The imaging platform “The Lifespan Machine” automates the lifelong observation of large populations. We show the steps required to perform lifespan, stress resistance, pathogenesis, and behavioral aging assays. The quality and scope of the data allow researchers to study interventions in aging despite the presence of biological and environmental variation.
Genetically identical animals kept in a constant environment display a wide distribution of lifespans, reflecting a large non-genetic, stochastic aspect to aging conserved across all organisms studied. This stochastic component means that in order to understand aging and identify successful interventions that extend the lifespan or improve health, researchers must monitor large populations of experimental animals simultaneously. Traditional manual death scoring limits the throughput and scale required for large-scale hypothesis testing, leading to the development of automated methods for high-throughput lifespan assays. The Lifespan Machine (LSM) is a high-throughput imaging platform that combines modified flatbed scanners with custom image processing and data validation software for the life-long tracking of nematodes. The platform constitutes a major technical advance by generating highly temporally resolved lifespan data from large populations of animals at an unprecedented scale and at a statistical precision and accuracy equal to manual assays performed by experienced researchers. Recently, the LSM has been further developed to quantify the behavioral and morphological changes observed during aging and relate them to lifespan. Here, we describe how to plan, run, and analyze an automated lifespan experiment using the LSM. We further highlight the critical steps required for the successful collection of behavioral data and high-quality survival curves.
Aging is a complex, multifaceted process characterized by a decline in the physiological function of an organism, which leads to an increase in the risk of disease and death over time1. Lifespan, measured as the time from birth or the onset of adulthood until death, provides an unambiguous outcome of aging2 and an indirect but rigorously quantitative proxy for measuring the relative rate of aging between populations3. Aging studies often depend on accurate measurements of lifespan, similar to clinical trials, to compare outcomes between one population exposed to an intervention and an unexposed control group. Unfortunately, reproducibility issues pervade aging research, sometimes due to statistically underpowered experiments4 and often because of the inherent sensitivity of lifespan assays to subtle variations in the environment5. Robust experiments require multiple replicates of large populations, and this process particularly benefits from the experimental scalability offered by automation6.
The rigorous demands of lifespan assays originate from the unpredictability of the aging process itself. Isogenic individuals housed in identical environments display different death times and rates of physiological decline7, suggesting that lifespan involves a high degree of stochasticity7,8. Therefore, large populations are required to measure quantitative changes in the aging process, such as changes in the mean or maximum lifespan, and to overcome biases arising from individual variability. In addition, a capacity for high-throughput lifespan assays is crucial to support studies of survival curve shapes and models of the dynamics of aging9.
The nematode Caenorhabditis elegans is an invaluable model for aging research due to its short lifespan, genetic tractability, and rapid generation time, which underscore its suitability for high-throughput aging and lifespan assays. Traditionally, the lifespan in C. elegans has been measured by following a synchronized, small population of about 50-100 animals over time on solid media and writing down the time of individual deaths. As animals age and lose mobility, manually scoring the death times requires individually prodding the animals and checking for small movements of the head or tail. This is usually a tedious and laborious process, though efforts have been made to accelerate it10,11,12. Importantly, slow experimental pipelines hinder progress in our understanding of aging and the effectiveness of tested interventions.
To meet the demands of aging research for quantitative data, many technologies have been developed for automating data collection, including a remarkable range of approaches from microfluidic chambers to flatbed scanners13,14,15,16,17,18. The LSM differs from other methods in its extensive optimization for the collection of highly precise and accurate lifespan data, which is achieved through the development of careful equipment calibration protocols combined with an extensive software suite that allows users to validate, correct, and refine automated analyses13. Though the software can, in principle, be applied to diverse imaging modalities, in practice, most users use flatbed scanners modified to allow for fine-tuned control over the environmental temperature and humidity – factors of critical importance due to their major effect on lifespan19. The LSM takes images of nematodes every 20 min over intervals ranging from days to months, depending on the environmental conditions and genotype. The data produced are of much higher temporal resolution compared to data from manual assays, and the images collected provide a permanent visual record of the nematode position across the lifespan. Using machine-learning methods, death times are automatically assigned to each individual. These results can be rapidly, manually validated using a client software called "Worm Browser". As a result of its hardware and software, the LSM can generate survival curves that are statistically indistinguishable from manual death scoring at the hands of experienced researchers, with the added advantage of decreased workload and higher scalability13.
The latest version of the LSM also allows for the study of behavioral aging by collecting morphological and behavioral data throughout the nematode's life and reporting it along with the lifespan of each individual. In particular, the LSM captures the time of each animal's vigorous movement cessation (VMC), a landmark often used to quantify the "healthspan" of an individual as distinct from its lifespan. By simultaneously collecting lifespan and behavioral aging data, the LSM supports the study of interventions that may have differential effects on different phenotypic outcomes of aging20. A variety of macroscopically observable phenotypes can be used to study behavioral aging, such as body movement or pharyngeal pumping21, tissue integrity22, and movement speed or stimulus-induced turning17. Comparisons between different aging phenotypes can support analyses of the causal structure of aging processes. For instance, the comparison between VMC and lifespan was recently used to characterize two distinct aging processes in C. elegans23.
While initially developed to measure lifespan in C. elegans, the LSM supports the collection of survival and behavioral data from a range of nematode species, including C. briggsae, C. tropicalis, C. japonica, C. brenneri, and P. pacificus23. The technology facilitates the study of the effect of biological and environmental interventions on lifespan, stress resistance, and pathogen resistance and can be coupled to experimental tools such as targeted assays of RNA interference or auxin-inducible protein degradation systems. To date, it has been used in the scientific literature for a wide range of applications6,24,25,26,27,28,29,30.
Here, we outline a step-by-step protocol for performing a Lifespan Machine experiment using agar plates, from the initial stages of the experimental setup to the output of the resulting survival curves. A distinctive feature of the LSM is that the effort is highly front-loaded, meaning that the majority of the researcher's time is spent during experimental setup and, to a small degree, during post-image acquisition. The data collection is completely automated for the whole duration of the experiment and allows the researcher to have a "hands-free" experience. The steps described here are held in common among many different types of survival assays – the same experimental setup is performed for lifespan, thermotolerance, oxidative stress, and pathogenesis assays. In the representative results section, we discuss a subset of data from a recently published manuscript to illustrate the effectiveness of the analysis pipeline and highlight the most important steps during image analysis23.
1. Software and hardware requirements
Supplementary Figure 1: Lifespan Machine hardware. One flatbed scanner unit with an open lid to show the loaded plates, which are placed facing down into 16 openings cut on a rubber mat. The rubber mat is placed on the surface of a glass scanner. Labels for the conditions are written on the sides of the plates to avoid issues during image analysis. Marking tape with the number ("1") and/or name of the device ("Jabba") facilitates later verification of the sample location when working with multiple scanner devices. More details about the LSM hardware components are found elsewhere13. Please click here to download this File.
2. Setup prior to the day of the experiment
3. Setup on the day of the experiment
4. Pre-image acquisition
NOTE: A comprehensive flowchart summarizing all the software-based steps during image acquisition is depicted in Figure 1.
Figure 1: Graphical overview of the Lifespan Machine image analysis pipeline. The pre-, during, and post-image acquisition steps are largely performed on the web interface (WI, in red) and on the Worm Browser (WB, in green). Some steps are performed in other platforms (O, in blue), such as TXT documents in step 3a, Photoshop or equivalent in step 4b, and JMP or equivalent in step 13. Please click here to view a larger version of this figure.
Figure 2: Preview capture image and scan area selection. (A) For each scanner in the experiment, a preview capture image is generated. (B) Selection of one row of plates at a time (red boxes), which increases the speed of scanning and prevents worm motion blur as a result of scanning areas that are too wide. Please click here to view a larger version of this figure.
5. Image acquisition
NOTE: The following steps can be performed both while the experiment is running or after it has finished.
Figure 3: Specification of the plate locations for each scanner using sample masks. To ensure the independent analysis of plates within the column selections shown in Figure 1, individual plates must be identified by generating an image mask composite. (A) A capture of the scans of the scanners is opened with an image manipulation software (note the name of the scanner "han" above a scanned selection, and "a-d" referring to each of the columns). (B) The individual steps of mask generation to mark the location of each plate in the mask composite require the background to be set to black, (C) the removal of jagged edges and edges of non-selected plates by the expanding and then shrinking of the background, and (D) selecting the foreground plates and filling the areas entirely with white pixels. (E) For the LSM to recognize individual plates in the scanned rows, each white region in a row is filled with a different shade of gray, usually in increasing brightness. (F) At this stage, the mask is saved (LZW compression with no layers specified if generated in Photoshop). The mask is then scanned by the Worm Browser, and a visualization of the mask by the software is generated. A correct mask visualization should display one defined square per plate with a small cross at the center and a different color for each row. Please click here to view a larger version of this figure.
Figure 4: Plate quality control using the web interface. Censoring of suboptimal plates on the web interface before the worm movement analysis is crucial for speeding up the image processing pipeline. Examples of plates subject to removal include conditions of (A) desiccation, (B) contamination, or (C) fogging, as opposed. (D) Optimal plates to be included in further analysis. A scale bar of 10 mm is superimposed onto a preview capture image. Please click here to view a larger version of this figure.
6. Post-image acquisition
NOTE: After worm detection is completed, all data collected from the experiment must be aggregated over time to track each individual across their lifespan and identify all the individuals' death times. Wait until all animals in the experiment have died and until all the worm detection jobs have been completed, and then perform the following steps:
Figure 5: Animal storyboard on the Worm Browser. (A) All stationary worms are shown in chronological order of machine-annotated death time. To navigate the storyboard, press the buttons on the (B) bottom-right corner, and (C) save the annotations often. (D) The images with a non-gray background depict two worm death events (early death as green, later death as red), which can either occur when two worms die close to each other, or when dead worms are moved by a passing worm and are, thus, detected as dead twice. (E) A red tag in the bottom corner of an image identifies worms with a detected death time; (F) a green tag indicates where an object did not remain still long enough to record a death time. (G) Multiple worms in the same frame can be flagged by pressing shift and left-clicking. (H) Non-worm objects are excluded from the analysis by a right-click.(I) Exploded worms are censored from the analysis by clicking on the corresponding image (a by-hand annotation window opens) and pressing shift and right-clicking until an "animal exploded" message appears. A scale bar of 0.5 mm and labels are superimposed on the screenshot of a Worm Browser window. Please click here to view a larger version of this figure.
Figure 6: Inspecting objects and annotation of death times on the Worm Browser. Left-clicking on any object on the Worm Browser storyboard opens a new interface and allows the user to inspect the object's movement dynamics. On the right side, the (A) movement score is displayed, which quantifies the object movement; this is estimated by the change in pixel intensities between consecutive observations. Additionally, on the right side, (B) the change in the total object intensity is displayed, which quantifies changes in the object size. On the left side, the upper bar shows the (C) machine estimate of the death time, while the bottom bar is the (D) human by-hand annotation. Clicking on any point of the bars and pressing the space key allows the user to move through the time frames in which the worm has been imaged. On these bars, pink represents the time spent in vigorous movement, red represents the time spent in death, and yellow is everything in between. The time spent in expansion and contraction after the death time is shown in green. Labels are superimposed on the screenshot of a Worm Browser window. Please click here to view a larger version of this figure.
Figure 7: Population summary statistics on the Worm Browser. Population statistics for the scanner device "obiwan", with a plot of the survival (left panel) and a scatter plot of the vigorous movement cessation (VMC) time versus the death time (right panel). The plotted are details of (A) one condition, obtained from (B) one scanner achieved by first selecting (C) the survival grouping by strain. (D) The square shapes in the scatter plot depict the by-hand annotated events, while (E) the circular shapes depict the machine-annotated events. (F) By-hand annotation is often required for death events that occur early or (G) those where the time of vigorous movement cessation time coincides with the death time. Labels are superimposed on the screenshot of a Worm Browser window. Please click here to view a larger version of this figure.
Experimental reproducibility in lifespan assays is challenging and requires both tightly controlled experimental conditions and large populations to achieve sufficient statistical resolution4,36. The LSM is uniquely suitable for surveying large populations of animals in a constant environment with high temporal resolution. To demonstrate the capability of the LSM, highlight the crucial steps of analysis, and help researchers to prioritize their labor efforts, we present a subset of data from an optimal, previously published experiment23. Due to the nature of this experiment as a dose-response assay, a substantial number of animals were required for reliably detecting lifespan alterations. By hand, this experiment would require an intense time commitment from multiple researchers or, if scaled down, lead to underpowered results. The dataset measured quantitative changes in lifespan after removing the RNA polymerase II (b) subunit gene (RPB-2), required for messenger RNA transcription. Using the auxin-inducible system37 in C. elegans, the endogenous locus of RPB-2 was tagged with a degron sequence to conditionally ubiquitylate and degrade RPB-2 using different concentrations of auxin (α-naphtaleneacetic acid). The experiment was performed on AMP100 [rpb–2 (cer135); eft-3p::TIR-1] with cer135 corresponding to the CRISPR-inserted AID tag rpb-2::GFPΔpiRNA::AID::3xFLAG23. The experimental conditions were at a temperature of 20 °C and using UV-inactivated NEC937 (OP50 ΔuvrA; KanR)38 E. coli. Hermaphrodites were sterilized by transferring nematodes during the late L4 stage to plates containing 5-fluoro-2-deoxyuridine (FUdR). Nematodes were transferred during day 2 of adulthood to plates containing different concentrations of auxin. In the original study, the authors showed that in the presence of auxin, RPB-2 degradation shortened lifespan in a dose-dependent manner23.
Here, we demonstrate the contribution that post-experiment data validation and annotation make to the final survival curve output (Figure 8). In the last steps of image analysis within the Worm Browser storyboard, we compared raw survival curves produced before storyboard annotation to the survival curves produced after the manual exclusion of non-worm objects and also to the survival curves produced after the annotation of both non-worm objects and individual death times (Figure 8A–C). We found that the initial survival curves produced before storyboard annotation (Figure 8A) were distorted by the improper inclusion of non-worm objects, which was most apparent in the tails of the survival curves. Before manual annotation, the survival curves included all objects detected by the machine as potential worm objects, approximately half of which, due to the intentionally high false positive rate, were non-worm objects and had to be manually excluded (Figure 8D). This is by design, as the algorithms are calibrated to have a relatively high false-positive rate in order to avoid false negatives, as it is much easier to exclude incorrectly included objects than to search for and recover incorrectly excluded objects. After excluding non-worm objects during manual storyboard annotation, the resulting survival curves were of much higher resolution (Figure 8B,E), and further manual annotation of the death times on the storyboard produced changes of no more than ~4% in the estimated mean lifespan. We, therefore, demonstrate that the removal of non-worm objects during the image analysis pipeline of the LSM is the crucial step for obtaining well-resolved survival curves.
Figure 8: The effect of manual data validation on survival curves. The degradation of RPB-2 shortens the C. elegans lifespan in the presence of auxin (K-NAA: α-napthaleneacetic acid) in a dose-dependent manner. (A) Survival curves plotted from the LSM raw output after quality control of the plates and without manual annotation of the worm objects on the Worm Browser storyboard. (B) Survival curves plotted after manual annotation of the worm objects on the storyboard. (C) Survival curves plotted after manual annotation of the worm objects and death times on the storyboard. (D) Summary of all the objects detected by the LSM. The censored worm objects included objects excluded automatically (for example, due to worms moving outside of the scanned area) or manually by the experimenter (for example, due to worms bursting). (E) Tabular representation of the estimated mean lifespan after by-hand worm object annotation (left) and additional by-hand death time annotation (right). Percentage difference in the mean lifespan between neighboring groups of different auxin concentrations and the statistical power of detection. All graphs were plotted on the statistical software JMP. The data for this figure was adapted with permission from Oswal et al.23. Please click here to view a larger version of this figure.
Moving beyond lifespan as a single end-point for studying aging, we then considered behavioral aging and investigated which steps of the post-experimental image analysis were most crucial for measuring it. Focusing on the relationship between vigorous movement cessation (VMC) and lifespan, we compared the LSM output at different stages of image analysis and validation (Figure 9A–D). We found that non-worm objects showed a unique relationship between the apparent VMC and lifespan, with the two being annotated as occurring nearly simultaneously in every object (Figure 9A). In contrast, true nematodes usually ceased moving vigorously several days prior to their death time (Figure 9A). This difference in the relationship between VMC and lifespan provides an additional means for rapidly identifying and excluding non-worm objects.
Figure 9: The effect of manual data validation on vigorous movement cessation (VMC) analysis. (A) Behavioral aging data without manual annotation of worm objects on the Worm Browser storyboard, displaying the relationship between the death times and VMC times in non-worm objects (in black) versus worm objects (in red). (B) Behavioral aging data plotted after manual annotation of worm objects on the storyboard. (C) Behavioral aging data plotted after manual annotation of worm objects and death times on the storyboard.(D) Tabular representation of the average remaining lifespan (ARL; obtained from the intercept between the death age and VMC age) across different steps of the image analysis pipeline and percentage difference in the ARL between worm object annotation and further death time annotation. (E) Graphic representation of the estimated ARLs across different steps of the post-image acquisition analysis, and their relationship to the worm lifespan (which is dependent on the auxin concentration: α-napthaleneacetic acid). The spline fit was performed using a cubic spline method. All graphs were plotted on the statistical software JMP. The data for this figure was adapted with permission from Oswal et al.23. Please click here to view a larger version of this figure.
We found that annotation and exclusion of non-worm objects using the Worm Browser was sufficient to provide a rough estimate of the relationship between the VMC and death times (Figure 9A–C), recapitulating the expected dynamics of the physiological decline in nematodes23. To further explore this, we considered the same RNA polymerase II data set and estimated the average remaining lifespan (ARL) after VMC for each subpopulation as the intercept of a linear regression model relating the lifespan to VMC. To understand the effect of data annotation on the ARL, we recalculated the ARL after each step of data annotation (Figure 9E). We found that the manual annotation of death times in behavioral aging analysis was especially important in longer-lived worms, in this instance, those exposed to the lowest concentrations of auxin (Figure 9D,E). In contrast to its minimal effect on the survival curves, manual annotation of the death times had a substantial effect on the quantitative relationship between the VMC and lifespan, increasing the estimated ARL, for instance at 0 µM of auxin, from 8.09 days to 10.42 days after death time annotation; this represents an ARL difference of 29%. Therefore, we found that the relationship between VMC and death times explained by ARL was much more sensitive to by-hand annotation of the death times compared to measurements of death times for lifespan alone. This sensitivity can be explained by the relative durations of the ARL and the lifespan; the same adjustments to death time will usually be small relative to the lifespan but large relative to the ARL. Thus, adjustments to the death times will have a larger relative effect on the ARL compared to the lifespan.
Supplementary File 1: Structure of the Lifespan Machine experiment schedule file. The experiment schedule consists of three parts. First, the basic information about the experiment is included, such as the name and capture specification. From this part, generally only the name of the experiment needs to be changed for each new experiment. The second part of the document is generated by exporting scan areas from the Worm Browser and specifies the physical location of the scan areas for each scanner ("asuna", "bulma", "moscow", "rio", "yuno", and "yuki" in this example). These are replaced for each new experiment and are copied from the TXT files generated for each scanner individually during step 4.1.2.3. The third part of the document provides information about the duration of the experiment, which should also be modified for each new experiment, and the capture frequency. The device will scan each defined area one at a time at specified intervals for the duration of the entire experiment. Please click here to download this File.
Here, we provide a detailed, accessible protocol for performing an experiment using the latest version of the Lifespan Machine. We have shown that the critical step for achieving well-resolved survival curves is the manual exclusion of non-worm objects during post-image acquisition. Manual death time annotation has a small effect on the overall shape of the survival curves, demonstrating that fully automated death time estimation is efficient even without manual annotation (Figure 8). On the contrary, the acquisition of high-quality behavioral aging data requires more careful annotation of the death times, especially in long-lived individuals (Figure 9). Therefore, the amount of time required for by-hand storyboard annotation will ultimately depend on the specific outcome being measured. Overall, we find that when working with the LSM, the researcher's efforts are most crucial during the experimental setup and, to a smaller degree, during the post-image acquisition analysis. Lastly, we highlight the value of high-throughput, automated assays in producing highly resolved survival and behavioral aging data while increasing the productivity of researchers and supporting experimental reproducibility.
The LSM houses nematodes on agar plates, recreating the classical by-hand lifespan assay in an automated context. Other tools have been developed to automate the measurement of lifespan in C. elegans using different methods of nematode confinement. These include approaches in which single nematodes are housed in solid media (WorMotel15) or within a microfluidic device (Lifespan-on-a-chip11) or those that track a larger population of animals using microfluidics (WormFarm14). The advantages of microfluidic platforms include the possibility of precise, real-time environmental control and the automated removal of progeny by size exclusion. However, the aforementioned devices so far have not proved easily scalable, as they require extensive manual handling and often rely on daily image or video capture triggered by an experimentalist. Other platforms, such as the Stress-Chip39, use the modified flatbed scanners designed for the LSM to image custom microfluidic devices.
In contrast to other methods, the LSM has extensive data annotation and validation facilities, thus allowing users to systematically perform the quality control required to collect high-resolution, precise, and accurate lifespan data in a high-throughput context13. The technique is versatile due to its use of current agar-plate-based laboratory protocols and offers a unique advantage for experimental scalability due to the relative ease of arraying large groups of flatbed scanners. Though the LSM requires an initial time investment to build and operate and to train users on the specialized software, these costs are balanced by the robust, high-throughput production of lifespan data. Lifespan Machine clusters of 50 scanners or more have been deployed in several labs, running continuously for more than a decade40.
The LSM does have limitations. Animals are housed in scanners during the automated collection of survival data, thus limiting the ability of researchers to perform experimental interventions simultaneous to observation and requiring either sterilization of the animals or starting experiments after the nematodes' reproductive phase. Changes in temperature are a rare exception to the limitation on interventions, as the environmental temperature can be modulated without the need to open the scanners and access the animals. In cases where hands-on interventions must be applied to the nematodes mid-life, a common solution is to delay the start of automated observation until after the necessary handling of the animals has been performed. Additionally, there is an inherent variation in the location of plates within and across the scanners. These can be subject to minute, local differences in environmental conditions (such as in temperature, light, ventilation, etc.), which might influence C. elegans lifespan19. This environmental influence can be further quantified and studied by using accelerated failure time regression models41. A way to mitigate this effect is by simply scaling the number of plates and scanners to achieve a rigorous measurement independent of environmental fluctuations. Commonly, plates of the same condition are randomly distributed within each scanner, and population sizes larger than 500 individuals per condition and across scanners have demonstrated statistically robust survival estimates31.
Altogether, the LSM allows for the high-accuracy and large-population collection of survival and behavioral aging data, and could enable previously unfeasible screens to be performed in a quantitative manner. In this way, the LSM contributes a major technical advance for the standardized collection of survival curves and provides a novel framework for the coupled study of lifespan and behavioral aging in nematodes.
The authors have nothing to disclose.
We thank Julian Ceron and Jeremy Vicencio (IDIBELL Barcelona) for producing the rpb-2(cer135) allele. This project was funded by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (Grant Agreement No. 852201), the Spanish Ministry of Economy, Industry and Competitiveness (MEIC) to the EMBL partnership, the Centro de Excelencia Severo Ochoa (CEX2020-001049-S, MCIN/AEI /10.13039/501100011033), the CERCA Programme/Generalitat de Catalunya, the MEIC Excelencia award BFU2017-88615-P, and an award from the Glenn Foundation for Medical Research.
1-Naphtaleneacetic acid (Auxin) | Sigma | N0640 | Solubilize Auxin in 1M potassium hydroxide and add into molten agar |
5-fluoro-2-deoxyuridine (FUDR) | Sigma | F0503 | 27.5 μg/mL of FUDR was used to eliminate progeny from populations on UV-inactivated bacteria |
Glass cleaner | Kristal-M | QB-KRISTAL-M125ml | |
Hydrophobic anti-fog glass treatment | Rain-X Scheibenreiniger | C. 059140 | |
Rubber matt | Local crafstman | Cut on a high-strength EPDM rubber sheet stock | |
Scanner glass | Local hardware supplier | 9" x 11.5" inch glass sheet | |
Scanner plates | Life Sciences | 351006 | 50 mm x 9 mm, polystyrene petri dish |
USB Reference Thermometer | USB Brando | ULIFE055500 | For calibrating temperature of scanners |