

To examine neurite outgrowth clearly, we created a two-color strain with GFP-labeled untwisting markers and a pceh-10::mCh construct to label neuronal cell bodies and neurites. We observed substantial mosaicism in terms of which cells were labeled from one embryo to the next with this strain. Although neurons were labeled with an extrachromosomal array and a certain degree of mosaicism could be anticipated, labeling differences from one animal to the next hindered our ability to track both ALA and CAN outgrowths. Nevertheless, we were able to obtain two datasets where the ALA neuron was labeled throughout most of our imaging period.

ALA is a single neuron with a cell body located in the dorsal portion of the head; a pair of long neurites extend ventrally from this cell body into the nerve ring, and then turn and extend posteriorly along the lateral nerve cord (White et al., 1986). Left and right ALA outgrowths could be readily identified and annotated in both twisted and untwisted embryos (Figure 6D, Figure 6—figure supplement 1). In modeling the left and right neurite shapes, we simplified them by annotating them as three distinct points (ALA: cell body; AxR1 or AxL1: point at which the neurite turns to extend posteriorly; AxR2/L2: neurite terminus; Figure 6E). We then measured the 3D displacements (relative to the nose, as before) of each independent point, shifting, averaging and fitting the data derived from two embryos (as outlined in Figure 4 and illustrated in Figure 6—figure supplement 1), to yield a noise-free representation of the neurite (Table 1). Aligning the fitted neurites to our reference embryo allowed inspection of ALA neurite growth in the context of the elongating embryo (Figure 6A, Videos 5,6), revealing that neurite outgrowth continued to occur for ~240 min after the other cells assumed their final positions at the end of elongation. We also segmented growing ALA neurites at several points in development to demonstrate that straightened images can be used to generate volumetric reconstructions of cell morphology throughout development (Figure 6—figure supplement 2.) We are unaware of any other work that has modeled the growth and positions of neurites in the post-twitching embryonic regime (for C. elegans or any other model organism).

Figure 6 with 2 supplements see all

Download asset Open asset

The C. elegans cell lineage is invariant (Sulston et al., 1983), and tracking cells in the L1 larva has revealed that cellular positions in post-hatching animals are relatively stereotyped (Long et al., 2009). Our work suggests that this positional stereotypy extends to the cells in the late embryo as well. However, we also found that in the case of cells or structures which actively migrate, such as the CAN neurons and ALA neurites, there seems to be greater variability in terms of end position and growth rate. To some extent this is not surprising; as these cells and neurites move longer distances than most other cells, and depend on actively finding their way in a complex environment (as opposed to passive movement in response to elongation), there may be more room for variability in how they travel and reach their destinations.

On a more general level, we also observed variability in the temporal shifts necessary to align each elongating embryo to the reference dataset (embryo 1). Some of this variability may be due to relatively mundane explanations: embryos were at slightly different ages when imaging began, and temperature was moderately controlled (to within 2 to 3°C both during imaging and strain growth). Intrinsic developmental variability, caused by maternal effects or exposure to imaging could also have played a role in the slightly different patterns of development we observed across embryos.

Expanding the work we describe here to other migrating and non-migrating neurons should make clear whether there actually is a difference in positional variability between migrating and non-migrating cells. Adding additional data to our 4D model is conceptually straightforward: strains with distinguishable neurons can be crossed into the untwisting background, untwisted, trajectories of cells and outgrowths fitted, and subsequently registered with previously derived data. 'Filling in' the positions of all neurons and outgrowths in the developing embryo would form the basis of the 4D atlas of neurodevelopment, and could be combined with functional activity mapping and gene expression data to provide a more comprehensive picture of animal development in late embryogenesis.

Our untwisting and annotation plugin is designed to be flexible, so that it can be applied to most problems involving tracking position and morphology of distinguishable structures in the nematode embryo. The core of the plugin relies on defining the sides of the worm embryo; although our work uses a specific set of markers, we note that any other markers which define the edges of the worm body should also work. The annotation capability is also flexible; as it is based on manual annotation, almost any distinct structure can be annotated. Finally, while the isotropic resolution of the diSPIM is very helpful in resolving fine embryonic detail (Figure 1—figure supplement 2), our untwisting algorithm is compatible with other high-resolution imaging methods. For example, we used a super-resolution two-photon instant structured illumination microscope (2P ISIM) (Winter, 2014) to image and untwist a bent L2 larval worm, obtaining clear images of this relatively large specimen (Figure 1—figure supplement 4, Video 7). Our plugin is designed specifically for untwisting nematode embryos, and as such is unlikely to be immediately applicable to other biological systems without substantial modification (we know of few non-nematode systems that have the same vermiform shape and degree of twisting and movement). However, some of the more general concepts we implement, such as the benefit of aligning and pooling information derived different datasets to generate an overall 4D view of development, are likely applicable to more systems than just the worm.

Video 7

Download asset

Despite the power of our semi-automated approach, several areas for improvement remain. Automated lattice-building assumes the embryo has 20-22 seam cell nuclei on which the lattice is based; in early periods of elongation (especially the 1.5- to 2-fold transition) expression is absent in some seam cell nuclei, requiring manual lattice-building. In addition, time spent in editing automatic segmentation and lattice generation results in ~8 hr of manual work when untwisting an embryo spanning 100–150 timepoints. Fully automated untwisting is not currently feasible, but the development of alternative markers may enable this goal. Second, although the positions of cells and neurites in the growing embryo can be determined with micron-scale precision, and placed in context with their neighbors, additional methods are needed to place the full morphological volume of a given cell within the untwisted embryo. While our data are of sufficient quality to segment such morphology in an untwisted animal (Figure 6—figure supplement 2, Video 8), the general question about how to combine morphological segmentations from distinct, untwisted embryos remains. New methods developed for pre-twitching embryos may prove useful in this regard (Santella et al., 2015).

Video 8

Download asset

A more significant and long-term set of technical problems for completing the neurodevelopmental atlas relates to the generation of fluorescent markers and strains that provide sparse, optically resolvable neurons. Most fluorescent strains label multiple neurons that are too close in space and time to be easily resolved – possible strategies to 'separate' these neurons might include 'Brainbow' (Livet et al., 2007) (spectral separation of densely labeled neurons) or heat-shock-based approaches (Halfon et al., 1997; Bacaj and Shaham, 2007) (temporal separation of densely labeled cells). Even if such strains are built, the identity of the resulting neurons will need to be verified. As lineaging (Bao, 2006) in C. elegans has been carried out to just before twitching begins (Giurumescu et al., 2012), in principle neurons can be identified by matching early expression to lineage data. If expression turns on after twitching, lineaging would also need to be extended into the post-twitching regime. Such 'deep lineaging', or tracking the coordinates of all nuclei through twitching would be a valuable and complementary effort to untwisting. Finally, we note that the expression pattern of fluorescent proteins within individual neurons could be further optimized. For almost all strains (except DCR4209 which contained membrane-targeted mCherry), fluorescent proteins were expressed cytoplasmically. An improved strategy would combine such cytoplasmic labeling with membrane targeting, better filling out very thin neuronal outgrowths that otherwise might be missed due to low expression; a similar strategy was adopted in super-resolution microscopy to trace thin neurites (Lakadamyali et al., 2012).

1. 1. Ahrens MB
   2. Orger MB
   3. Robson DN
   4. Li JM
   5. Keller PJ

   (2013) Whole-brain functional imaging at cellular resolution using light-sheet microscopy

   Nature Methods 10:413–420.

   https://doi.org/10.1038/nmeth.2434

   • PubMed
   • Google Scholar
2. 1. Amat F
   2. Lemon W
   3. Mossing DP
   4. McDole K
   5. Wan Y
   6. Branson K
   7. Myers EW
   8. Keller PJ

   (2014) Fast, accurate reconstruction of cell lineages from large-scale fluorescence microscopy data

   Nature Methods 11:951–958.

   https://doi.org/10.1038/nmeth.3036

   • PubMed
   • Google Scholar
3. 1. Bacaj T
   2. Shaham S

   (2007) Temporal control of cell-specific transgene expression in caenorhabditis elegans

   Genetics 176:2651–2655.

   https://doi.org/10.1534/genetics.107.074369

   • PubMed
   • Google Scholar
4. 1. Bao Z
   2. Murray JI
   3. Boyle T
   4. Ooi SL
   5. Sandel MJ
   6. Waterston RH

   (2006) Automated cell lineage tracing in caenorhabditis elegans

   Proceedings of the National Academy of Sciences of the United States of America 103:2707–2712.

   https://doi.org/10.1073/pnas.0511111103

   • Google Scholar
5. 1. Bao Z
   2. Murray JI

   (2010)

   Imaging in Developmental Biology: A Laboratory Manual, New York, Cold Spring Harbor Laboratory Press

   Imaging in Developmental Biology: A Laboratory Manual, New York, Cold Spring Harbor Laboratory Press.

   • Google Scholar
6. 1. C. elegans Sequencing Consortium

   (1998) Genome sequence of the nematode c. elegans: a platform for investigating biology

   Science 282:2012–2018.

   https://doi.org/10.1126/science.282.5396.2012

   • PubMed
   • Google Scholar
7. 1. Chilton JK

   (2006) Molecular mechanisms of axon guidance

   Developmental Biology 292:13–24.

   https://doi.org/10.1016/j.ydbio.2005.12.048

   • PubMed
   • Google Scholar
8. 1. Chin-Sang I
   2. Chisholm AD

   (2000) Form of the worm: genetics of epidermal morphogenesis in c. elegans

   Trends in Genetics 16:544–551.

   https://doi.org/10.1016/S0168-9525(00)02143-0

   • Google Scholar
9. 1. Christensen R
   2. Bokinsky A
   3. Santella A
   4. Wu Y
   5. Marquina J
   6. Kovacevic I
   7. Kumar A
   8. Winter P
   9. McCreedy E
   10. Mohler W
   11. Bao Z
   12. Colón-Ramos D
   13. Shroff H

   (2015) An imaging and analysis toolset for the study of caenorhabditis elegans neurodevelopment

    SPIE BIOS 9334:93340C–93349.

   https://doi.org/10.1117/12.2082394

   • Google Scholar
10. 1. Ciarletta P
    2. Ben Amar M
    3. Labouesse M

    (2009) Continuum model of epithelial morphogenesis during caenorhabditis elegans embryonic elongation

    Philosophical Transactions of the Royal Society A 367:3379–3400.

    https://doi.org/10.1098/rsta.2009.0088

    • Google Scholar
11. 1. Ding M
    2. Woo W
    3. Chisholm A

    (2004) The cytoskeleton and epidermal morphogenesis in

    Experimental Cell Research 301:84–90.

    https://doi.org/10.1016/j.yexcr.2004.08.017

    • Google Scholar
12. 1. Dudanova I
    2. Klein R

    (2013) Integration of guidance cues: parallel signaling and crosstalk

    Trends in Neurosciences 36:295–304.

    https://doi.org/10.1016/j.tins.2013.01.007

    • PubMed
    • Google Scholar
13. 1. Gally C
    2. Wissler F
    3. Zahreddine H
    4. Quintin S
    5. Landmann F
    6. Labouesse M

    (2009) Myosin II regulation during c. elegans embryonic elongation: LET-502/ROCK, MRCK-1 and PAK-1, three kinases with different roles

    Development 136:3109–3119.

    https://doi.org/10.1242/dev.039412

    • PubMed
    • Google Scholar
14. 1. Giurumescu CA
    2. Kang S
    3. Planchon TA
    4. Betzig E
    5. Bloomekatz J
    6. Yelon D
    7. Cosman P
    8. Chisholm AD

    (2012) Quantitative semi-automated analysis of morphogenesis with single-cell resolution in complex embryos

    Development 139:4271–4279.

    https://doi.org/10.1242/dev.086256

    • PubMed
    • Google Scholar
15. 1. Haak D
    2. Page C-E
    3. Deserno TM

    (2015) A survey of DICOM viewer software to integrate clinical research and medical imaging

    Journal of Digital Imaging.

    https://doi.org/10.1007/s10278-015-9833-1

    • Google Scholar
16. 1. Halfon MS
    2. Kose H
    3. Chiba A
    4. Keshishian H

    (1997) Targeted gene expression without a tissue-specific promoter: creating mosaic embryos using laser-induced single-cell heat shock

    Proceedings of the National Academy of Sciences of the United States of America 94:6255–6260.

    https://doi.org/10.1073/pnas.94.12.6255

    • PubMed
    • Google Scholar
17. 1. Holekamp TF
    2. Turaga D
    3. Holy TE

    (2008) Fast three-dimensional fluorescence imaging of activity in neural populations by objective-coupled planar illumination microscopy

    Neuron 57:661–672.

    https://doi.org/10.1016/j.neuron.2008.01.011

    • PubMed
    • Google Scholar
18. 1. Huisken J
    2. Swoger J
    3. Del Bene F
    4. Wittbrodt J
    5. Stelzer EH

    (2004) Optical sectioning deep inside live embryos by selective plane illumination microscopy

    Science 305:1007–1009.

    https://doi.org/10.1126/science.1100035

    • PubMed
    • Google Scholar
19. 1. Ito M

    (2013) Error detection and representation in the olivo-cerebellar system

    Frontiers in Neural Circuits 7:1–11.

    https://doi.org/10.3389/fncir.2013.00001

    • PubMed
    • Google Scholar
20. 1. Keller PJ
    2. Schmidt AD
    3. Wittbrodt J
    4. Stelzer EH

    (2008) Reconstruction of zebrafish early embryonic development by scanned light sheet microscopy

    Science 322:1065–1069.

    https://doi.org/10.1126/science.1162493

    • PubMed
    • Google Scholar
21. 1. Koh K
    2. Rothman JH

    (2001)

    ELT-5 and ELT-6 are required continuously to regulate epidermal seam cell differentiation and cell fusion in C. elegans

    Development 128:2867–2880.

    • PubMed
    • Google Scholar
22. 1. Kolodkin AL
    2. Tessier-Lavigne M

    (2011) Mechanisms and molecules of neuronal wiring: a primer

    Cold Spring Harbor Perspectives in Biology 3:a001727.

    https://doi.org/10.1101/cshperspect.a001727

    • PubMed
    • Google Scholar
23. 1. Kumar A
    2. Colón-Ramos DA
    3. Shroff H

    (2015) Watching a roundworm develop with a sheet of light

    Physics Today 68:58–59.

    https://doi.org/10.1063/PT.3.2856

    • Google Scholar
24. 1. Kumar A
    2. Wu Y
    3. Christensen R
    4. Chandris P
    5. Gandler W
    6. McCreedy E
    7. Bokinsky A
    8. Colón-Ramos DA
    9. Bao Z
    10. McAuliffe M
    11. Rondeau G
    12. Shroff H

    (2014) Dual-view plane illumination microscopy for rapid and spatially isotropic imaging

    Nature Protocols 9:2555–2573.

    https://doi.org/10.1038/nprot.2014.172

    • PubMed
    • Google Scholar
25. 1. Lakadamyali M
    2. Babcock H
    3. Bates M
    4. Zhuang X
    5. Lichtman J

    (2012) 3D multicolor super-resolution imaging offers improved accuracy in neuron tracing

    PloS One 7:e30826.

    https://doi.org/10.1371/journal.pone.0030826

    • PubMed
    • Google Scholar
26. 1. Livet J
    2. Weissman TA
    3. Kang H
    4. Draft RW
    5. Lu J
    6. Bennis RA
    7. Sanes JR
    8. Lichtman JW

    (2007) Transgenic strategies for combinatorial expression of fluorescent proteins in the nervous system

    Nature 450:56–62.

    https://doi.org/10.1038/nature06293

    • PubMed
    • Google Scholar
27. 1. Long F
    2. Peng H
    3. Liu X
    4. Kim SK
    5. Myers E

    (2009) A 3D digital atlas of C. elegans and its application to single-cell analyses

    Nature Methods 6:667–672.

    https://doi.org/10.1038/nmeth.1366

    • PubMed
    • Google Scholar
28. 1. McAuliffe M
    2. Lolande F

    (2001)

    Medical image processing, analysis, and visualization in clinical research

    Proc. 14th IEEE Symp. Computer-Based Medical Systems pp. 381–386.

    • Google Scholar
29. 1. Norman KR
    2. Moerman DG

    (2002) Alpha spectrin is essential for morphogenesis and body wall muscle formation in caenorhabditis elegans

    The Journal of Cell Biology 157:665–677.

    https://doi.org/10.1083/jcb.200111051

    • PubMed
    • Google Scholar
30. 1. O'Donnell M
    2. Chance RK
    3. Bashaw GJ

    (2009) Axon growth and guidance: receptor regulation and signal transduction

    Annual Review of Neuroscience 32:383–412.

    https://doi.org/10.1146/annurev.neuro.051508.135614

    • Google Scholar
31. 1. Peng H
    2. Long F
    3. Liu X
    4. Kim SK
    5. Myers EW

    (2008) Straightening caenorhabditis elegans images

    Bioinformatics 24:234–242.

    https://doi.org/10.1093/bioinformatics/btm569

    • PubMed
    • Google Scholar
32. 1. Priess JR
    2. Hirsh DI

    (1986) Caenorhabditis elegans morphogenesis: the role of the cytoskeleton in elongation of the embryo

    Developmental Biology 117:156–173.

    https://doi.org/10.1016/0012-1606(86)90358-1

    • PubMed
    • Google Scholar
33. 1. Santella A
    2. Catena R
    3. Kovacevic I
    4. Shah P
    5. Yu Z
    6. Marquina-Solis J
    7. Kumar A
    8. Wu Y
    9. Schaff J
    10. Colón-Ramos D
    11. Shroff H
    12. Mohler WA
    13. Bao Z

    (2015) WormGUIDES: an interactive single cell developmental atlas and tool for collaborative multidimensional data exploration

    BMC Bioinformatics 16:189.

    https://doi.org/10.1186/s12859-015-0627-8

    • PubMed
    • Google Scholar
34. 1. Santella A
    2. Du Z
    3. Nowotschin S
    4. Hadjantonakis AK
    5. Bao Z

    (2010) A hybrid blob-slice model for accurate and efficient detection of fluorescence labeled nuclei in 3D

    BMC Bioinformatics 11:580.

    https://doi.org/10.1186/1471-2105-11-580

    • PubMed
    • Google Scholar
35. 1. Santi PA

    (2011) Light sheet fluorescence microscopy: a review

    The Journal of Histochemistry and Cytochemistry  59:129–138.

    https://doi.org/10.1369/0022155410394857

    • PubMed
    • Google Scholar
36. 1. Sulston JE
    2. Schierenberg E
    3. White JG
    4. Thomson JN

    (1983) The embryonic cell lineage of the nematode caenorhabditis elegans

    Developmental Biology 100:64–119.

    https://doi.org/10.1016/0012-1606(83)90201-4

    • PubMed
    • Google Scholar
37. 1. Terns RM
    2. Kroll-Conner P
    3. Zhu J
    4. Chung S
    5. Rothman JH

    (1997)

    A deficiency screen for zygotic loci required for establishment and patterning of the epidermis in caenorhabditis elegans

    Genetics 146:185–206.

    • PubMed
    • Google Scholar
38. Website

    1. Thomson N

    (2014) EM image 2-fold_993, from 2-fold image set

    http://www.wormimage.org
39. 1. Totong R
    2. Achilleos A
    3. Nance J

    (2007) PAR-6 is required for junction formation but not apicobasal polarization in c. elegans embryonic epithelial cells

    Development 134:1259–1268.

    https://doi.org/10.1242/dev.02833

    • PubMed
    • Google Scholar
40. 1. Wenick AS
    2. Hobert O

    (2004) Genomic cis-regulatory architecture and trans-acting regulators of a single interneuron-specific gene battery in c. elegans

    Developmental Cell 6:757–770.

    https://doi.org/10.1016/j.devcel.2004.05.004

    • PubMed
    • Google Scholar
41. 1. White JG
    2. Southgate E
    3. Thomson JN
    4. Brenner S

    (1986) The structure of the nervous system of the nematode caenorhabditis elegans

    Philosophical Transactions of the Royal Society B 314:1–340.

    https://doi.org/10.1098/rstb.1986.0056

    • Google Scholar
42. 1. Winter PW
    2. York AG
    3. Nogare DD
    4. Ingaramo M
    5. Christensen R
    6. Chitnis A
    7. Patterson GH
    8. Shroff H

    (2014) Two-photon instant structured illumination microscopy improves the depth penetration of super-resolution imaging in thick scattering samples

    Optica 1:181–191.

    https://doi.org/10.1364/OPTICA.1.000181

    • PubMed
    • Google Scholar
43. 1. Wu Y
    2. Christensen R
    3. Colón-Ramos D
    4. Shroff H

    (2013a) Advanced optical imaging techniques for neurodevelopment

    Current Opinion in Neurobiology 23:1090–1097.

    https://doi.org/10.1016/j.conb.2013.06.008

    • PubMed
    • Google Scholar
44. 1. Wu Y
    2. Ghitani A
    3. Christensen R
    4. Santella A
    5. Du Z
    6. Rondeau G
    7. Bao Z
    8. Colon-Ramos D
    9. Shroff H

    (2011) Inverted selective plane illumination microscopy (iSPIM) enables coupled cell identity lineaging and neurodevelopmental imaging in caenorhabditis elegans

    Proceedings of the National Academy of Sciences of the United States of America 108:17708–17713.

    https://doi.org/10.1073/pnas.1108494108

    • Google Scholar
45. 1. Wu Y
    2. Wawrzusin P
    3. Senseney J
    4. Fischer RS
    5. Christensen R
    6. Santella A
    7. York AG
    8. Winter PW
    9. Waterman CM
    10. Bao Z
    11. Colón-Ramos DA
    12. McAuliffe M
    13. Shroff H

    (2013b) Spatially isotropic four-dimensional imaging with dual-view plane illumination microscopy

    Nature Biotechnology 31:1032–1038.

    https://doi.org/10.1038/nbt.2713

    • PubMed
    • Google Scholar

Author details

1. Ryan Patrick Christensen

   Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States

   Contribution

   RPC, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Alexandra Bokinsky, Anthony Santella and Yicong Wu

   For correspondence

   ryan.christensen@nih.gov

   Competing interests

   The authors declare that no competing interests exist.

2. Alexandra Bokinsky

   Biomedical Imaging Research Services Section, Center for Information Technology, National Institutes of Health, Bethesda, United States

   Contribution

   AB, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Ryan Patrick Christensen, Anthony Santella and Yicong Wu

   Competing interests

   The authors declare that no competing interests exist.

3. Anthony Santella

   Developmental Biology Program, Sloan-Kettering Institute, New York, United States

   Contribution

   AS, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Ryan Patrick Christensen, Alexandra Bokinsky and Yicong Wu

   Competing interests

   The authors declare that no competing interests exist.

4. Yicong Wu

   Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States

   Contribution

   YW, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Ryan Patrick Christensen, Alexandra Bokinsky and Anthony Santella

   Competing interests

   The authors declare that no competing interests exist.

5. Javier Marquina-Solis

   Program in Cellular Neuroscience, Neurodegeneration and Repair, Department of Cell Biology, Yale University School of Medicine, New Haven, United States

   Contribution

   JM-S, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

6. Min Guo

   1. Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States
   2. State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, Zhejiang University, Hangzhou, China

   Contribution

   MG, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

7. Ismar Kovacevic

   Developmental Biology Program, Sloan-Kettering Institute, New York, United States

   Contribution

   IK, Conception and design, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

8. Abhishek Kumar

   1. Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States
   2. Program in Cellular Neuroscience, Neurodegeneration and Repair, Department of Cell Biology, Yale University School of Medicine, New Haven, United States

   Contribution

   AK, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

9. Peter W Winter

   Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States

   Contribution

   PWW, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

10. Nicole Tashakkori

    Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States

    Contribution

    NT, Analysis and interpretation of data, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

11. Evan McCreedy

    Biomedical Imaging Research Services Section, Center for Information Technology, National Institutes of Health, Bethesda, United States

    Contribution

    EM, Conception and design, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

12. Huafeng Liu

    State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, Zhejiang University, Hangzhou, China

    Contribution

    HL, Conception and design, Analysis and interpretation of data

    Competing interests

    The authors declare that no competing interests exist.

13. Matthew McAuliffe

    Biomedical Imaging Research Services Section, Center for Information Technology, National Institutes of Health, Bethesda, United States

    Contribution

    MM, Conception and design, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

14. William Mohler

    Department of Genetics and Developmental Biology, University of Connecticut Health Center, Farmington, United States

    Contribution

    WM, Conception and design, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

15. Daniel A Colón-Ramos

    Program in Cellular Neuroscience, Neurodegeneration and Repair, Department of Cell Biology, Yale University School of Medicine, New Haven, United States

    Contribution

    DAC-R, Conception and design, Analysis and interpretation of data, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

16. Zhirong Bao

    Developmental Biology Program, Sloan-Kettering Institute, New York, United States

    Contribution

    ZB, Conception and design, Analysis and interpretation of data, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

17. Hari Shroff

    Section on High Resolution Optical Imaging, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, United States

    Contribution

    HS, Conception and design, Analysis and interpretation of data, Drafting or revising the article

    Competing interests

    The authors declare that no competing interests exist.

• 4,618

  views

• 747

  downloads

• 31

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.