Quantitative high-throughput analysis of synthetic genetic interactions in Caenorhabditis elegans by RNA interference.

Biological processes are highly dynamic but the current representation of molecular networks is static and largely qualitative. To investigate the dynamic property of genetic networks, a novel quantitative high-throughput method based on RNA interference and capable of calculating the relevance of each interaction, was developed. With this approach, it will be possible to identify not only the components of a network, but also to investigate quantitatively how network and biological processes react to perturbations. As a first application of this method, the genetic interactions of a weak loss-of-function mutation in the gene efl-1/E2F with all the genes of chromosome III were investigated during embryonic development of Caenorhabditis elegans. Fifteen synthetic genetic interactions of efl-1/E2F with the genes of chromosome III were detected, measured and ranked by statistical relevance.

Introduction

Genes and their products are part of a complex system of physical and genetic interactions that regulate cell activity. Therefore, the function of genes can be fully understood only if each gene is analyzed in the context of the module or network of membership [1,2].

Molecular networks are often depicted as an interaction web of molecules [3,4] or genes [5] that describe the possible interactions between molecules. However, these representations of molecular networks have the limit of being static and largely qualitative. For instance, we cannot predict how the genetic interactions of a gene can be modulated over time in relation to the other components of that same module, but only whether an interactions is active or not. Thus, the dynamic behavior of a genetic network cannot be described if it is perturbed by mutations (e.g. tumors) or external factors (e.g. chemicals). To solve this bottleneck, a new quantitative high-throughput RNAi [6] method was developed to measure the value of each interaction at the genome level, after a specific perturbation.

The screening for synthetic phenotype is a powerful method to detect genetic interactions. For instance, two genes may not display phenotype if targeted individually by RNAi, however, they could have a lethal phenotype when both genes are targeted together: this phenotype is defined as synthetic.

As a first application of this method, the synthetic genetic interactions of the gene efl-1, the worm homolog of the human E2F, with all genes of the chromosome III were quantified during embryonic development of Caenorhabditis elegans after RNAi perturbations. E2F is part of the retinoblastoma complex and alterations of its function determine the insurgence of tumors and other human diseases [7,8]. Following the application of this method, 15 genes of chromosome III that genetically interact with efl-1/E2F were detected and ranked by statistical relevance.

Results

Method outline

In order to obtain quantitative data at the genome level, worms were grown in liquid culture in 96-well plates targeting each gene by RNAi feeding library [9-12]. The RNAi feeding library [10,12] is largely used and it is as effective as the other methods used for RNAi delivery in C. elegans [13]. After purification of embryos through two filtering steps (maintaining the original 96-well format), images of the embryos, soon after the purification and after 24 h were taken, with an automatic microscope. Since, images were processed with computer software, there was the capacity to discriminate and quantify the embryos. Comparing the number of embryos at time 0 and time 24, the lethal effect of treatment on embryos was measured. Comparing the differences of embryonic lethality between the wild-type and the mutated strain JJ1549 genotype efl-1 (se1), the value of each synthetic interaction was detected and calculated.

Embryo isolation and quantification

After 96 h of worms grown in 96-well plates (liquid culture), the content of each plate was transferred to a 96-well Millipore nylon filter plates (MANMN2050), mounted on top of Costar assay block plates. The larvae (L1/L2) are small enough to pass through the filter mesh with an active behavior; instead, adults and embryos cannot pass through this mesh size. In addition, plates were centrifuged for 30 s at 1000 rpm to collect larvae (L1/L2) and the medium with bacteria at the bottom of the block plates. Immediately fallowing centrifugation, 150 μl of M9 buffer with levamisole was added in each well of the filter plates to resuspend the adult worms and embryos. Levimasole serves to paralyse the worms, but does not penetrate the eggshell of the embryos. After 3 min, plates were centrifuged for 15 s to remove the buffer with levimasole. Filter plates were stuck onto a PCR plastic lid to close the mesh to allow resuspension of the content of each well in 200 μl of NGM plus 4 mM concentration of IPTG and 100 μg/ml ampicillin and fungisone. The content of each plate (buffer, embryos and paralysed adults) was then transferred to 96-well Millipore nylon filter plates (MANMN4050), mounted on a 96-well flat-bottomed tissue culture plates. In this last step, without centrifugation, the embryos pass through the mesh, instead, the paralysed worms remained on the mesh. At the end of this procedure, a purified culture of embryos at the bottom of each well of the plates was obtained. In addition, the embryos adhered firmly to the bottom of the plate, without changing position during all successive experimental steps, allowing to monitor whether a specific embryo has hatched or not. In all steps, the original 96-well format array of the RNAi feeding library was preserved. Soon after the purification of embryos, images of the embryos of each well were taken using an automatic stage Olympus microscope. Plates containing embryos were incubated at 20 ° C and following 24 h, images of the embryos were repeated. To quantify embryos, macros that automate the process of counting were developed. The purification of embryos by the filtering steps facilitates images analysis by computer software.

Validation of the automatic counting

In order to validate the accuracy of the automatic quantization, embryos were counted by eye in addition to the automatic method on the NL2099 strain, following RNAi treatments of a random sub-set of 68 genes of the chromosome III. This automatic method was found to correctly count the number of embryos, discriminating the embryos from larvae, egg shells and detritus (Figs. 1 and 2).

Distribution of the frequency of embryonic lethality in the chromosome III

All genes of chromosome III were silenced by RNAi in the wild-type genotype and in the mutated background of the JJ1549 genotype efl-1 (se1). The profile frequency of synthetic lethality in the wild-type and JJ1549 was found to be similar but the average level of embryonic lethality was different. In fact, in the wild-type, embryonic lethality was 21% compared to the mutated background whereas the average lethality was increased to 27% (Fig. 3). Comparing the results of each gene targeted by RNAi in the wild-type and the JJ1549 strain, differences in percentage of embryonic lethality between the two strains were statistically analyzed. Fifteen synthetic genetic interactions were found to be statistically significant, ANOVA, F = 1.64 df = 63,412, p < 0.0001 (Fig. 4, Table 1).

Fig. 3

Frequency distribution of embryonic lethality obtained by RNAi in the wild-type strain N2 (A) and in the mutated strain JJ1549 (B) of the full chromosome III.

Fig. 4

Linear regression between the mean percentage of embryonic lethality between the wild type strain, N2 and the mutated strain JJ1549, n = 2084, R2 = 0.58, p < 0.0001. The top left triangle shows the genes with a synthetic lethality value exceeding 20%.

Table 1

Analysis of embryonic synthetic lethality ranked by statistical significance (one-way ANOVA and Bonferroni's multiple comparison tests)

Sequence name	Gene name	Emb. let. JJ1549	n	Emb. let. N2	n	Diff. JJ1549-N2	p summary	t
ZK632.6	cnx-1	0.95 ± 0.02	7	0.31 ± 0.07	8	0.64 ± 0.07	⁎⁎⁎	7,13
F54F2.5	Ztf-1	0.83 ± 0.04	11	0.23 ± 0.04	12	0.60 ± 0.06	⁎⁎⁎	8,31
T26A5.8		0.65 ± 0.08	6	0.18 ± 0.05	6	0.47 ± 0.10	⁎⁎⁎	4,73
T24C4.1	ucr-2.3	0.64 ± 0.08	6	0.18 ± 0.05	7	0.46 ± 0.09	⁎⁎⁎	4,80
C14B1.8		0.57 ± 0.04	11	0.17 ± 0.03	9	0.40 ± 0.05	⁎⁎⁎	5,09
K10D2.4		0.57 ± 0.06	7	0.18 ± 0.03	8	0.39 ± 0.07	⁎⁎⁎	4,34
Y55B1A_115.c	Mat-3	0.72 ± 0.06	7	0.36 ± 0.09	9	0.37 ± 0.11	⁎⁎⁎	4,20
C34E10.2	Gop-2	0.64 ± 0.08	6	0.24 ± 0.04	7	0.40 ± 0.09	⁎⁎	4,15
T16G12.5		0.54 ± 0.09	7	0.19 ± 0.03	9	0.35 ± 0.09	⁎⁎	4,02
C05D11.12	let-721	0.85 ± 0.06	7	0.51 ± 0.07	8	0.34 ± 0.09	⁎⁎	3,81
T04A8.7		0.80 ± 0.06	6	0.47 ± 0.04	9	0.33 ± 0.07	⁎⁎	3,66
H19M22.2	let-805	0.60 ± 0.07	7	0.29 ± 0.07	7	0.31 ± 0.10	⁎	3,33
F31E3.5	eft-3	0.50 ± 0.07	7	0.20 ± 0.05	9	0.30 ± 0.09	⁎	3,43
ZC21.3		0.53 ± 0.07	7	0.25 ± 0.06	9	0.28 ± 0.09	⁎	3,18
R151.9	pfd-5	0.52 ± 0.07	7	0.24 ± 0.04	9	0.28 ± 0.08	⁎	3,23

The table shows the percentage of lethality and the differences in embryonic lethality between the wild-type and mutated strain ± s.e.m.

Validation of results

In order to validate this new method, experiments were repeated according to the standard protocol on plates [10,12] for the genes selected. On NGM agar plates 13 of the 15 genes detected with the new method, considering all the possible phenotypes, were found to again present with the synthetic phenotype (Table 2). However, 2 genes (C34E10.2 and H19M22.2) did not show any difference between the two strains. The embryonic phenotype of the gene H19M22.2 was not detectable, because the worms were strongly sterile on NGM agar plates.

Table 2

Validation of results by comparing with a previously standard method on NGM agar plates

Sequence name	Diff. JJ1549-N2	p summary	N2 emb	JJ1549 emb	N2 ste	JJ1549 ste	N2 gro	JJ1549 gro	N2 pep	JJ1549 pep
ZK632.6	0.64 ± 0.07	⁎⁎⁎	0	0	0	0	0	1
F54F2.5	0.60 ± 0.06	⁎⁎⁎	0	70	0	1	0	0
T26A5.8	0.47 ± 0.10	⁎⁎⁎	0	50	0	0	0	1
T24C4.1	0.46 ± 0.09	⁎⁎⁎	0	80	0	2	0	0
C34E10.2	0.40 ± 0.09	⁎⁎	50	50	0	0	2	2
C14B1.8	0.40 ± 0.05	⁎⁎⁎	0	40	0	0	0	0
K10D2.4	0.39 ± 0.07	⁎⁎⁎	0	40	0	0	0	0
Y48A6C.5	0.37 ± 0.11	ns	0	0	0	1	0	3
Y55B1A_115.c	0.37 ± 0.11	⁎⁎⁎	90	90	0	0	0	1
T17E9.1	0.35 ± 0.13	ns	10	10	1	1	0	0
T16G12.5	0.35 ± 0.09	⁎⁎	0	20	0	0	0	0
C05D11.12	0.34 ± 0.09	⁎⁎	80	90	0	0	1	0
T04A8.7	0.33 ± 0.07	⁎⁎	50	80	0	0	0	2
R01H10.1	0.33 ± 0.07	ns	90	100	0	0	0	0
H19M22.2	0.31 ± 0.10	⁎	-	-	3	3	0	0	lvl	lvl
F31E3.5	0.30 ± 0.09	ns	0	0	0	0	0	1
Y119D3_456.a	0.28 ± 0.17	ns	0	10	0	0	0	1
ZC21.3	0.28 ± 0.09	⁎	10	10	0	0	0	2
T07C4.7	0.28 ± 0.09	ns	70	90	0	0	2	2	rup
R151.9	0.28 ± 0.08	⁎	40	50	0	0	0	0
ZK112.2	0.27 ± 0.08	ns	0	10	0	0	0	0
ZK637.8	0.26 ± 0.09	ns	-	-	3	3	0	0
F31E3.3	0.26 ± 0.07	⁎	0	0	0	0	0	0
C34E10.6	0.26 ± 0.07	ns	-	-	3	3	3	3
Y76A2A.2	0.25 ± 0.10	ns	0	15	0	0	1	1	lon, adl	adl
Y55B1A_122.c	0.25 ± 0.10	ns	0	10	0	0	0	1
F43D9.3	0.24 ± 0.11	ns	-	-	3	3	0	0
K01G5.7	0.23 ± 0.11	ns	80	100	0	0	0	0
R01H2.3	0.23 ± 0.09	ns	0	10	0	0	0	0
C36E8.5	0.20 ± 0.10	ns	50	90	0	0	0	0	pvl, rup	pvl
Y71H2_375.b	0.20 ± 0.09	ns	0	100	0	0	0	3
F54C8.3	0.19 ± 0.12	ns	50	10	0	0	0	1
T28A8.3	0.17 ± 0.08	ns	0	10	0	0	0	1
W04D12.1	0.16 ± 0.11	ns	0	10	0	0	0	0
Y37D8A.10	0.16 ± 0.06	ns	5	0	3	3	1	1	lvl	lvl
F10C5.1	0.15 ± 0.09	ns	80	80	0	0	1	1	pvl
Y66A7A.1	0.13 ± 0.13	ns	0	0	0	0	0	1
Y71H2_389.e	0.10 ± 0.09	ns	-	-	3	3	0	0
T12D8.6	0.08 ± 0.13	ns	100	100	3	3	0	1	unc, pch, pvl	rup

The embryonic lethality was evaluated qualitatively (score from 0 to 100). For sterility and growing defect using three classes: 1 = low; 2 = medium; 3 = high. Legend: emb: embryonic lethality, ste = sterility, gro = growing defect, pep = post embryonic phenotype (lvl = larval lethal, rup = ruptured, lon = long, adl = adult lethal, pvl = protruded vulva, unc = uncoordinated, pch = patched). Diff. JJ1549-N2 = percentage of the differences in embryonic lethality between the wild-type and the mutated strain ± s.e.m. after repeated experiments.

Discussion

The ideal features of a genomic phenotypic screening should be a high-throughput method providing quantitative read-out. A new quantitative high-throughput method to conduct a phenotypic analysis of about 1200 genes per day was optimized. Previously described methods [5,10,12] required a manual scoring of the phenotypes and were largely qualitative. With this novel quantitative method it is possible to screen the full genome of C. elegans measuring the phenotypic effect for each gene targeted by RNAi. To obtain the same data by manual counting would be extremely time consuming and a strategy not considered practical. In addition, this method can be fully automatized and extended to other phenotypes such as determining the level of sterility or identifying growth defects.

This new method was able to detect all genetic interactions found with the standard protocol, showing a good overlap between the new method and the standard protocol on NGM plates. Regardless, there were some differences that could be explained with an environmental effect (liquid versus agar plates) upon gene activity. In addition, experiments performed in liquid were performed at the population level (40 worms per well), on NGM agar plates, according to the standard protocol [10,12], few worms were used, increasing the probability of less accurate detection of phenotypes. In the worst scenario case, considering false positive genes that were detected with the novel method, but not with the standard method, we can accurately estimate 13% of genes that are false positives.

The development of a quantitative high-throughput method has important applications: firstly, novel genetic interactions can be identified and measured quantitatively in different mutated or natural strains. Secondly, experiments can be performed to examine how the relevance of genetic interactions in a network can be modulated if perturbed by mutations or external factors, quantifying exactly the phenotypic effect of specific perturbations and thus, the dynamic response of a genetic network.

As a first application of this method, 15 genes of the chromosome III of C. elegans that genetically interact with the transcription factor efl-1 were identified and ranked by their statistical relevance. Some of these interactions have been previously described to interact with efl-1 (T26A5.8 [5] and mat-3 [14]. In addition, thirteen interactions with genes of unknown function or previously unknown to interact with efl-1 were also identified. Some of these genes are involved in the insurgence of human disease; for instance, T04A8.7 is orthologous to human acid beta-d-glucosidase gene (GBA), which when mutated leads to Gaucher disease [15,16]. Considering that complex diseases are characterized by the alteration of many different genes, the identification of new interactions with a gene involved in the insurgence of a disease could provide a significant contribution to identifying key players involved in the same disease in addition to potentially identifying novel drug targets.

Methods

Worms' growth

Worms were grown on Nematode Growth Medium (NGM) agar plates seeded with E. coli OP50 strain at 20 °C. Worms were collected by washing plates with M9 buffer. To select L1 larvae, the different stages of worm population were filtered using a 96-well Millipore nylon filter plate (MANMN1150), mounted on top of a Costar assay block plate followed by a 60 s centrifugation step at 1000 rpm. The staged-larvae were then transferred again onto NGM agar plates seeded with E. coli OP50 strain allowing the larvae to grow at 20 °C until development to L3/L4 stages. The larvae were then collected by washing the plates with M9 buffer and centrifuged twice for 60 s at 1000 rpm, to clean worms from bacteria. The larvae were then resuspended in M9 buffer and transferred in to experimental 96-well plates.

Culture of bacterial RNAi feeding strains

Bacterial RNAi feeding clones were grown in 96-well format Costar assay plates (2 ml deep) in 800 μl of 2XTY medium per well and 100 μg/ml ampicillin at 37 °C at 220 rpm for 16 h. Bacteria were induced to express the dsRNA by incubating them for 1 h at 37 °C at 220 rpm with 4 mM isopropyl-beta-d-thiogalactopyranoside (IPTG). Following centrifugation for 5 min at 3500 rpm, bacteria were pelleted to the bottom of the plates. The supernatant was removed and bacteria resuspended in a 180 μl NGM plus 4 mM concentration of IPTG and 100 μg/ml ampicillin and fungisone. Forty larvae at L3/L4 development stage were added to each well and worms were grown for 96 h at 16 °C on a horizontal shaker at 150 rpm.

Strains used

For this work the wild type N2 strain, the JJ1549 strain, genotype efl-1 (se1) and the NL2099 strain, genotype rrf-3 (PK1426) were used. The JJ1549 strain is a weak loss-of-function mutant and it has already been used in RNAi screening [5]. The JJ1549 strain has a background of embryonic lethality of only 6% more than the wild-type (see Results section) and thus it is a good strain to use in synthetic genetic screening for embryonic lethality. The strains of worms and the E. coli OP50 strain were obtained from the Caenorhabditis Genetics Center, University of Minnesota, USA.

Data analysis

All experiments were performed independently on different days. The screening of the entire chromosome III for N2 was repeated 5 and 3 times for JJ1549. Genes that showed at least a difference in embryonic lethality between N2 and JJ1549 greater than 20% with a variation less than 25% were selected. From the analysis all well with less then 10 embryos were excluded, since, the worms have shown to be strongly sterile. These criteria were arbitrary and could be modified in a manner to be more or less conservative. The experiments for the selected genes were repeated at least for another 4 times for the two stains JJ1549 and N2. After passing Kolmogorov–Smirnov normality test, one-way ANOVA and Bonferroni's multiple comparison tests were performed to determine the statistical significance of results. R01H10.1, T17E9.1, W04D12.1, Y119D3_456.a, Y48A6C.5, Y66A7A.1, and ZK112.2 did not pass the normality test and were analyzed with Kruskal–Wallis test followed by Dunn's multiple comparison test. Statistical analysis was performed using Graph Pad Prism 5 software, USA.