LFASS: Label-Free Automated Survival Scoring for High-Throughput Nematode Assays

Caenorhabditis elegans is an excellent model for high-throughput experimental approaches, but lacks a robust, versatile and automated means to pinpoint time of death. Here we describe an automated, label-free, high-throughput method using death-associated fluorescence to monitor nematode survival, which we apply to stress and infection resistance assays. We demonstrate its use to define correlations between age, longevity and stress-resistance, and reveal an autophagy-dependent increase in acute stress resistance in early adulthood.

include microfluidics and automated robotic arms 7,8 . Yet full automation of C. elegans survival assays as been limited by death scoring techniques. Worms are deemed dead when they fail to move in response to touch, making death scoring by movement-tracking impracticable in quiescent, old or mutant animals with impaired mobility or touch-response, or in assays that affect worms' ability to sense or move.
We recently discovered that an endogenous burst of blue fluorescencedubbed "death fluorescence" (DF) -generated by de-quenching of anthranilic acid species, occurs in nematodes' intestine at the onset of death 9 . From time-lapse recordings of blue fluorescence during killing assays, we observed that the timing of individual DF events follows a normal distribution. Considering the total fluorescence of a worm population over time, we found that its half-maximal fluorescence corresponds to half the worms undergoing DF (Supplementary Figure 1a, 1b).
Hence, the time of half-maximal fluorescence corresponds to the median time of death, which is the main parameter sought in survival assays. Because it is endogenous and easily detectable by fluorescence microscopes, spectrophotometers and plate-readers, DF recording could replace vital dyes and movement tracking algorithms in survival assays in a variety of laboratory setups.
We used DF measurement to develop new, label-free, high-throughput and automated oxidative-stress, heat-shock and infection survival assays. To enable multiple conditions to be tested in parallel and to limit phototoxicity (data not shown), we opted for a 384-well plate format to be read by a light-emitting diode (LED)illumination plate-reader (Fig. 1a). In these conditions, DF was optimally recorded with excitation/emission wavelengths of 360/435 nm (Supplementary Figure 1c) using as few as 16 worms per well. With this setup, each measurement takes 0.8 s, allowing for 150 samples to be measured in under 2 min. This enables accurate and unbiased measurement of median time of death in assays spanning from hours (acute stress assays, Fig. 1b, 1c) to days (bacterial infection assays, Fig. 1d). Hence this label-free automated survival scoring method (henceforth referred to as LFASS) may be adapted for high-throughput approaches.
To demonstrate its practicality, we used LFASS to re-examine in unprecedented depth the controversial link between stress resistance and longevity [10][11][12] , which is tedious to study by traditional means. Survival stress assays have long been used to dissect mechanisms of disease and ageing, revealing that animals lose their ability to endure chronic stress as they age 13-15 , in line with molecular damage theories of ageing [16][17][18] . However, this correlation has not been rigorously tested with respect to intense, acute stress paradigms that challenge initial resistance to stress, as opposed to longer-term resistance engaged under chronic milder stress, which might primarily reflect adaptability to stress.
We exposed aged worm cohorts to either acute oxidative-stress or heat-shock, seeking conditions that would kill 100% of worms within a few hours, relying on the timing of median survival to detect differences in stress resistance. Wild-type adult hermaphrodites exposed to 0.002% -7% tert-butyl-hydroperoxide (t-BOOH) showed simple dose-dependency in oxidative stress resistance (Fig. 1b). Thermal stress, from 35C -42C, also had a dose-dependent effect (Fig. 1c). To limit the influence of uptake mechanisms and stress adaptation on measured stress resistance, we used 7% t-BOOH and 42C as acute stress challenges in subsequent studies, which are both characterized by a median time of death of 1.5 h.
We next tested these protocols on well-characterized longevity mutants, monitoring stress resistance throughout life, which the high-throughput capability of LFASS affords. The insulin/IGF-1 signalling (IIS) pathway is highly conserved from invertebrates to mammals, regulating growth, metabolism and ageing 19 . Mutations of the daf-2 IIS receptor extend lifespan and increase stress resistance, while those of the downstream transcription factor DAF-16/FoxO shorten lifespan and cause stress hypersensitivity 20 . As expected, ageing cohorts of daf-2 and daf-16 mutants showed resistance and hypersensitivity, respectively, to acute stress (Fig. 1e, 1f, Supplementary Table 1). Against expectation, wild-type worms showed a daf-16independent age increase in resistance to acute oxidative stress, which peaked at day 10 and reached a level of resistance comparable to that of stress resistant daf-2 mutants (Fig. 1e, Supplementary Tables 3, 4). Because this effect persisted in animals with genetically or pharmacologically impaired feeding ( Supplementary   Figure 2a, 2b), and was supressed by specific mutations (Fig. 2b-d), it is unlikely to be due to an age-associated decrease in t-BOOH uptake. By contrast, wild-type animals exhibited a daf-16-dependent rise and fall in heat-shock resistance, peaking at day 6 ( Fig. 1f, Supplementary Table 2, 4).
To explore the genetic basis of these phenomena, we screened selected ageing mutants. As manual analysis of high-throughput data generated by LFASS is very time consuming, we developed a software package to automatically extract median The <5% unfitted curves can be re-analysed individually with user guidance.
Automated analysis yielded near identical results to manual analysis (Fig. 1g), but in 1/100 th of the time.
Screening ageing mutants showed, again, major differences in the relationship between longevity and resistance to acute oxidative stress versus heat-shock (Supplementary Figure 2). Firstly, acute oxidative stress resistance correlated very poorly with lifespan across genotypes or treatments, while there was a strong positive correlation between heat-shock resistance and lifespan (Fig. 2a, Supplementary   Figure 4). The skn-1 (NRF2) antioxidant transcription factor 21 was required for the age increase in oxidative stress resistance, and the hsf-1 heat shock transcription factor 22 for the age increase in heat-shock resistance (Fig. 2b). Interestingly, autophagy proved to be essential for both effects (Fig. 2c), in line with its known increase in activity during early adulthood 23 and stress protective effects 23,24 . All autophagy mutations tested suppressed ageing-associated resistance to heat-shock and oxidative stress (Fig. 2c,  Consistent with these results, adult-specific RNAi targeted against daf-16, hsf-1 and atg-13 could all reduce ageing-associated heat-shock resistance in wild-type and atg-13(bp414) worms (Fig. 2d), but did not significantly affect acute oxidative stress resistance (Supplementary Figure 5). Thus, adult expression of daf-16, hsf-1 and atg-13 all contribute to the ageing-associated increase in heat-shock resistance but are dispensable for acute oxidative stress resistance.
While chronic stress resistance declines with age 13-15 , we find that acute stress resistance increases in early ageing. Moreover, the dynamics of heat-shock and acute oxidative stress resistance with ageing differ markedly. A hypothesis that reconciles these results is that older animals, being under "physiological stress", have a higher initial stress-buffering capacity, but lack the capacity to adapt to sub-lethal stress. The earlier fall in heat-shock resistance compared to acute oxidative stress resistance is consistent with the previously observed early decline in proteostasis 13 . Importantly, our correlation analysis predicts that screens for heat-resistant rather than prooxidantresistant variants are more likely to yield new determinants of ageing.
In summary, LFASS proved applicable to a range of nematode survival assays and its high-throughput capabilities enabled new insights about stress resistance and ageing to be obtained. It is unbiased, easily implemented and versatile, requiring no added reagents and is compatible with transgenic, frail or immobile worms. It does not require a strict sample size so that worm loading can be automated (e.g. using worm sorters, automatic dispensers). It is applicable to modern screening platforms employing transparent materials (e.g. microfluidic chips, multi-well plates) and is highly cost effective. It will be advantageous to accelerate anti-helminthic drug discovery, and studies of toxico-pharmacology, bacterial infection and ageing.
C. elegans strains were maintained at 15°C following standard culture conditions 26 , on NGM 60 mm-diameter agar plates seeded with E. coli OP50. Ageing worm cohorts were prepared as follows. Young adult hermaphrodites (30 per plate) were allowed to lay eggs for 24 hr. Several days later L4 animals were collected and transferred to NGM/OP50 plates containing 15 M fluorodeoxyuridine (FUdR) (Sigma-Aldrich ♯F0503) to block egg production; or to NGM with 15 M FUdR and 25g/mL carbenicillin (Sigma-Aldrich ♯C3416), and seeded with HT115 RNAiproducing bacteria, as described 27 , and in each case maintained at 25˚C. L4 larvae were collected in this way daily for 2-3 weeks, and then adult hermaphrodites of each age were picked into multi-well plates and subjected to LFASS all on the same day.
Time-lapse microscopy experiments. 100-120 1 day old adult hermaphrodites were mounted in M9 on 2% agarose pads between slide and coverslip without anaesthetic unless otherwise stated. Imaging was performed through a DAPI filter set (Chroma technology Corp, USA) using a 2.5x objective on a Leica DMRXA2 microscope (Leica Biosystems Nussloch GmbH, Germany). Successive brightfield and DAPI images were acquired every 30s using the Volocity 6.3 software (Perkin Elmer, USA). For rapid killing experiments (Supplemental Figure 1a), a 70% solution of tert-butylhydroperoxide (Luperox TBH70X, Sigma-Aldrich ♯458139, Switzerland) was added volume to volume to the mounting M9 medium prior to coverslip apposition. The imaging protocol was started exactly 1 min later. For heat-killing experiments (Supplemental Figure 1b), a PE120 heating/cooling platform (Linkam Scientific Instruments Ltd, UK) was attached to the microscope stage. The heating protocol (ramping up to 42°C at 1.2°C per min) was initiated simultaneously with the time-lapse imaging.
Time-lapse microscopy analysis. We used the Volocity 6.3 Quantitation module to generate graphic representations (kymographs) of single worm traces from the 2.5x time-lapse imaging series. The time of death for each worm was deduced from the time of the intestinal blue fluorescence burst. Individual times of death during a single time-lapse were fitted into bins and count distributions plotted and fitted with a Gaussian curve using GraphPad Prism 6.0 software (GraphPad Software Inc., USA). Overall fluorescence for each time point was measured using the ImageJ-based opensource package Fiji (http://fiji.sc/Fiji), plotted and analyzed using GraphPad Prism 6.0.
Plate-reader assays. For oxidative stress and heat shock assays, we picked 16 worms into 60 L M9 per well for 384-well plates, and 50 worms in 150 L M9 for 96-well plates, together with a pellet of E. coli OP50 bacteria to prevent starvation. For infection assays, Enterococcus faecalis GH10 bacteria were streaked onto Brain Heart Infusion Kanamycin (BHIK) agar plates and used within a week, as described 28 . Liquid (BHI) E. faecalis cultures were grown for 3-5h at 37°C to saturation on the day. We then picked 100 worms per well into 50 L M9 + 30 L OP50 medium (for 384-well plates), and supplemented with 10 L freshly saturated E. faecalis solution cooled to room temperature. A Tecan Infinite 200 plate-reader (Tecan Group Ltd., Switzerland) was pre-warmed at 25°C to match the temperature at which aged cohorts were raised and E. faecalis infections assays performed. Blue fluorescence (excitation: 360 nm / emission: 435 nm) was recorded for each well every 2min for 8h or every 5min for 4 days for stress and infection assays, respectively.
Death fluorescence (DF) curve manual analysis. Fluorescence time-lapse recording data for each well were normalised. The maximum was chosen where a significant peak of fluorescence was observed. Fluorescence values for the first 15-20 time points were often inaccurate, yielding local maxima and minima, due to worm thrashing in the wells, and were therefore omitted for the determination of the fluorescence minimum and maximum. After normalization, the time of half-maximum fluorescence was determined. (1) Matlab separates text and number matrices so that tags and values are stored in separate matrices. In .xlsx plate-reader files each row represents the fluorescence of a single well over time. The last column is the well identity/tag (attributes of the sample: age, genotype, drug treatment, bacterial type).
(2) For the fit to perform optimally and return median time of death in minutes, several assay parameters have to be informed by the user (see supplemental online methods). The time interval between two measurements of the same well allows for the results to be expressed in minutes. The noise threshold allows for discarding empty wells, and wells in which no fluorescence peak can be detected. The noise fluorescence threshold should be chosen above the fluorescence values measured in empty wells, and below the peak blue fluorescence value samplecontaining wells. This threshold also depends on the number of worms per well, and all data treated in the same bulk analysis should have roughly the same number of worms per well. Max and min have to be identified for the normalisation of the data. Early fluorescence values can greatly fluctuate due to worm thrashing/swimming (in the absence of anaesthetic) associated with high or low fluorescence values that can exceed the relevant maximum and minimum. For this reason, the user must indicate in which time intervals min and max are to be expected. These intervals usually exclude the first 10-20 time points and have to include all the times of minimum or maximum fluorescence of all the data sets included in one analysis. Once the data are normalised, the sigmoid fit needs to be constrained within initial and final plateaux that match the min and the max values. Because of inaccuracies in measurements, to chose the best plateaux, they need to be fitted over several time points. To achieve this, a tolerance threshold is given for the min/max (i.e., 0.5/0.95). The fit function will then take into account time points around the min/max that are found within these tolerance thresholds (between 0 and 0.5, and between 0.95 and 1, respectively) and stop looking for additional points beyond. This will effectively define a fit region that encompasses the death-associated blue fluorescence burst, ignoring all other parts of the curve (see green dotted lines in supplemental figure 3 (3b) and (b) lower panels).
(3) To speed up computing, the program first excludes rows that do not need to be fitted such as parameter rows (date, time points, temperature with time, etc.) by keeping only rows with at least 6 consecutives numbers (non-data rows contain letters). It then uses the noise threshold defined earlier to exclude data rows that do not have values above this threshold, which eliminates most of the empty wells (3a). Inaccuracies in measurements are associated with noise spikes that can complicate fitting. To limit their influence, the data are smoothened twice using the "smooth" function (4 other smoothing options were compared and this performed best). Min and max are found, data are normalized, and a fit interval is found. In the region to fit, curves are typically sigmoidal in shape. Because we are only interested in extracting the time of half-maximum (corresponding to the median time of death, Fig. S1), the critical region to fit is around this time point, and the sigmoid fit performs optimally.
(4) As the analysis progresses, a 4-column wide .txt result table is filled. Column 1 contains all the tags, column 2 reports the half-maximum time inferred from unfitted normalised raw data, column 3 reports half-maximum times obtained with the bulk fitting analysis, and column 4 reports the updated values obtained from bulk fitting and user-guided analyses. Dataset names and parameters (temperature, duration of assay) are filled in the first rows of column 1. "0" fill empty cells from non-data rows. "1" fill cells in data rows that did not pass the user-defined noise threshold. "NaN" fills cells in data rows that could not be fitted.
(5) When the fit does not converge for a given row, the user can re-analyse it giving attribute values that differ form the bulk analysis and that are better suited for this specific row. Typically, this allows for recovery of the 5% exploitable low quality data that are excluded by the bulk analysis. The result table is updated after each reanalysis until the user stops. (

6) Post-processing is performed in Microsoft Excel for data sorting and basic row/columns operations. Then statistical analysis and graphical representations are processed in GraphPad Prism.
Lifespan assays. With the exception of RNAi experiments, all worm cohorts used in reported stress or lifespan assays were hermaphrodites maintained at 15°C on OP50seeded NGM plates and switched at the L4 stage to OP50 plates supplemented with 15M FUdR, and subsequently maintained at 25°C.

Statistics.
For lifespan statistics we used the JMP 12.01 Pro software package from SAS (USA). Lifespans were compared using the non-parametric log rank test. Unless otherwise stated, all other statistics were performed using Prism 6.0 from GraphPad 26 Brenner, S. The genetics of Caenorhabditis elegans. Genetics 77, 71-94 (1974    -find/open each file -for each file identify data rows -for each file identify row tags (ages/genotypes, etc.)

3a) Sort data to fit
-exclude non-data rows -exclude empty wells based on noise threshold 3b) Data manipulation for each row -smoothen twice (Matlab smooth function) -find relevant min scanning user-defined time interval -find relevant max scanning user-defined time interval -bring values above max to max -bring values below min to min -normalise (scale values between 0-1) -define fittable region using user-defined tolerance thresholds 3b) Sigmoid curve fitting -fit upper plateau using user-defined tolerance threshold for max -fit lower plateau using user-defined tolerance threshold for min -find best fit within fittable region with max and min threshold constraints -find the half-max time within the fitted region on raw curve and on fit fit does not converge 4a) Create result .txt table -create 4-column table (row tag, estimated median TOD from raw data, fitted data, re-fitted data) -assign file name for each dataset -assign value "0" for columns 2-4 -assign tag for each row to column 1

4b) Fill in bulk analysis results
-assign value "0" for columns 2-4 of excluded rows -assign value "1" for columns 2-4 of rows excluded based on noise threshold -assign half-max time from raw data to 2nd column for all data rows -assign half-max time from fit to columns 3 (typically >95% data) if initial value is "0" -assign half-max time from fit to columns 4 -assign "NaN" to unfittable data (typically <5% of the data) for columns 2-4

5) Re-analyse unfitted data with user guidance
-display smoothened curves with fits and ask if user-guided fit is wanted (re-fitting will be performedl if data are deemed exploitable by user) -if "Y", then close displayed graphs and perform steps 2a to 3b and 4b, 5 until user decides to stop.

Supplementary legends
Supplemental Figure 1 | Using population death fluorescence to determine median time of death. Time-lapse microscopy (exemplified by panels a3i-a3v) of worm population submitted to fast (a) and slower (b) killing assays reveal that the time of half maximum death fluorescence (a1, b1, red dotted line) corresponds to the median time of death (a2, b2, red dotted line). Optimal excitation (c1) and emission (c2) windows for anthranilate-dependent death fluorescence in 384 well plates read by a LED plate-reader, matched with excitation and emission spectra for two commercially available anthranilate compounds.
Supplemental Figure 2 | Effects of lifespan-affecting mutations and drug treatments on the age increase in acute stress resistance. (a) Inhibition of pharyngeal pumping by ivermectin or levamisole does not suppress the age increase in oxidative stress resistance. b) Slower pumping rate and associated genetically-induced dietary restriction (DR) do not correlate with acute oxidative-stress resistance dynamics (left) but correlates with heat-shock resistance dynamics: eat-20(nc4) is a mild pumpingdefective mutant, eat-2(ad465) is a strong pumping/genetic DR mutant, daf-2(e1370) pumps more slowly than daf-2(m577) and so combines a genetic DR phenotype with reduced IIS. (c) The lifespan extending drug metformin sensitizes daf-2 and wild type worms to acute oxidative stress but confers increased heat-shock resistance to young wild-type adults. (d) AMP-activated protein kinase alpha subunit (aak-2), vacuolar-H + -ATPase subunit 12 (vha-12) and hypoxia-induced factor 1 (hif-1) differentially affect age increases in acute oxidative-stress and heat-shock resistance. (e) The endoplasmic reticulum-associated protein degradation (ERAD) and unfolded protein response pathways are dispensable for the age increase in heat stress resistance. Error bars, SEM. Comparison with age-matched wild-type: * p<0.05 down to p<0001.