A Delicate Balance Between Bacterial Iron and Reactive Oxygen Species Supports Optimal C. elegans Development

SUMMARY

Iron is an essential micronutrient for all forms of life; low levels of iron cause human disease, while too much iron is toxic. Low iron levels induce reactive oxygen species (ROS) by disruption of the heme and iron-sulfur cluster-dependent electron transport chain (ETC). To identify bacterial metabolites that affect development, we screened the Keio Escherichia coli collection, and uncovered 244 gene deletion mutants that slow Caenorhabditis elegans development. Several of these genes encode members of the ETC cytochrome bo oxidase complex, as well as iron importers. Surprisingly, either iron or anti-oxidant supplementation reversed the developmental delay. This suggests that low bacterial iron results in high bacterial ROS and vice versa, which causes oxidative stress in C. elegans that subsequently impairs mitochondrial function and delays development. Our data indicate that the bacterial diets of C. elegans provide precisely tailored amounts of iron to support proper development.

Graphical Abstract

eTOC blurb

Zhang et al. identify more than 200 E. coli deletion mutants that, when fed to C. elegans as diet, slow animal development. This phenotype can be rescued by iron or anti-oxidant supplementation, indicating that low bacterial iron causes oxidative stress in the animal, which in turn decelerates development.

INTRODUCTION

In addition to providing macronutrients, our diet is an important source of essential micronutrients such as vitamins and trace elements that are required for a broad range of cellular and physiological functions. Such vitamins and trace elements often support metabolic enzyme function. For instance, vitamin B12 is an essential cofactor for two metabolic enzymes to enable flux through the methionine/S-adenosylmethionine or one-carbon cycle, and to break down the short chain fatty acid propionate (Banerjee and Ragsdale, 2003; Krautler, 2012). The complexity of human diets, the genetic heterogeneity of humans and the additional complexity of resident microbiota, make it challenging to systematically dissect the effects of individual nutrients on human development. While mammalian model systems such as mice have been extremely useful, the microbiota of these model systems have mostly been studied with complex populations of natural bacteria rather than by the much easier to deconvolute approach of testing one bacterial species at a time. Further, mammalian model systems are limited in scale, throughput and cost.

The nematode Caenorhabditis elegans is a bacterivore that can thrive on different bacterial diets that provide necessary macronutrients, as well as vitamins and trace elements. The rapid development of C. elegans makes it suitable to study the mechanisms by which different nutrients affect animal development using large-scale, and high-throughput experiments. After hatching, C. elegans progresses through four larval stages (L1-L4) and four molting stages (M1-M4) to adulthood in about two days (Byerly et al., 1976). Different bacterial diets can greatly affect C. elegans life history traits, including developmental rate (MacNeil and Walhout, 2013; Zhang et al., 2017). For instance, when fed a diet of the soil bacteria Comamonas aquatica DA1877 the animals develop much faster than when fed the standard laboratory diet of Escherichia coli OP50 (MacNeil et al., 2013). By comprehensive forward and reverse C. elegans and bacterial genetics, this difference in developmental rate control by bacterial diet was found to be due to the fact that C. aquatica DA1877 supplies the animal with high levels of vitamin B12, while E. coli OP50 does not (Watson et al., 2013; Watson et al., 2014).

Iron is an essential trace element for all forms of life that is involved in multiple cellular processes such as oxygen transport, energy metabolism, DNA synthesis, and transcription regulation (Wang and Pantopoulos, 2011). Under aerobic conditions, ferrous iron (Fe²⁺) is oxidized to ferric iron (Fe³⁺) which is highly insoluble and therefore has low bioavailability. Maintaining iron homeostasis is crucial for general health. While iron deficiency can cause serious health issues such as anemia, iron becomes toxic at high physiological concentrations (Anderson and Frazer, 2017; Andrews, 2000). Specifically, high levels of iron facilitate the Fenton reaction in which ferrous iron is oxidized to ferric iron upon the conversion of hydrogen peroxide to the highly reactive oxygen species (ROS) hydroxyl radical (Winterbourn, 1995). Conversely, low levels of iron result in the perturbation of iron-dependent processes such as oxidative phosphorylation, thereby generating high levels of ROS. Therefore, free intracellular iron is tightly regulated at cellular and organismal levels (Andrews and Schmidt, 2007).

Living cells constantly encounter ROS, which are generated under normal physiological conditions in an oxygen-rich environment (Dixon and Stockwell, 2014; Gorlach et al., 2015). ROS can be beneficial in host defense, developmental processes and cell survival pathways and function as a signaling molecule (Bystrom et al., 2014; Dixon and Stockwell, 2014). However, excess ROS is toxic due to the ability of different ROS to react with and damage DNA, proteins and lipids (Cross et al., 1987; Dizdaroglu and Jaruga, 2012). Organisms have evolved complex mechanisms to mitigate the damaging effects of active oxygen and to keep balanced levels of ROS. These include the production of small molecule anti-oxidants such as glutathione, the use of dietary anti-oxidants such as vitamins C and E, and the synthesis of anti-oxidant enzymes such as catalase, superoxide dismutase, and thioredoxin (Gorlach et al., 2015).

Since different bacterial diets can affect C. elegans growth, we asked whether we could identify additional metabolites that modify the animal’s developmental rate. We reasoned that by identifying mutants in bacterial genes encoding metabolic enzymes, we may be able to infer the lack or overproduction of metabolites that affect C. elegans development. To do so, we screened the E. coli Keio collection that consists of 3,985 strains, each harboring a kanamycin resistance cassette in a single, non-essential gene that are hereafter referred to as deletion mutants (Baba et al., 2006). We identified 244 E. coli deletion mutants that slow C. elegans development. These deletion mutants involve genes encoding proteins that function in different cellular processes, including energy metabolism and iron transport. Remarkably, the delay in C. elegans development elicited by most of these mutant strains could be rescued by supplementation of either the anti-oxidant N-acetylcysteine or by supplementing iron. Altogether, our data show that the delicate balance between ROS and iron, as provided by the bacterial diet, supports optimal C. elegans development. These findings further establish the utility of C. elegans-bacteria interspecies systems in understanding the connections between diet and life history traits.

RESULTS

A High-throughput Screen for E. coli Single Gene Deletions that Affect the Rate of C. elegans Development

We first asked whether we could use C. elegans body size at a fixed time point after L1 feeding as a proxy for developmental rate. We grew wild type, synchronized C. elegans L1 larvae on wild type E. coli BW25113 bacteria (the Keio parent strain) for 60 hours and measured the animal’s body size using ImageJ (Figure 1A) (Schindelin et al., 2012). To capture the greatest differences in growth in C. elegans fed different bacteria, we decided to take 48 hours, when wild type animals reach the young adult stage, as a time point for our next experiments (Figure 1A). We validated that we can observe differences in developmental rate in this assay by recapitulating the acceleration of C. elegans development when animals were fed C. aquatica DA1877 relative to animals fed the standard laboratory diet of E. coli OP50 (Figure 1B) (MacNeil et al., 2013). Finally, we determined the variance in body size after 48 hours post-L1 feeding of animals fed wild type E. coli BW25113 bacteria. Two independent experiments found the variability to be less than 10% (Figure 1C). Therefore, we used a 48-hour time point and a change in body size greater than 10% as a condition to screen the E. coli Keio single gene deletion collection (Baba et al., 2006). As a primary screen, the collection was screened five times and only those deletion strains that conferred, on average, a greater than 10% reduction in C. elegans body size were considered (we did not find any strains that reproducibly increased C. elegans body size). These deletion strains were screened again in triplicate, on larger plates and with more animals. Deletion strains that conferred a body size decrease greater than 10% in each of these three experiments were genotyped for the deleted gene (see Methods). The 244 verified strains were considered final positives in the screen and are henceforth referred to as “E. coli hits”. The range of C. elegans growth inhibition for the E. coli hits was large; from 90% to 30% of wild type C. elegans body size (Figure 1D, Table S1).

Figure 1. Genome-Scale Screen to Identify E. coli Single Gene Deletion Mutants that Affect C. elegans Development.

(A) Representative growth curve of wild type C. elegans fed wild type E. coli BW25113. Animal body size was measured by ImageJ. The dashed red line indicates the 48-hour time point used in the screen. Error bars indicate ± standard error of the mean (SEM). Number of animals used for 0, 12, 24, 36, 48, and 60 hours are 30, 34, 45, 31, 54, and 43, respectively.

(B) Relative body size of C. elegans grown for 48 hours on Comamonas aquatica DA1877 relative to animals grown on the standard laboratory diet of Escherichia coli OP50. Error bars indicate ± SEM.

(C) Relative body size of individual C. elegans fed the wild type E. coli BW25113 strain. Two independent experiments are shown. Each data point indicates the body size of a single animal relative to the average body size of the population. The red dashed line indicates the relative body size of 0.9, a 10% difference from the average wild type body size of 1.0.

(D) Relative average body size of C. elegans fed each of 244 E. coli deletion strains (hits) that confer a >10% reduction in body size. The red dashed line is the same as described in part C.

(E) Comparison of the effect of E. coli deletion strains on C. elegans development versus growth of each E. coli strain. The mean of three experiments is shown. Error bars indicate standard deviation (SD). See Figure S1A for details.

(F) Luciferase developmental rate experiment for C. elegans fed a selection of 44 E. coli hits. The top line indicates animals fed the wild type E. coli BW25113 strain. L = larval stage; M = molt. The dashed red line indicates the time point at which C. elegans fed wild type E. coli BW25113 completes development (48 hours). A representative of three experiments is shown. See also Figure S2A.

Even though the animals were never starved in our experiments, we wondered whether the effects of bacterial gene deletions on C. elegans development were correlated with bacterial growth. Careful comparison between bacterial growth and the effect on C. elegans development showed that there is no correlation between the two; some bacterial strains grow slowly but have a modest effect on C. elegans growth rate while other strains elicit a strong delay in animal growth rate whilst exhibiting no change in growth (Figure 1E and Figure S1A, compare colored bars). This indicates that specific bacterial metabolites rather than growth or biomass affect C. elegans developmental rate.

To validate that the E. coli hits slow rather than arrest C. elegans development, we incubated animals fed E. coli hits for 72 hours or more and found that all had completed development and most started to produce offspring (Figure S1B, data not shown). Finally, we performed luciferase-based development assays (Olmedo et al., 2015) and found that C. elegans fed 44 of 45 E. coli hits tested developed slower than animals fed wild type E. coli (Figure 1F, Figure S2A). Importantly, the animals had sufficient amounts of bacteria to feed on during the experiments, and we did not observe any dauer or developmental arrest phenotypes in C. elegans, indicating that the slow developmental rate was not because of starvation.

The E. coli Mutants that Reduce C. elegans Developmental Rate Involve a Broad Variety of Functions

Gene ontology analysis of the 244 hits did not reveal any informative enriched functional categories. However, systematic manual annotation showed that the largest proportion of hits (70 or 29%) involve deletions in metabolic genes, followed by transporters (31 or 13%)(Figure 2A, Table S1). Other categories include transcription (19), DNA repair (5) and peptidases (9). Thirteen hits are annotated as pseudogenes, and therefore the effects of these mutants on C. elegans development is difficult to interpret (Table S1). Ten of these are not in operons or are in operons with other pseudogenes and can therefore not be functionally annotated. Of the remaining three, wbbL may affect the downstream gene gnd, which is also a hit in our study and is involved in the pentose phosphate pathway. Further, we have another hit in the wwb operon, wbbI. Within the category of metabolism, different types of genes were found (Figure 2A). Several mutants function in anabolic biomass generation, such as genes involved in nucleotide metabolism (9), fatty acid metabolism (6) and the pentose phosphate pathway (2). Other genes encode proteins involved in energy production, including those involved in pyruvate metabolism (5), the electron transport chain (ETC)(7), and nitrate reduction (4).

Figure 2. Deletion of Different E. coli Metabolic Genes Slow C. elegans Development.

(A) Pie chart showing manual functional annotations of E. coli hits that slow C. elegans development (left) and different categories of metabolic enzymes (right) among the E. coli hits.

(B) E. coli electron transport chain gene deletion strains slow C. elegans development. A representative of three experiments is shown. Error bars indicate ± SEM. P≤0.05 for all of the mutants relative to wild type by Student’s t-test.

(C) E. coli electron transport chain gene deletion strains contain more ROS compared to wild type E. coli. The mean of three experiments is shown. Error bars indicate ± SEM.

(D) Supplementation with 10 mM of the anti-oxidant N-acetylcysteine rescues the C. elegans developmental delay caused by E. coli electron transport chain mutants. The mean of three experiments is shown. For the electron transport chain mutants, all N-acetylcysteine treated animals displayed significantly increased body size relative to untreated animals. A representative of three experiments is shown. Error bars indicate ± SEM.*P≤0.05 by Student’s t-test; n.s., not significant.

(E) Cartoon illustrating the framework for determining statistical associations between metabolite increases or decreases in E. coli deletion strains and their effect on C. elegans developmental rate. The 1513 E. coli genes include the 244 hits that slow C. elegans development. The rest are non-hits for which sufficient body size data was generated. To find metabolite-gene associations, columns of the z-score matrix (ions) were correlated with the body size data (y vector). See methods for details.

(F) Increased bacterial choline (left) and bacterial enterobactin (right) are significantly associated with a reduction in C. elegans developmental rate, based on the correlation analysis defined in (E). Each data point in these ion plots indicates the adjusted z-score of the ion for a particular strain. See also Figure S2B, Tables S1 and S2.

E. coli Mutants in Cytochrome bo Oxidase Complex of the Electron Transport Chain Slow C. elegans Development

Prominent hits in our screen, cyoA, cyoB, cyoC, and cyoD, encode members of the cytochrome bo oxidase complex of the ETC, and sdhC, a succinate:quinone oxidoreductase that is part of both the ETC and the TCA cycle (Figure 2B). A previous study found that feeding cyo mutants to C. elegans causes developmental delay (Govindan et al., 2015). Disruption of oxidative phosphorylation generates high levels of ROS (Dixon and Stockwell, 2014; Sedensky and Morgan, 2006), leading to the prediction that cyo mutants may produce high levels of ROS that in turn slows the growth of C. elegans (Govindan et al., 2015). We found that the cyo and sdhC mutants indeed produce increased levels of ROS (Figure 2C). To further test whether elevated ROS contributes to the C. elegans developmental delay, we added the antioxidant N-Acetylcysteine and found that it partially rescued the developmental delay phenotype (Figure 2D). These results indicate that high bacterial ROS slows C. elegans development.

Metabolites Associated with E. coli Mutants that Delay C. elegans Development

To identify bacterial metabolites that may cause the slow C. elegans developmental rate, either when overproduced or when depleted, we integrated our data with a metabolomic dataset in which metabolite ions were measured for each of the Keio deletion strains (Fuhrer et al., 2017). In the reference study, signal intensities of detectable metabolite ions were measured by high-throughput, non-targeted liquid chromatography-mass spectrometry (LC-MS), and the dataset was presented in the form of z-scores that quantify the deviation of metabolite concentrations for each strain from the median of all strains, including wild type. We first processed this dataset to remove background noise, to focus on strains for which we obtained systematic C. elegans body size measurements, and to evaluate only those ions that were annotated (about half of all detected ions (Fuhrer et al., 2017)). If an ion has a positive z-score for a strain in this dataset, then the parent metabolite of this ion is enriched in that strain, or if the ion has a negative z-score, then the parent metabolite is depleted. Since z-scores indicate the variation of the concentration of a metabolite across all strains, it is possible to analyze the statistical association between relative metabolite levels in different strains, as represented by z-scores of matching ions, and C. elegans body size, based on simple correlation (Figure 2E). After calculating correlations of all ions with body size and multiple hypothesis correction, we identified 42 ions that were significantly associated with C. elegans body size (P≤0.01, FDR<0.1) (Table S2). More than 80% of the significant ions (35 out of 42) anti-correlated with C. elegans body size, which indicates that an increase in the concentration of the source metabolites corresponding to these ions is associated with a decrease in animal growth rate.

The precise metabolite that gave rise to each ion measured in the metabolomics dataset cannot be unambiguously identified. Since different metabolites accumulated in different bacterial mutants could independently affect C. elegans developmental rate, getting a high absolute correlation score for individual metabolites is unlikely. The most statistically significant candidate bacterial metabolites associated with C. elegans developmental rate are choline and cyclic nucleotides, which accumulated in several of the E. coli hits, relative to wild type bacteria (Table S2, Figure 2F). However, the connection of these hits to the production or breakdown of these metabolites was not immediately evident (Figure S2B). This is in contrast to enterobactin (also known as enterochelin), which was a candidate metabolite corresponding to three different ions detected in the mass spectrometry experiment (Fuhrer et al., 2017) (Table S2). There was a significant association between the E. coli hits and an increase in enterobactin (Figure 2F). Enterobactin is a non-ribosomal peptide that is excreted from bacteria and chelates iron in the media (Raymond et al., 2003). The enterobactin-iron complex is then imported into bacteria by Fep proteins, and iron is released from enterobactin by Fes (Crosa and Walsh, 2002; Raymond et al., 2003; Wilson et al., 2016). The genes responsible for the association between increased bacterial enterobactin and delayed C. elegans development include fepC, fepD, fepG and fes (Figure 2F). These observations indicate that, when the import of the enterobactin-iron complex is perturbed and there is a reduction in iron entering the bacterial cytoplasm resulting in low bioavailable iron there is an increase in enterobactin synthesis. Indeed, E. coli is known to regulate ent gene expression in response to low iron (Escolar et al., 1999; McHugh et al., 2003).

Low Bacterial Iron Delays C. elegans Development

Is it low bacterial iron or high bacterial enterobactin that slows C. elegans development? We recovered only one annotated enterobactin biosynthesis gene, ybdZ, in our screen (Table S1). This could indicate that either high or low enterobactin may affect C. elegans development. However, enterobactin levels were not changed significantly in ybdZ mutant bacteria (Fuhrer et al., 2017). YbdZ encodes a small MbtH-like protein predicted to adenylate amino acids required by non-ribosomal peptide synthetases. Experimental evidence about the function of this gene in enterobactin synthesis is limited and contradictory. One study proposed that the YbdZ protein enhances the catalytic function of EntF in vitro (Felnagle et al., 2010), while another study found that enterobactin biosynthesis can efficiently occur without YbdZ (Gehring et al., 1998). Since we did not recover any enterobactin biosynthesis genes in our primary screen (Table S1), we decided to select the relevant mutants and test their effects on C. elegans developmental rate directly and compared them to different fep mutants. Confirming our primary screen results, fep mutants delayed C. elegans development (Figure 3A). However, deletion in none of the five enterobactin biosynthesis genes affected C. elegans development (Figure 3A). Our finding of ybdZ deletion mutants slowing animal development suggests that, while the protein encoded by this gene may increase the catalytic activity of enterobactin biosynthesis (Felnagle et al., 2010), it likely has other functions, potentially related to iron metabolism, as well.

Figure 3. Low or High Iron both Slow C. elegans Development.

(A) While bacterial fep B, C, D, E, and G, and fes mutants affect C. elegans development, fepA and enterobactin synthesis mutants (ent genes) do not. A representative of three experiments is shown. Error bars indicate ± SEM.*P≤0.01 by Student’s t-test.

(B) The iron chelators bipyridine and phenanthroline slow wild type C. elegans development in a dose-dependent manner. All treated samples are statistically significantly reduced in body size relative to the DMSO (vehicle control) treated animals P≤0.0001 by Student’s t-test. A representative of three experiments is shown. Error bars indicate ± SEM.

(C) The slow C. elegans development on a diet of fepG mutant E. coli is rescued by supplementation of 4 mM iron. A representative of three experiments is shown. Data are represented as mean ± SEM. P≤0.05 by Student’s t-test; n.s, not significant.

(D) Excess iron slows wild type C. elegans development and is lethal at the highest dose.

Since deletion of enterobactin biosynthesis genes did not affect C. elegans development, we hypothesized that low levels of bioavailable bacterial iron may explain slow animal growth in C. elegans fed fep or fes mutant bacteria. However, iron cannot be directly measured by LC-MS and therefore was not present in the published metabolomic dataset (Fuhrer et al., 2017). To test whether low iron can delay C. elegans development, we performed several experiments. First, we supplemented C. elegans fed wild type E. coli with the iron chelators bipyridine or phenanthroline and found that this caused a dose-dependent developmental delay (Figure 3B). Second, we found that supplementing C. elegans fed E. coli fepG mutants with 4 mM iron fully rescued the C. elegans developmental delay (Figure 3C). It is well known that high levels of iron can be toxic. Indeed, supplementing animals fed wild type E. coli with increasing concentrations showed that a dose of 8 mM greatly slowed C. elegans growth while a dose of 16 mM was highly toxic (Figure 3D). These data indicate that C. elegans needs bacterial diets with precisely tailored levels of iron to support optimal animal development.

Cross-Talk Between Bacterially Generated Reactive Oxygen Species and Iron Affects C. elegans Development

Surprisingly, when we tested for the specificity of iron rescue, we found that 4 mM iron supplementation rescued the C. elegans developmental rate for most E. coli hits tested (Figure 4A). This indicates that these mutant E. coli strains supply the animal with insufficient iron. While a higher dose of 8 mM iron greatly slowed the development of wild type bacteria, it was able to rescue the developmental delay elicited by a set of E. coli hits that slowed C. elegans development to different degrees (Figure 4B). This suggests that the slow C. elegans growth caused by these E. coli mutants is indeed due to low iron levels and further suggests that most E. coli hits slow C. elegans development by a similar, low-iron-induced mechanism. Interestingly, the slow C. elegans development elicited by E. coli mutants in cytochrome bo oxidase genes was not only rescued by anti-oxidant supplementation (Figure 2D), it was also rescued by iron supplementation (Figure 4C). Are all bacterial mutant effects on C. elegans development caused by low iron and/or high ROS? Compared to a random set of 88 non-hits, E. coli mutants that delay C. elegans development harbor greater levels of ROS (Figure 4D). Moreover, the C. elegans developmental delay phenotype elicited by E. coli hits could be rescued by supplementation with the anti-oxidant N-acetylcysteine (Figure 4E).

Figure 4. Crosstalk Between Low Bacterial Iron and High Bacterial ROS Slows C. elegans Development.

(A) Iron supplementation (4 mM) rescues the slow C. elegans development elicited by most E. coli hits. The average of three biological replicate experiments is shown.

(B) A dose of 8 mM iron greatly slows the development of C. elegans fed wild type bacteria, whilst rescuing developmental rate of animals fed randomly selected set of E. coli hits. A representative of two experiments is shown. Error bars indicate ± SEM.

(C) Iron supplementation (4 mM) rescues the slow development elicited by E. coli electron transport chain mutants. A representative of three experiments is shown and all iron treated animals are significantly larger in body size relative to untreated animals. Data are represented as mean ± SEM.*P≤0.05 by Student’s t-test; n.s, not significant.

(D) ROS levels are elevated in E. coli hits compared to wild type E. coli and non-hits. Dashed red line indicates normalized ROS levels in wild type E. coli. The mean of three biological replicate experiments is shown and error bars indicate ± SD.

(E) The anti-oxidant N-acetylcysteine (10 mM) rescues the slow C. elegans development elicited by most E. coli hits. The average of three biological replicate experiments is shown.

(F) Fluorescent microscopy shows that feeding most E. coli hits to C. elegans activates the hsp-6 promoter, which is sensitive to the mitochondrial unfolded protein response. Insets show corresponding bright field images.

(G) Fluorescent microscopy shows that iron supplementation (4 mM) can turn off the hsp-6 promoter in animals fed E. coli containing the fepG mutant. Insets show corresponding bright field images.

(H) Metabolically inactive bacterial powder made from fepG or cyoD mutant E. coli slows C. elegans development compared to powder made from wild type E. coli. A representative of three experiments is shown and data are represented as mean ± SEM.

How do the E. coli hits delay C. elegans development? We hypothesized that low-iron and/or high-ROS bacterial diets would generate ROS within the animal, thereby interfering with mitochondrial function and slowing development. The hsp-6 promoter is activated by perturbed mitochondrial protein folding and processing (Yoneda et al., 2004) and by ROS (Runkel et al., 2013). We found that the majority of the E. coli hits robustly activate hsp-6 promoter activity (Figure 4F), indicating that these bacterial diets negatively affect C. elegans mitochondrial function, thereby delaying the animal’s development. Iron supplementation repressed hsp-6 promoter activity induction in fepG mutant-fed animals and all other mutants tested (Figure 4G, data not shown). Our data suggest that bacterial hits that affect C. elegans development might have both low iron and high ROS levels. Indeed, fep and fes mutants in which bacterial iron levels are low harbor high levels of ROS (Figure S3A). Further, bacteria respond to oxidative stress by regulating iron homeostasis (Cornelis et al., 2011). Is it high bacterial ROS or low bacterial iron that perturbs C. elegans mitochondrial function and slows development? We generated metabolically inactive bacterial powder (Garcia-Gonzalez et al., 2017) that should reflect the iron content of live bacteria but does not actively produce ROS, and found that powder generated from both fepG and cyoD mutants slowed C. elegans development relative to powder generated from wild type E. coli (Figure 4H). These results indicate that instead of high bacterial ROS, low bacterial iron may be the predominant cause of C. elegans developmental delay and suggests that low dietary iron generates ROS in C. elegans, which affects mitochondrial function and slows development. We tested four more hits from different metabolic pathways as bacterial powder, and confirmed that they slowed C. elegans development, when compared to powder obtained from wild type E. coli (Figure S3B). However, while iron supplementation rescued C. elegans developmental rate with live bacteria, it had little effect on bacterial powder (Figure 4H), indicating that iron supplementation requires active bacterial metabolism to rescue the developmental delay in C. elegans. This is likely because free iron is highly insoluble and not readily absorbed in a bioavailable form by C. elegans (Guerinot and Yi, 1994).

Metabolic Network Analysis Links E. coli Hits to both ROS and Anti-Oxidant Production

The connection to ROS production or iron depletion is straightforward to infer for several of the E. coli. These include the iron import (fep and fes) genes and ETC (cyo) genes. Mutations in ETC genes directly perturb oxidative phosphorylation, which increases ROS production, while mutations in iron import genes result in low bacterial iron and indirectly perturb oxidative phosphorylation activity. However, our data indicate that most E. coli hit strains seem to disturb the delicate balance between dietary iron and ROS production because they generally harbor increased levels of ROS, and because the growth delay in C. elegans they elicit can be rescued by either supplementation of anti-oxidant or iron. The connections of most E. coli hit genes to iron and ROS production was not immediately apparent (Table S1). We therefore asked whether these genes are functionally connected at a systems level. To answer this question, we focused on metabolic genes and transporters. We mapped E. coli hits that encode enzymes or transporters to a published genome-scale metabolic network model of E. coli named iJO1366 (Orth et al., 2011)(Figure 5). This network model includes reactions associated with 52 of the 244 E. coli hits, most of which reside in the core metabolic network (TCA cycle, oxidative phosphorylation, pentose-phosphate pathway), while some participate in peripheral pathways in the immediate vicinity of this core (e.g., nucleic acid metabolism) (Figure 5, Table S1).

Figure 5. Overlay of E. coli metabolic mutant hits on the E. coli metabolic network map.

75% of E. coli metabolic hits (39 of 52) can be mapped to the core metabolic network and its immediate vicinity. A minimal flux distribution that activates all 52 genes was calculated and is indicated by blue arrows.

While pathway visualization on a network reconstruction gives an idea about gene function, it is not sufficient to get a systems-level understanding of how these genes are connected to each other and to the rest of the network, or what type of metabolic functions they carry out as a subnetwork. To develop a better and quantitative understanding of metabolic network function, metabolic reconstructions are converted to mathematical models, which allow constraint-based flux balance analysis (FBA) (Orth et al., 2010). Assuming a nutritional input, constraining network reactions with respect to reversibility, and forcing reactions associated with active genes to carry flux, a metabolic output can be predicted (see Methods). We used FBA to calculate a flux distribution where all reactions associated with E. coli hits, which can be assumed to be active in bacterial metabolism because their deletion results in a C. elegans growth delay, were forced to carry flux, while flux in the rest of the metabolic network was minimized, since we do not have any knowledge regarding the activity of other enzymes and transporters. By doing so, we obtained a minimal flux distribution that connected the abovementioned pathways by directional fluxes (Figure 6A). We then asked if the E. coli hits are overall associated with ROS production at the network level. Using published criteria to determine reactions that can potentially generate ROS (Brynildsen et al., 2013), we found 166 ROS-generating reactions in the E. coli iJO1366 metabolic network model (Table S3). There are 77 genes associated with these reactions and also associated with a C. elegans growth phenotype in our assays (either as hits or as non-hits), 15 of which were found as hits in our screen (P<0.012, hypergeometric test) (Figure 6A). To evaluate ROS generation potential by FBA, we summed the fluxes of all ROS-generating reactions in the predicted flux distribution defined above and divided this sum with total flux in the entire network. The resulting flux ratio was about 0.07, indicating that ROS-generating reactions carried 7% of all flux in the predicted metabolic state (Figure 6B). To establish the significance of this result, we generated 10,000 flux distributions with the abovementioned method, but activating a random set of 52 genes (instead of the 52 E. coli hits) each time. In only about 8% of the randomized cases was the normalized value greater than 0.07 (Figure 6B), which indicates that flux in ROS generating reactions is larger than expected for a random set of 52 genes. While 8% does not reach formal statistical significance, it does indicate an association. To identify statistically significantly enriched pathways, we repeated this flux enrichment analysis for all subsystems in the metabolic network model (Figure 6D, Table S4) and found that E. coli hit genes are significantly associated with oxidative phosphorylation (P=0.031) (Figure 6C and D), nitrogen metabolism (P=0.04), and folate metabolism (P=0.05)(Table S4). The latter is interesting because we did not recover folate metabolism genes in our screen as most of these genes are essential and are therefore not part of the Keio collection (Baba et al., 2006).

Figure 6. Analysis of Bacterial Metabolic Network Indicates Perturbation of Redox Homeostasis in E. coli Hits.

(A) Association between E. coli hit genes and those directly linked to reactions that generate ROS (Table S3).

(B) Total flux (sum of absolute values) in ROS-generating reactions (Table S3) as a percentage of total flux in the entire network. The value shown by the red line was calculated based on the network predicted by activating 52 E. coli hits. The values in the histogram were calculated based on 10,000 networks obtained by activating the same number of (52) random metabolic genes. The P-value indicates total frequency of values to the right of the red line divided by 10,000.

(C) Same as (B) for the flux sum of oxidative phosphorylation pathway instead of ROS-generating reactions.

(D) Enrichment of pathways and metabolites in E. coli metabolic network. The analysis in (C) was extended to pathways other than oxidative phosphorylation.

The connection of oxidative phosphorylation perturbation to ROS production is well understood; this can occur by genetic mutation in the genes involved in the ETC such as cyo genes, or by functional perturbation due to low iron availability (Dixon and Stockwell, 2014). Our modeling analysis cannot predict the latter relationship, as the concentration of iron, or the concentration of any metabolite for that matter, is not taken into account by FBA.

It is interesting that nitrogen metabolism was also enriched in the metabolic network model. Nitrogen metabolism presents an alternative redox mechanism that can result in the generation of reactive nitrogen species (RNS), many of which have similar effects as ROS (Koskenkorva-Frank et al., 2013). Indeed, four genes annotated to be involved in nitrate reduction, narY, narH, narJ and napH, were found as hits in our screen (Table S1). Why would the network associate folate metabolism with redox biology? Recently, it has been shown that a major source of cellular anti-oxidant NADPH generation is by the folate cycle (Fan et al., 2014). This is in addition to the well-known generation of NADPH by the pentose phosphate pathway. Importantly, genes involved in the latter pathway were found in our screen as well (Table S1). Therefore, perturbation of the folate cycle likely results in a reduction in anti-oxidant capacity.

We also found that total flux through many redox metabolites was enriched in the predicted flux distribution, although several were just around a threshold of P=0.05 (Figure 6D, Table S4). Interestingly, most of these are directly related to cellular redox state and anti-oxidant function, including glutathione (GSH, P<0.02), electron carriers (P=0.05), quinol-related molecules (P=0.06), NADH/NAD (P=0.06) and glutaredoxin (P<0.02). Together, our observations indicate that the hits enhance ROS (or RNS) production in E. coli by two mechanisms: directly via the perturbation of energy production by mutation of the genes involved (e.g., cyo genes), or indirectly by low iron availability (e.g., fep and fes genes), or perturbation of anti-oxidant production (e.g., pentose phosphate pathway and glutaredoxin)(Figure 7).

Figure 7. Model of how E. coli mutants can slow C. elegans development.

Deletions of bacterial genes involved in the cytochrome bo oxidase (ETC) or other anti-oxidant genes when fed to C. elegans cause increases in animal ROS leading to mitochondrial dysfunction and slow development. Animal development can be rescued by the addition of antioxidants. Likewise, deletions of the ferric enterobactin import system genes fepB, fepC, fepD, fepE, fepG, and fes cause a decrease in bioavailable iron again leading to mitochondrial impairment and slow animal growth. Slow growth can be rescued by the supplementation of iron and growth can be slowed in wild type animals by iron chelators. Bioavailable iron can react with intracellular hydrogen peroxide to cause an increase in ROS via the Fenton Reaction, and a reduction in bioavailable iron.

DISCUSSION

The bacterivore C. elegans can be fed individual bacterial strains and mutants and provides a facile interspecies system to study the effects of macro- and micronutrients on life history traits such as development and fecundity (MacNeil and Walhout, 2013; Watson et al., 2015; Zhang et al., 2017). In this study, we identified 244 E. coli deletion mutants that, when fed to C. elegans, delay the animal’s development. We did not discover any deletion mutants that increased the developmental rate, indicating that at least for this E. coli species, the wild type bacteria provide optimal nutrition.

Several observations support the high quality of our screen. First, the level of false positives is likely low since we only considered those mutant strains that, in the retest, elicited a change in C. elegans body size of at least 10% in each of the three replicate experiments. However, our first test with only wild type bacteria shows a variance of ~10%, which illustrates the inherent or experimental variability of C. elegans growth upon release from the arrested L1 stage. Second, we uncovered members of the cytochrome bo oxidase complex, which were reported to delay C. elegans development in a previous study (Govindan et al., 2015). The cytochrome bo oxidase is part of the ETC and functions in oxidative phosphorylation. Interestingly, members of other ETC complexes such as complex I (NADH dehydrogenase) were not found, which confirms the previous study as well (Govindan et al., 2015). Third, the metabolic genes and transporters in our dataset are connected in a coherent network that produces ROS. However, it is likely that the level of false negatives (missed genes) is considerable, because of experimental stringency. Indeed, not all genes in the metabolic network related to ROS production were uncovered.

Our results indicate that the balance between bioavailable bacterial iron and ROS is critical. However, while we did measure ROS in the bacterial hits, our inference of low levels of bioavailable iron in these hits is indirect based on (1) the identification of fep and fes genes in our screen, (2) the developmental delay caused by supplementation with iron scavengers, and (3) the rescue of slow C. elegans development by iron supplementation. We extensively tried to obtain more direct measurements of iron in bacteria and C. elegans but neither Calcein AM (which is not specific for iron), nor Pftn-2::GFP transgenic C. elegans worked in our hands. Further, the fact that the genes corresponding to the E. coli hits were not essential and that these hits were capable of supporting C. elegans development, albeit slowly, suggests that the changes in iron level may not be dramatic. Future studies using inductively coupled plasma-mass spectrometry will likely provide a more direct measurement of iron content in different bacterial strains.

In contrast to previous work (Qi and Han, 2018), the enterobactin biosynthesis mutants did not delay C. elegans development in our hands. This is likely because E. coli has multiple iron import systems that act redundantly (Braun, 2003). In contrast, Fep/fes mutants did delay development. These mutants have high levels of enterobactin (Fuhrer et al., 2017), the biosynthesis of which is upregulated when bacterial iron is low (Escolar et al., 1999; McHugh et al., 2003). Enterobactin is a powerful iron scavenger, and any iron that could be imported through other iron import systems may be locked in with enterobactin and not be available to the bacterial cell.

Remarkably, the C. elegans developmental delay caused by most if not all E. coli hits could be rescued either by supplementing either anti-oxidant or by adding iron. Adding additional iron or anti-oxidant to C. elegans fed wild type E. coli did not affect the animal’s developmental rate, suggesting that, with this bacterial strain, an optimal amount of iron is provided. Our finding that metabolically inactive bacterial powder from fepG and cyoA mutants delay C. elegans development relative to powder made from wild type bacteria suggests that low iron is the primary cause for the developmental delay, likely by interfering with mitochondrial function.

It is likely that low iron not only generates ROS but that high ROS also results in low iron in bacteria via the Fenton reaction where the ROS H₂O₂ can be converted into highly reactive hydroxyl radicals (•OH). In this reaction, ferrous iron (Fe²⁺) is converted into ferric iron (Fe³⁺), lowering iron availability for biological processes. We associated bacterial metabolites found in a large-scale metabolomics screen (Fuhrer et al., 2017) with the C. elegans developmental delay. The increase in enterobactin in fep and fes mutants could be explained by low enterobactin-iron import in the relevant bacteria, which leads to an increase in enterobactin production. Several other metabolites that can be found in our association list (Table S2) can also be associated with ROS production. For example, tetrathionate is a molecule that has been observed to be produced in a ROS‐dependent process during inflammation in mammals (Daeffler et al., 2017) and shikimic acid is produced by a pathway that has been related to ROS production (Gomes et al., 2017).

The sources of ROS production by the majority of E. coli hits remain elusive. Only 52 of 101 genes encoding (predicted) metabolic enzymes and transporters are included in the reconstructed E. coli metabolic network (Orth et al., 2011). The other 49 genes may be associated with ROS production as their enzymatic or transport function is elucidated. For genes annotated as transcription factors, it may be that they control the expression of ROS generating enzymes or iron transporters. Excessive ROS can damage different biomolecules, including lipids, proteins, and DNA. In light of this, it is interesting that we retrieved multiple genes encoding proteins related to lipid and nucleotide metabolism, DNA repair and protein degradation. It is tempting to speculate that these metabolic processes feedback to the delicate balance between ROS and iron, at least in bacteria. Finally, we retrieved more than 50 bacterial genes with unknown function, which can now be associated with ROS production (Table S1). Taken together, an unanticipated outcome of our study is that the C. elegans-bacteria interspecies system can not only be used to gain insight into the animal’s biology, this system can find functions for E. coli genes as well.

STAR METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for reagents may be directed to and will be fulfilled by the corresponding author A.J.M. Walhout (marian.walhout@umassmed.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

C. elegans strains

C. elegans strains were cultured and maintained by standard protocols (Brenner, 1974). N2 (Bristol) was used as the wild type strain. The zcIs13[Phsp-6::GFP]; strain was provided by the Caenorhabditis Genetics Center (CGC). PE254 (feIs4 [Psur-5::luc+::gfp; rol-6 (su1006)]V) was used for the developmental assay. All experiments were performed with synchronized L1 stage hermaphrodite animals that were incubated for 48 hours on nematode growth media (NGM) at 20°C.

Bacterial Strains

E. coli strain OP50 and wild type E. coli BW25113 were grown at 37°C in Lysogeny Broth (LB). E. coli deletion mutants (Baba et al., 2006) were grown at 37°C in LB with 50 μg/mL kanamycin. Comamonas aquatica strain DA1877 was grown at 37°C in LB with 10 μg/mL gentamycin. Overnight cultures were seeded on NGM plates and dried for 1 hour at room temperature.

METHOD DETAILS

E. coli Deletion Collection Screen for Changes in C. elegans Developmental Rate

Bacteria were grown overnight at 37°C in LB medium with 50 μg/mL kanamycin in 96-well deep plates. 20 μL of overnight culture was seeded onto 96-well plate containing NGM agar dried at room temperature prior to the addition of C. elegans. Animals were fed the parental E. coli BW25113 strain (wild type) for at least two generations prior to screening. C. elegans embryos were harvested from gravid animals using buffered bleach solution and incubated overnight in M9 buffer at room temperature to allow embryo hatching and L1 arrest. Approximately 25 synchronized L1 animals were placed in each well of 96-well NGM agar plates containing individual E. coli single-gene deletion strains and allowed to develop for 48 hours at 20°C. Animals were then washed with M9 buffer and transferred to fresh 96-well plates. Bright-field images of animals in each well were collected using an Invitrogen EVOS FL imaging system, and body size was determined by measuring the surface area of the animals using ImageJ (Schindelin et al., 2012). Each E. coli single gene deletion strain was screened five times and mutants that cause an average of at least a 10% animal body size change were retested in triplicate in 24-well plates using approximately 40 L1 animals per well. Bacterial mutants that caused at least a 10% reduction in animal body size in each of the three retest experiments were genotyped to verify the deletion and were considered hits.

Genotyping Bacterial Deletion Strains

Single gene deletion E. coli strains were streaked onto LB plates containing 50 g/mL kanamycin. PCR was performed on individual colonies using genomic and kanamycin-cassette-specific primers. Genomic primers were designed for individual strains at starting 100–300 bases upstream of the start codon of the gene, and 100–300 bases downstream of the stop codon. Primers are listed in Table S5. PCR products were analyzed for the correct size by agarose gel electrophoresis.

Bacterial Growth Rate

Overnight cultures of wild type and hit E. coli strains were diluted 1:1000 in LB and grown for 12 hours in 96-well plates at 37°C on a Tecan Spark plate reader with the OD₆₀₀ (nm) recorded every 10 minutes. To quantify the growth properties of the bacterial strains, Gompertz growth function (Equation 1) (Kahm et al., 2010) was fitted to OD data. In this equation, λ, μ, and A characterize the lag time, growth rate (maximum slope), and maximum growth level reached, respectively, while e designates Euler’s number. For each culture, first, a background OD level based on the average of eight wells from the same plate with no bacteria was subtracted from OD values. Then, all three growth parameters were obtained for each culture by best-fitting Equation 1 to the time series of corrected OD values, using the nlinfit function of Matlab (The MathWorks, Inc., Natick, MA). The final growth properties of each bacterial strain were obtained by averaging the growth parameters from three independent tests, and the relative growth rate was normalized to the growth rate of the wild type strain.

$$OD(t)=A.\text{exp}\left\{-\text{exp}\left[\frac{\mu .e}{A}(\lambda -t)+1\right]\right\}$$ (1)

Luciferase Developmental Rate Assay

Luciferase developmental rate assays were performed as described (Olmedo et al., 2015). Briefly, single arrested L1 animals were transferred into one well of a 96-well plate containing 100 μl of S-basal medium with 200 μM D-Luciferin and simultaneously resumed development by adding 100 μl of S-basal with bacteria. Plates were sealed with a gas-permeable cover (Breathe Easier, Diversified Biotech) and luminescence was measured in a Berthold Centro LB960 XS3 for 1 sec, at 5-min intervals.

Fluorescent Dye-Based ROS Detection

The fluorescent dye 6-carboxy-2’,7’-dichlorodihydrofluorescein diacetate (carboxy-H2DCFDA, Thermo Fisher Scientific) was used to quantitate ROS production in bacteria. A 10 mM stock solution of carboxy-H2DCFDA was made in DMSO. Overnight cultures of E. coli were diluted 1:100 with LB, and bacteria were grown in 200 μL of LB in 96 deep well culture plates at 37°C for another 4 hour s. Cultures were centrifuged at 3,000 rpm for 30 minutes and washed with phosphate buffered saline. Carboxy-H2DCFDA was added to the bacteria to a final concentration of 10 μM and incubated in the dark at 37°C for 1 hour. Fluorescence signals of carboxy-H2 DCFDA were detected using a Tecan microplate reader with maximum excitation at 495 nm and emission at 520 nm.

Metabolite Supplementation

A 1 M stock solution of FeCl₃ (Sigma Aldrich) and 0.5 M solution of N-Acetyl-cysteine (Sigma) were made in ddH₂O. 0.5 M stock solutions of 2,2′-dipyridyl (Sigma Aldrich) and 1,10-phenanthroline (Sigma Aldrich) were made in DMSO. All stock solutions were stored at −20°C. For each experiment, each stock solution was diluted to the final concentration in NGM agar prior to plate pouring.

Phsp-6::GFP Reporter Assay

C. elegans embryos were harvested from gravid Phsp-6::GFP stress reporter animals using buffered bleach solution, and incubated overnight in M9 at room temperature. Approximately 40 synchronized L1 animals were placed on 96-well NGM agar plates containing individual Keio E. coli single-gene deletion strains and allowed to develop for 72 hours or longer for strains that slow C. elegans development. Animals were then washed with M9 and transferred to new 96-well plates. Fluorescent images of animals were collected using an Invitrogen EVOS FL imaging system. Bacterial mutants that caused a change in two out of three experiments were considered hits.

Statistical Analysis with Published Metabolomics Dataset

The metabolomics dataset used in this study was previously obtained with a high throughput chromatography-free method that omits ion separation before mass spectrometry (Fuhrer et al., 2017). While the exclusion of chromatography was necessary to enable high-throughput analysis at genome scale, substantial noise is introduced as a result, which was evident from the high z-scores of metabolites in wild type bacteria. Based on the wild type profile, we set thresholds of 0.4 and 1.5 for absolute z-score values of positive and negative ions, respectively. Subsequently, scores below these thresholds were collapsed to zero while those greater than these thresholds (in magnitude) were reduced by the pertaining threshold. Also, the analysis was limited to strains that were quantitatively studied with respect to the effect on C. elegans body size in this study and to ions that were annotated, yielding a metabolome data matrix of 1513 bacterial gene deletions by 3114 ions. Correlation between columns (ions) of this matrix and relative C. elegans body size data was analyzed (Fig. 2E). Specifically, Pearson correlation coefficients and corresponding p-values were calculated using corrcoef function of Matlab (The MathWorks, Natick, MA). Then, mafdr function of Matlab was used to estimate P-value-based FDR corrected for multiple hypothesis testing, following the procedure of Benjamini and Hochberg (Benjamini and Hochberg, 1995). An FDR threshold of 10% was used to define ions that are significantly associated with C. elegans body size (Table S2).

Metabolic Network Analyses

A modified version of E. coli metabolic network model iJO1366 (Orth et al., 2011) was used for mapping E. coli hit genes to metabolic pathways and for Flux Balance Analysis (FBA). The modifications included the addition of sink/demand reactions to fix network gaps by rescuing dead-end metabolites, and constraints on reactions associated with nine genes that are in iJO1366 reconstruction (based on E. coli K12 MG1655 strain), but absent in Keio parent (E. coli K12 BW25113 strain), which prevented these reactions from carrying flux (see Table S6 for all additions and constraints). Constraint-based FBA was performed using Cobra Toolbox (Vlassis et al., 2014) in MatLab following standard mass balance and reversibility constraints on metabolites and reactions, respectively. In addition, original maintenance energy requirements were preserved (Orth et al., 2011), while no constraints were imposed to force growth (biomass formation) (Table S6). To define the main nutritional input, exchange reactions were constrained according to (Tawornsamretkit et al., 2012) assuming an LB growth medium. However, the lower boundary of all exchange and sink reactions that are not part of these medium constraints was set at −0.1 mmol/gDW/h to allow other nutrients to be used by the model so that all reactions could carry flux during FBA.

To calculate a flux distribution that incorporates all E. coli hit genes a MatLab implementation was developed based on iMAT algorithm (Zur et al., 2010). The original algorithm receives a set of reactions which are to be activated (R_high), and another set of reactions, which are to remain inactive (R_low). Then the number of reactions in R_high that carry a flux greater than a threshold and the number of reactions in R_low that carry no flux are added together and this sum is maximized to find an optimal flux distribution. In this study, all reactions associated with the 52 hit genes defined R_high. No R_low was defined as there is no basis for such definition. After maximizing the number of active reactions in R_high, the total flux was minimized while keeping the maximized number from the first step constant, to obtain a minimal flux distribution where all hit genes were functional. This method combines iMAT (Zur et al., 2010) and parsimonious FBA (Lewis et al., 2010) algorithms as we have done previously (Yilmaz and Walhout, 2016). The default threshold used to impose flux on reactions in R_high was 0.05. However, if the maximum flux that a reaction could carry was less than this amount, then half of this maximum was used as the threshold. Reversible reactions could have different flux thresholds in forward and reverse directions when necessary. For randomization of optimized flux distribution, 52 genes were randomly picked from the model, and R_high was defined based on this random set of genes. Then the abovementioned procedure was repeated 10,000 times. For comparison between different flux distributions, sum of fluxes over defined subsets of reactions or sum of fluxes through selected metabolites were normalized with the total flux. All summations used absolute flux values.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was performed with Prism 7 software (GraphPad). Statistical significance for experimental data was determined using a Student’s t-test and P-values less than 0.05 were taken to indicate statistical significance. For the published metabolomics data, Pearson correlation coefficients and corresponding P-values regarding the association between ion columns in this matrix and relative C. elegans body size were calculated using corrcoef function of Matlab.

DATA AND SOFTWARE AVAILABILITY

This study did not generate any new software. The published article includes all datasets generated during this study.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and Virus Strains
Escherichia coli K-12 Keio Knockout Collection	Dharmacon	Cat#: OEC4988
Escherichia coli OP50	Caenorhabditis Genetics Center (CGC)	N/A
Comamonas aquatic DA1877	CGC	N/A
Chemicals, Peptides, and Recombinant Proteins
D-Luciferin	Biothema	Cat#: BT11–100
N-Acetyl-L-cysteine	Sigma Aldrich	Cat#: A7250
Iron(III) chloride	Sigma Aldrich	Cat#: 157740
2,2’-bipyridyl	Sigma Aldrich	Cat#: D216305
1,10-phenanthroline	Sigma Aldrich	Cat#: 131377
Dimethyl sulfoxide	Sigma Aldrich	Cat#: D4540
Levamisole Hydrochloride	Sigma Aldrich	Cat#: PHR1798
6-carboxy-2’,7’-dichlorodihydrofluorescein diacetate	Thermo Fisher Scientific	Cat#: C2938
Experimental Models: Organisms/Strains
Caenorhabditis elegans N2 (wild type)	CGC	N/A
C. elegans feIs4 [Psur-5::luc+::gfp; rol-6 (su1006)]V	(Olmedo et al.,2015)	Strain PE254
C. elegans zcls13[Phsp-6::GFP]	CGC	Strain SJ4100
Oligonucleotides
List of Oligonucleotides	This study	Table S5
Software and Algorithms
ImageJ/Fiji	(Schindelin et al., 2012)	https://fiji.sc/
MATLAB	Mathworks	https://www.mathworks.com/products/matlab.html

HIGHLIGHTS.

A screen of E. coli mutants reveals bacterial genes essential for C. elegans development

This developmental delay can be rescued by anti-oxidants or iron supplementation

Low bacterial iron raises ROS production likely triggering oxidative stress in C. elegans