A fitness assay for comparing RNAi effects across multiple C. elegans genotypes
- Mark Elvin†1,
- Laurens B Snoek†2,
- Martin Frejno1,
- Ulrike Klemstein1,
- Jan E Kammenga2Email author and
- Gino B Poulin1Email author
© Elvin et al; licensee BioMed Central Ltd. 2011
Received: 11 August 2011
Accepted: 17 October 2011
Published: 17 October 2011
RNAi technology by feeding of E. coli containing dsRNA in C. elegans has significantly contributed to further our understanding of many different fields, including genetics, molecular biology, developmental biology and functional genomics. Most of this research has been carried out in a single genotype or genetic background. However, RNAi effects in one genotype do not reveal the allelic effects that segregate in natural populations and contribute to phenotypic variation.
Here we present a method that allows for rapidly comparing RNAi effects among diverse genotypes at an improved high throughput rate. It is based on assessing the fitness of a population of worms by measuring the rate at which E. coli is consumed. Critically, we demonstrate the analytical power of this method by QTL mapping the loss of RNAi sensitivity (in the germline) in a recombinant inbred population derived from a cross between Bristol and a natural isolate from Hawaii. Hawaii has lost RNAi sensitivity in the germline. We found that polymorphisms in ppw-1 contribute to this loss of RNAi sensitivity, but that other loci are also likely to be important.
In summary, we have established a fast method that improves the throughput of RNAi in liquid, that generates quantitative data, that is easy to implement in most laboratories, and importantly that enables QTL mapping using RNAi.
The basic principle of the method is to assess the fitness of a population of animals by measuring the rate at which E. coli is consumed. Typically, an experiment assesses the fitness over a period of 8 days. This generates Fitness Curves (FCs), which are amenable to statistics. Various analyzes can be performed on the curves using parameters such as the slope, the mean, the time required to consume 50% of the food, the difference between first and last readings, or even the surface above the area of the curve. Here, we mainly used variation on the slope, since this parameter can capture most of the effects observed on the FCs.
We demonstrate that our approach is robust by mapping the C. elegans loss of RNAi sensitivity trait using diverse genetic backgrounds. We took advantage of Recombinant Inbred Lines (RILs) that were produced by mating two diverse genetic backgrounds N2 (Bristol) and CB4856 (Hawaii). This RIL population has extensively been used to map a range of quantitative traits [3–5]. Hence, we used these previously described RILs [6–8] and a number of new RILs (Additional file 1, Table S1). The N2 (Bristol) background is RNAi sensitive whereas the CB4856 (Hawaii) background is RNAi insensitive in the germline . Using Quantitative Trait Loci (QTL) mapping, we identified ppw-1, a gene causal to the RNAi insensitivity of the CB4856 germline . Importantly, we performed a detailed genotype-to-phenotype analysis that provides evidence that loss of RNAi sensitivity is a complex trait, in which ppw-1 is one of probably several RNAi sensitivity modifiers. In total, 56 genetic backgrounds were tested using 12 RNAi treatments. Analyzing such a number of genetic backgrounds has never been performed before, but critically this is the first time that quantitative data have been produced for RNAi treatments in C. elegans.
Bristol (N2) and Hawaii (CB4856) respond differently to RNAi treatments
Bristol and Hawaii react very differently when exposed to RNAi. RNAi in Bristol is highly effective, but in Hawaii RNAi sensitivity for germline expressed genes has been lost . Therefore, we reasoned that our fitness assay should be ideal to differentiate between these two RNAi sensitivity behaviors. For example, if we target an essential gene that functions in the germline, Bristol will die, but not Hawaii. We performed RNAi in both strains on a panel of 12 RNAi clones: four clones (targeting par-1, par-6, pos-1, and mel-26) are known to be more effective in Bristol than in Hawaii ; two clones (targeting rab-5 and tag-214) are known to be effective in both natural isolates ; one clone (targeting lin-31) is used as a negative control since it affects vulval development, a non-essential organ for viability ; four clones (targeting smo-1, mpk-1, gld-1, and let-502) have never been tested before for comparison between Bristol and Hawaii; and the empty vector clone is used as a control or a reference. We have analyzed the data generated from the fitness assay in two ways: i) by comparing each RNAi condition to the empty vector for each strain and ii) by analyzing whether Bristol responds differently from Hawaii to specific RNAi treatments.
Three-way analysis of RNAi treatments performed on Bristol (N2) and Hawaii (CB4856)
EV vs Target in Bristol
EV vs Target in Hawaii
P value (Bristol vs Hawaii)
In the second and most innovative part of our analysis, we compared the FCs obtained from the same RNAi treatment between Bristol and Hawaii. Accordingly, Bristol and Hawaii do not respond differentially to empty vector, mpk-1(RNAi), the control lin-31(RNAi), and rab-5(RNAi) (Table 1). However, we observed differential responses with RNAi targeting mel-26, pos-1, par-1, par-6, gld-1, smo-1 and let-502 (Figure 2 (bottom 3 panels) and Table 1). Surprisingly, tag-214 appears to produce a differential response suggesting that the RNAi treatment was more effective in Hawaii than Bristol. Detailed analysis would be required to confirm this observation. Taken together, these results show that RNAi treatment effects can produce data consistent with previous studies [1, 2]. Importantly, these data can be quantified when applied onto diverse genetic backgrounds or genotypes.
PPW-1 is a modifier of RNAi sensitivity
RNAi treatments of 56 RILs identifies a shared QTL at the ppw-1 locus
A genotype-to-phenotype analysis suggest that loss of RNAi sensitivity is a complex trait
Factors affecting the fitness assay
Here we show that our method is efficient at addressing the contribution of genetic variation on a specific trait, herein loss of germline RNAi sensitivity. We think that the Fitness Assay will be useful to other groups interested in identifying the contribution that diverse genotypes may have to their phenotypes of interest. This may involve different treatments such as environmental stress or chemical compounds. The Fitness Assay is easy to implement in any C. elegans laboratory, but future users need to take into account two critical parameters for future adaptations of the fitness assay: the number of worms and the quantity of food. These parameters need to be optimized, in particular for strains displaying a low brood size or lethality.
For our purpose, the worms were seeded manually, but a multiwell dispenser could also be used. This apparatus can process a 96-well plate in about 10 seconds, but tends to be more variable than the manual method using a single channel pipette (+/- 4.5 versus +/- 3 worms, respectively). We aimed at seeding 20 L1 worms per well for our experiments, and used duplicates for the manual method, however when using the multiwell dispenser we performed the experiments in triplicate. We directly tested the efficiency of our assay using the multiwell dispenser by performing six RNAi treatments (lin-31, rab-5, smo-1, par-1, par-6, and mel-26) over 38 RILs. We found that we can still map a QTL at the ppw-1 locus for smo-1, par-1, par-6 and mel-26 RNAi treatments (Additional file 2, Figure S1). This confirms that the multiwell dispenser can also be used to seed the worms to perform the Fitness Assay. Even though not tested here, it is very likely that using a worm sorter, such as the COPAS (Union Biometrica), could reduce or perhaps eliminate the worm number issue and increase resolution.
Sources of variation
Here we have presented a new assay that is both cost efficient and easy to set-up in any laboratory. The main advantage of the Fitness Assay is its ability to distinguish and quantify the effect of a specific treatment across multiple genetic backgrounds or genotypes (Figure 1). We show that this is particularly powerful to identify the QTL responsible for loss of germline RNAi sensitivity in RILs generated from Bristol (RNAi sensitive) and Hawaii (RNAi insensitive) [6–9]. This study provides evidence that modifiers of this trait remain to be identified. The method is also versatile and other treatments could be used to decipher the contribution of natural genetic variation to specific traits, i.e. a drug response.
The Fitness Assay indicates the capacity of an animal population to eat at a certain rate. This rate is determined by multiple factors, including the number of progeny, the level of viability, and the rate of growth. In our selected set of RNAi, we found that most targeted genes produced a significant effect on the FCs (Fitness Curves), which is in accordance with their previously described function. In addition, using data from Kamath et al, 2003, we found that most phenotypes, around 75%, are due to embryonic lethality, a reduced brood size, or a growth defect. Indicating that the majority of genes that produce a phenotype by RNAi in Bristol could be identified by our method. We tested this prediction by targeting 40 genes: 20 known to produce 'viability-related' defects, and 20 with no 'viability-related' defects reported by . However, these latter 20 targets have been shown to have functions associated with signaling pathways, apoptosis or transcription (see Wormbase). We found that 75% of the genes required for viability produced a phenotype in the Fitness Assay, indicating that most RNAi treatments that produce a 'viability-related' phenotype would also produce a phenotype in the Fitness Assay (Additional file 3, Figure S2). Using the other set of targeted genes, we found that 60% of the RNAi treatments have no effect. Interestingly, the remaining 40% producing an effect on the Fitness Assay have been shown to play roles in WNT signaling (lin-44), nuclear excision repair (xpg-1), Notch signaling (sup-17 and aph-1), apoptosis (ikb-1 and vps-18), translational repression (fbf-1), and transcription (mab-5) (Additional file 4, Figure S3). Collectively, these data indicate that the phenotypes identified using the Fitness Assay mostly overlap with the data from Kamath et al. 2003  and that overall if we were to re-perform a genome-wide screen, we would miss some, but also detect a few new ones.
Even though phenotypes confined to non-viable organs such as the vulva could be missed by our method, the signaling pathways involved are often functional in other organs, and these organs may be required to maintain viability. For example, vulval development requires the RAS signaling pathway, but this pathway is also essential to the development of the excretory cell; and a malfunction of the excretory cell will cause larval lethality . The Fitness Assay can identify this phenotype. There are alternatives methods that provide high-resolution phenotypic analysis. For example, high throughput image analysis increases phenotypic information, albeit at a reduced throughput [14, 15]. However, most C. elegans laboratories do not have an up and running automated image-capture system. For these laboratories, the Fitness Assay remains the affordable option.
In summary, this is the first study that shows that RNAi effects can be quantified in C. elegans. This is important since it will allow us to understand the contribution of diverse genetic backgrounds to complex traits. For example, there is the concept that medicine should be personalized, in part because of genetic variation between humans. Perhaps fundamental principles of genetic variation could be derived from studies performed in Recombinant Inbred Lines in C. elegans, and these could help develop the field of personalized medicine research.
Experimental set-up and timeline
Strains and growth conditions
Wild type C. elegans strains (N2 Bristol and CB4856 Hawaii) were used. Other strains: NL2550 ppw-1(2505) I and WM27 rde-1(ne219) V. For QTL mapping we used a total of 56 Recombinant Inbred Lines (RILs) generated from a cross between Bristol and Hawaii, of which 27 are partly described in [6–8]. We added 29 newly generated RILs, the genotype of each RIL used is described in Additional file 1, Table S1. All strains were maintained on E. coli OP50-seeded Nematode Growth Medium (NGM) plates as previously described . All experiments were conducted at 20°C.
RNAi by feeding in liquid 96-well format
1. Preparation and induction of RNAi bacteria
Inoculate RNAi bacteria into 1 ml of LB containing 100 μg/ml ampicillin in a deep 96-well plate (BD Biosciences 353966) and incubate overnight at 37°C in a shaking incubator at 250 rpm. To induce the production of dsRNA add IPTG to a final concentration of 4 mM and incubate at 37°C for 1 hr in a shaking incubator at 250 rpm. After 1 hr pellet the bacteria by centrifugation at 4000 rpm for 5 mins. Resuspend bacterial pellets in 200 μl S-medium containing 100 μg/ml ampicillin and 4 mM IPTG .
2. Preparation of worm strains
On the day prior to setting up RNAi by feeding in liquid bleach gravid adults and hatch embryos overnight in M9 buffer to obtain synchronized L1 population. The following day pellet synchronized L1 worms by centrifugation at 4000 rpm for 1.5 mins and resuspend in M9 for a concentration of approximately 20 L1 worms per 5 μl of M9 buffer.
3. RNAi by feeding in liquid
For direct comparisons to be made between strains under study then approximately 20 synchronized L1 worms should be added to each well of a 96-well plate (Corning® Costar® CLS3596) in 5 μl of M9 buffer (either with a pipette or with the multiwell dispenser). Immediately afterwards add 60 μl of the resuspended RNAi bacteria in S-medium (from 200 μl volume). Perform all RNAi experiments in triplicate (therefore using 180 μl out of the 200 μl bacterial resuspension). Incubate 96-well plates in a humidity chamber at 20°C for the duration of the experiment.
4. OD measurements using plate reader
Measure the absorbance of the triplicate 96-well plates at 600 nm using a plate reader (Biotek® EL808) between 0 hrs - 192 hrs (0 - 8 days).
Normalization of FC data (with unequal start ODs)
Within an RNAi treatment the sources of variation could be either technical or genotypic. When testing RILs our interest is in the genotypic variation, therefore it is desirable to correct for some of the technical variation. Each of our experiments was performed in triplicate, always including Hawaii and Bristol as reference strains, and additionally included a row dedicated to RNAi bacteria without worms (bacteria blank). Hence, controlling for contamination and spurious bacterial growth, respectively. We observed that variation of bacterial growth (without worms) is minimal between different replicas, experiments or treatments. However, the starting ODs can vary (mean: 0.79 sd: 0.065). Therefore, we investigated if the bacterial growth was affected by the start OD. The 312 growth curves of all the batches of the RNAi bacteria without the worms were taken together and the source of variation studied by an anova. The linear model used was: " y(OD) ~ x(time) + x(time):x(start OD) ". This showed that, although the starting OD is a highly significant (p < 1*10-16) source of variation in the bacterial growth curves the effect size is minimal. The growth curves are nearly parallel and indicate no significant bacterial growth. However, to correct for the variation observed between starting OD, we divided the raw ODs of all time-points by the OD of the first time-point. In this way all curves start at relative OD of 1.
The start OD of the bacterial suspensions in which the worms were exposed to RNAi were also different (mean: 0.83; sd: 0.069). Above we have shown that this could influence the FCs. To normalize and correct for this difference, we compared the significance and the effect of the start OD on the variation found in the uncorrected and the start OD corrected FCs. All the 160 empty vector FCs were taken together and used in an anova. The linear model used was: " y(OD) ~ x(time) + x(time):x(start OD)". Both the significance and the effect of the start OD interaction with the slope of the FCs were lower when the FCs were corrected by dividing by the start OD (sig: 1.8*10-28 → 5.6*10-8 ; eff: 5.4*10-3 → 3.0*10-3). A similar reduction of the role of the start OD was found after correction for specific RNAi used (data not shown).
We further investigated the effect of the correction of the ODs on genotype as an explanatory factor for the variation in the feeding curves. All the RNAi data was analyzed together in an anova. The model used in the anova was: " y(OD) ~ x(time) + x(genotype) + x(RNAi) + x(time):x(start OD)". The start OD is far less significant after correction whereas the genotype as source of the variation is as significant as when the uncorrected values are used. We conclude that it is best to start with equal amounts of food (start OD) but since this is not always possible, it is beneficial to correct for the start OD when analyzing the feeding rate dynamics.
Differences between the FC signatures of Empty Vector and RNAi Treatment and between the FC signatures of Bristol and Hawaii
Matrix used to perform a chi-square test
Genotyping the new recombinant inbred lines
The new recombinant inbred lines were genotyped for 96 SNPs by Illumina "Golden gate" SNP genotyping  (Additional file 1, Table S1). SNPs correspond to previously used SNPs to genotype 80 RILs . Information from Illumina can be found at: (http://www.illumina.com/technology/goldengate_genotyping_assay.ilmn)
All FC data were normalized per RNAi treatment. ODs per time point of all experiments were divided by the average start OD. In this way all RNAi treatments had a relative start OD of 1 in all replicate experiments. We calculated the significance of each of the 121 markers of the N2/CB RIL population (partly described in [6–8]) by a linear model for each of the individual RNAi treatments and the empty vector (ev). With this linear model the QTLs are calculated by explaining the variation in FCs by start OD, time and marker (y(OD)) ~ x(time) + x(time):x(startOD) + x(marker). All measured FCs were used and only the approximate linear part of the FC was used (time-points: 24 to 196). We used 1000 permutation per RNAi treatment to determine a genome-wide -log10(p) threshold of 0.05. The FCs were randomized over the RILs for each round of QTL mapping in the permutation test. The QTL profiles were collected and for each profile the most significant score was put in a list. This list was ordered and the 50th highest value per RNAi treatment was used as the threshold.
QTL mapping for the 6 RNAi treatments using the multiwell dispenser (Additional file 2, Figure S1) was done by calculating the slopes of the FCs over time-points 2 to 5 (24, 96, 120 and 144 hours) of the three replicas and than take the average per genotype. We used 38 RILs to calculate the QTLs with a single marker model: y(mean slope)~x(marker).
Comparing FCs of individual RILs to Bristol and Hawaii
The FCs of 32 individual RILs, for which we sequenced ppw-1, were compared to the FCs of Bristol and Hawaii by a chi-square test. Averages per strain per time-point were calculated and normalized by dividing by the start OD. The FCs were further transformed to % OD of the normalized start OD. Two separate tests were performed for the RILs FCs against the Bristol's FCs and Hawaii's FCs.
As with any high-throughput experiment, the Fitness Assay generates large quantities of data. The nature of this data enables many types of detailed analysis. But to be used as a truly high-throughput method, a simple robust type of analysis needs to be applied. Also a software environment that enables high-throughput of statistical tests is recommended. The simplest level of analysis is to reveal the most severe RNAi effects. To this end, we used a t-test, which is performed on the relative ODs of the last time point(s). The next level of analysis aims at detecting less severe RNAi effects, which tend to affect the middle part of the curves. We found that comparing slopes using a t-test reflects the RNAi treatment effects better. Alternatively, a chi-square test can be used on the mean curve of the two samples. The time points of the individual replicates are tested to be above or below the overall mean and summed per treatment. The advantage of this latter method is that it can be used regardless of the shape of the curve and that it is easily adjusted to a specific part of the curve. In summary, many different tests can be applied to analyze the FCs, but in most cases simple tests are sufficient to detect RNAi treatments or genotypes effects.
ME, UK, LBS, JK, and GBP are supported by the FP7 PANACEA project project nr. 222936. GBP is a research fellow from the MRC (Career Development Award G0600127). Special thanks to Yang Li (University of Groningen, The Netherlands) for establishing the core set of RILs.
- Kamath RS, Fraser AG, Dong Y, Poulin G, Durbin R, Gotta M, Kanapin A, Le Bot N, Moreno S, Sohrmann M, et al: Systematic functional analysis of the Caenorhabditis elegans genome using RNAi. Nature. 2003, 421 (6920): 231-237. 10.1038/nature01278.View ArticlePubMedGoogle Scholar
- Lehner B, Tischler J, Fraser AG: RNAi screens in Caenorhabditis elegans in a 96-well liquid format and their application to the systematic identification of genetic interactions. Nat Protoc. 2006, 1 (3): 1617-1620. 10.1038/nprot.2006.245.View ArticlePubMedGoogle Scholar
- Gutteling EW, Doroszuk A, Riksen JA, Prokop Z, Reszka J, Kammenga JE: Environmental influence on the genetic correlations between life-history traits in Caenorhabditis elegans. Heredity. 2007, 98 (4): 206-213. 10.1038/sj.hdy.6800929.View ArticlePubMedGoogle Scholar
- Gutteling EW, Riksen JA, Bakker J, Kammenga JE: Mapping phenotypic plasticity and genotype-environment interactions affecting life-history traits in Caenorhabditis elegans. Heredity. 2007, 98 (1): 28-37. 10.1038/sj.hdy.6800894.View ArticlePubMedGoogle Scholar
- Kammenga JE, Doroszuk A, Riksen JA, Hazendonk E, Spiridon L, Petrescu AJ, Tijsterman M, Plasterk RH, Bakker J: A Caenorhabditis elegans wild type defies the temperature-size rule owing to a single nucleotide polymorphism in tra-3. PLoS Genet. 2007, 3 (3): e34-10.1371/journal.pgen.0030034.View ArticlePubMedPubMed CentralGoogle Scholar
- Li Y, Alvarez OA, Gutteling EW, Tijsterman M, Fu J, Riksen JA, Hazendonk E, Prins P, Plasterk RH, Jansen RC, et al: Mapping determinants of gene expression plasticity by genetical genomics in C. elegans. PLoS Genet. 2006, 2 (12): e222-10.1371/journal.pgen.0020222.View ArticlePubMedPubMed CentralGoogle Scholar
- Li Y, Breitling R, Snoek LB, van der Velde KJ, Swertz MA, Riksen J, Jansen RC, Kammenga JE: Global genetic robustness of the alternative splicing machinery in Caenorhabditis elegans. Genetics. 2010, 186 (1): 405-410. 10.1534/genetics.110.119677.View ArticlePubMedPubMed CentralGoogle Scholar
- Vinuela A, Snoek LB, Riksen JA, Kammenga JE: Genome-wide gene expression regulation as a function of genotype and age in C. elegans. Genome Res. 2010, 20 (7): 929-937. 10.1101/gr.102160.109.View ArticlePubMedPubMed CentralGoogle Scholar
- Tijsterman M, Okihara KL, Thijssen K, Plasterk RH: PPW-1, a PAZ/PIWI protein required for efficient germline RNAi, is defective in a natural isolate of C. elegans. Curr Biol. 2002, 12 (17): 1535-1540. 10.1016/S0960-9822(02)01110-7.View ArticlePubMedGoogle Scholar
- Miller LM, Gallegos ME, Morisseau BA, Kim SK: lin-31, a Caenorhabditis elegans HNF-3/fork head transcription factor homolog, specifies three alternative cell fates in vulval development. Genes Dev. 1993, 7 (6): 933-947. 10.1101/gad.7.6.933.View ArticlePubMedGoogle Scholar
- Parrish S, Fire A: Distinct roles for RDE-1 and RDE-4 during RNA interference in Caenorhabditis elegans. Rna. 2001, 7 (10): 1397-1402.PubMedPubMed CentralGoogle Scholar
- Tabara H, Sarkissian M, Kelly WG, Fleenor J, Grishok A, Timmons L, Fire A, Mello CC: The rde-1 gene, RNA interference, and transposon silencing in C. elegans. Cell. 1999, 99 (2): 123-132. 10.1016/S0092-8674(00)81644-X.View ArticlePubMedGoogle Scholar
- Sundaram MV: RTK/Ras/MAPK signaling. WormBook. 2006, 1-19.Google Scholar
- Green RA, Kao HL, Audhya A, Arur S, Mayers JR, Fridolfsson HN, Schulman M, Schloissnig S, Niessen S, Laband K, et al: A high-resolution C. elegans essential gene network based on phenotypic profiling of a complex tissue. Cell. 2011, 145 (3): 470-482. 10.1016/j.cell.2011.03.037.View ArticlePubMedPubMed CentralGoogle Scholar
- White AG, Cipriani PG, Kao HL, Lees B, Geiger D, Sontag E, Gunsalus KC, Piano F: Rapid and accurate developmental stage recognition of C. elegans from highthroughput image data. IEEE CVPR. 2010, 3089-3096.Google Scholar
- Brenner S: The genetics of Caenorhabditis elegans. Genetics. 1974, 77 (1): 71-94.PubMedPubMed CentralGoogle Scholar
- Fan JB, Oliphant A, Shen R, Kermani BG, Garcia F, Gunderson KL, Hansen M, Steemers F, Butler SL, Deloukas P, et al: Highly parallel SNP genotyping. Cold Spring Harb Symp Quant Biol. 2003, 68: 69-78. 10.1101/sqb.2003.68.69.View ArticlePubMedGoogle Scholar