The experimental system involved co-culture of >6,000 heterozygous (hemizygous) deletion strains in carbon-limited continuous culture. A number of similar studies have co-cultured the heterozygotes in batch culture 1521, whereas a more recent study with the heterozygotes has used the same continuous culture system as that employed here 22. The use of continuous culture for competitions enables detection of the more subtle phenotypes, expressed as small growth rate differences and revealed over a large number of generations. In addition, use of the chemostat for continuous culture offers high reproducibility, owing to a defined and constant growth rate and physicochemical environment at steady state. CrO₃ was supplied at a sub-lethal dose (0.1 mM), pre-determined to cause an approximately 30% increase in the mean doubling time of the mixed cultures. The relative growth of each strain in the cultures (that is, change in relative abundance between the start and end of a chemostat experiment) was derived from signals assigned to the strains' unique identifying (barcode) sequences (see Background, and Materials and methods). The effect of Cr was determined by comparing the relative growth of each strain in the Cr-treated cultures versus that in control cultures. This yielded a value for each strain for the size of the growth effect of Cr (see the Data analysis section in Materials and methods). The data for each strain are given in Additional data file 1.

The range of growth effects of Cr across the strains indicated a normal distribution centered around zero (Figure 1). The relative growth of some strains was decreased by Cr (negative growth effect; tendency towards haploinsufficiency with Cr), whereas others showed improved relative growth (tendency towards haploproficiency). A similar normal distribution was evident for growth effects on strains that were heterozygous specifically for essential gene functions. The growth data were analyzed further to identify strains that showed significant (false discovery rate, q < 0.05) haploinsufficiency or haploproficiency (see the Data analysis section in Materials and methods). There were fewer significantly haploinsufficient strains than haploproficient ones, that is, 115 strains exhibited a Cr-specific growth defect (indicating gene functions that normally protect against Cr), whereas the relative competitiveness of 203 strains was enhanced by Cr (indicating functions through which metal toxicity could be mediated). This suggests that S. cerevisiae has not been routinely exposed to Cr stress during its evolutionary history, as Fisher 23 demonstrated that when selection occurs in the environment to which an organism is adapted, then most mutations will be deleterious; whereas, when selection occurs in increasingly suboptimal conditions, then an increasing proportion of mutations will be beneficial. Knowledge of genes that exhibit haploproficient phenotypes in the presence of Cr could be exploited to increase the rate of biotechnological processes that may be limited by metal toxicity 47.

In contrast to the therapeutic compounds that were the subjects of previous haploinsufficiency analyses 14152124, there is no a priori expectation that the primary target of Cr will be a specific essential protein 5. Nonetheless, closer analysis of our data showed that, under the condition of Cr stress, five of the eight most significant haploinsufficient phenotypes (that is, those with the lowest q-values) were found in strains heterozygous for an essential gene (Additional data file 1). These included NHP2 (involved in 18S rRNA processing), and ARP3 and ARC19 (involved in actin nucleation and actin patch function). As with the drugs, these observations could, in principle, be explained by direct interference of the metal with the essential function of its target protein, reducing its activity to a level below that required to sustain the growth of a diploid cell at wild-type rates. Alternatively (and this also applies to the haploinsufficient phenotypes observed for non-essential genes), there may be a synthetic lethal interaction with the principal target of chromium, or the haploinsufficient protein may contribute to the intrinsic resistance of the cell to the toxic action of the metal.

GoMiner 25 was used to associate Gene Ontology (GO) terms with all genes whose heterozygous mutants exhibited significant haploinsufficiency or haploproficiency under Cr stress (Additional data file 2). Not unexpectedly, GO terms related to transport and metal homeostasis were significantly over-represented in the annotation of those genes that displayed either haploinsufficient or haploproficient phenotypes. Gene functions involved with chromatin structure were also significantly over-represented in the haploinsufficient and haploproficient data, indicating an involvement of chromatin organization and its possible effects on gene expression in Cr resistance. Schnekenburger et al. 26 have described how Cr cross-links complexes of histone deacetylase 1 and DNA methyltransferase 1 to gene promoters, inhibiting histone modifications and decreasing recruitment of RNA polymerase. Cr may also provoke aberrant DNA methylation, with the potential to silence tumor suppressor genes in higher cells 27. Genes involved in nucleotide excision repair also were evidently important for Cr resistance, consistent with previous work 28. Cr is well known to promote DNA damage, but it is unresolved whether this is a primary cause (versus a secondary effect) of Cr toxicity.

The Cr treatment revealed haploproficient phenotypes for several genes involved in sulfur metabolism. The flux of sulfur in these heterozygotes could be re-directed towards molecules that may promote metal resistance, such as glutathione (GSH). Such re-programming of sulfur metabolism occurs normally in wild-type yeast responding to other metals 2930. In addition, Cr uptake may occur through sulfate transporters, which are regulated in response to Cr stress 31. Actin was a highly over-represented haploinsufficient category for Cr, which might relate to targeting of actin function by Cr, as suggested above.

Genes involved in proteasome function and regulation of protein stability were among those most significantly over-represented in the set showing haploinsufficient phenotypes in the presence of Cr (Additional data file 2). We decided to subject this evidence of the mechanism of chromium's toxicity to further investigation. Initially, we validated the output from the library screen by confirming in independent batch-culture assays the haploinsufficient phenotypes of several Cr-treated proteasome mutants (Additional data file 3). These data point to a requirement for protein degradation in Cr resistance and, therefore, to an involvement of cellular proteins in Cr toxicity. This hypothesis was supported by experiments involving cycloheximide, an inhibitor of translational elongation. Exposure of cells to Cr for 3 h resulted in a marked loss of viability (Figure 2a). However, this toxicity was suppressed in cells that were blocked for protein synthesis using cycloheximide.

Previous evidence showed that Cr toxicity involves protein oxidation 32. Oxidized proteins are prone to forming potentially toxic aggregates 33, but this can be countered by proteasomal degradation of the abberant proteins. Combining those observations with our new haploinsufficiency data, we hypothesized that Cr toxicity could involve the formation of protein aggregates. Protein aggregation is also linked to cancer 33, and the carcinogenicity of Cr is well-documented 8. To test the effect of Cr on protein aggregation, insoluble aggregate fractions of proteins were isolated from cells (see Materials and methods) that had been incubated with or without Cr, and the levels of protein in these fractions were determined. The proportion of cellular protein occurring as insoluble aggregates was found to increase approximately two-fold during Cr exposure (Figure 2b), indicating that Cr promotes protein aggregation. Protein aggregates can be toxic, and the potential toxicities of aggregate preparations from cells can be tested by exposing fresh cells to these and measuring their inhibitory effect 34 on the growth of S. cerevisiae. Growth was not affected by supplementing the medium with soluble protein that was previously isolated from Cr-treated cells (Figure 2c). In contrast, growth was slowed in medium supplemented with an equivalent amount of aggregated protein from the Cr-treated cells. Therefore, aggregated protein formed in the presence of Cr can exert a toxic effect.

The observation that Cr resistance was enhanced by the simultaneous inhibition of protein synthesis (Figure 2a) suggested that proteins synthesized during Cr exposure were involved in toxicity. To explore this further, the source of Cr-induced protein aggregates was determined with pulse-chase experiments involving protein labeling with [³⁵S]methionine. These experiments showed that Cr-dependent aggregation was attributable primarily to proteins synthesized during Cr exposure, rather than to aggregation of pre-existing proteins: the aggregate fraction isolated from cells that were [³⁵S]methionine-labeled during the period of Cr exposure was enriched with labeled protein (Figure 3a), whereas the opposite was true for cells labeled prior to Cr exposure (Figure 3b). (In the latter case, there was a decrease in the proportion of labeled protein in the aggregate fraction following incubation with Cr. This could be due to dilution of pre-existing labeled aggregates with unlabelled aggregates formed during the incubation with Cr.) In other experiments, co-treatment with cycloheximide suppressed the Cr-dependent accumulation of protein aggregates (Figure 3c), supporting the conclusion that proteins that form insoluble aggregates in response to Cr are predominantly synthesized during Cr exposure.

The finding that Cr causes aggregation primarily among proteins being synthesized during exposure suggested that the metal might be targeting the protein synthesis or folding machineries. Mistranslation of mRNA transcripts provides a major potential source of aberrant proteins that form aggregates 33. To test whether Cr provokes mRNA mistranslation, the rate of translational read-through of a UAA nonsense (stop) codon was monitored in a short-term dual-luciferase assay (see Materials and methods). The rate of read-through was increased more than two-fold by the addition of CrO₃, at a concentration that increased the population doubling time by about 15% (Figure 4a). A similarly inhibitory dose of another metal, Cu(NO₃)₂, did not significantly affect read-through across the stop codon. The ribosome-targeting drug paromomycin caused a stimulation of mistranslation comparable to that observed with Cr. These data were supported by results from a longer-term qualitative assay, based on read-through of the ade1-14 UGA codon and suppression of the red pigmentation associated with this allele. Treatments with agents such as H₂O₂ or Cu(NO₃)₂ gave no change in colony color compared with untreated controls, whereas red pigmentation was suppressed with paromomycin or CrO₃ (Figure 4b), indicative of mistranslation 35. Red pigmentation was restored when pale colonies from Cr-supplemented medium were sub-cultured onto non-supplemented medium (not shown), indicating that Cr-dependent nonsense suppression did not stem from a prion switch or other heritable change. Translational read-through due to Cr, but not paromomycin, was abolished under anaerobic conditions; this indicates an oxidative basis for Cr-induced mistranslation (Figure 4c).

The hypothesis that induction of mRNA mistranslation causes Cr toxicity was tested first by assaying for synergistic toxicity between Cr and paromomycin. These agents together caused a far stronger growth-inhibitory effect than their combined individual effects (Figure 5a), indicating that paromomycin (which provokes mistranslation via ribosome binding) and Cr target a common process. No synergy was found between H₂O₂ and paromomycin (data not shown). Second, we examined Cr resistance in 18S ribosomal RNA mutants that carry out mRNA translation with differing degrees of accuracy 36. The L1583 mutant, which is characterized by highly error-prone translation, was markedly sensitized to Cr in comparison to the wild type (L1494); in contrast, increased translational accuracy (strain L1597) caused increased Cr resistance (Figure 5b). These results substantiated the proposal that induction of mRNA mistranslation is the main cause of chromium's toxic effect on yeast cells.

The heterozygous deletion strains, in the diploid BY4743 background (MATa/MATαhis3Δ 1/his3Δ 1 leu2Δ 0/leu2Δ 0 met15Δ 0/MET15 LYS2/lys2Δ 0 ura3Δ 0/ura3Δ 0) were obtained from the Saccharomyces deletion consortium 42. The strains were pooled as described elsewhere 22. BY4743 was used for aggregate extraction, protein labeling and the luciferase assay. The ribosomal mutant strains (L1494, L1597 and L1583) were kindly provided by Dr Susan Liebman (University of Illinois at Chicago).

Competition experiments in chemostat culture were carried out according to Colson et al. 43 using a small-scale multiple fermenter system (Fedbatch-pro, Das Gip Technology, Julic, Germany). Inoculation with the heterozygote pool and culture in carbon-limited medium were as described by Delneri et al. 22, but with the inclusion of CrO₃ (0.1 mM) or no stressor. In brief, an aliquot (1 × 10⁷ cells) of the pool of heterozygous strains was inoculated into 120 ml of carbon-limiting medium 44. These were grown in batch for 24 h at 30°C with shaking at 170 rev min^-1, before continuous culture was initiated at a dilution rate of 0.1 h^-1 and a constant pH of 4.5. Each competition experiment was conducted in two biological replicates for at least 24 generations. Other experiments were with strains cultured individually in YEPD or YNB media 4546. Where specified, organisms were cultured in 300 μl volumes in 48-well plates (Greiner Bio-One, Stonehouse, Gloucestershire, UK) with shaking at 30°C in a BioTek Powerwave microplate reader (BioTek, Vinooski, VT, USA). Where specified, an anaerobic atmosphere (H₂ + CO₂) was generated with an Oxoid Gas Generating Kit (Oxoid, Basingstoke, Hampshire, UK).

Samples (15 ml) of the organisms from competition experiments were collected from the culture outflow as soon as the continuous cultures reached steady state (time zero sample; approximately 72 h after original inoculation) and also after at least 24 generations of steady-state growth. Genomic DNA was extracted from these using the DNA tissue kit (Qiagen, Crawley, West Sussex, UK). The concentration of DNA in the extract was determined using a Nanodrop device (Agilent Technologies, South Queensferry, West Lothian, UK). The universal primers used for amplification of the unique barcodes in the genomic DNA of the heterozygotes and the hybridization protocol are those used by Winzeler et al. 10. Amplifications and hybridizations for each genomic DNA sample were carried out in duplicate, and each sample was from one of two biological replicates of the relevant competition.

Data from hybridizations were globally normalized by median centering the intensity values from tags corresponding to each heterozygous deletant (two tags per mutant). Log.-ratios for each strain were then calculated between the initial and final chemostat time points 22. This served to eliminate tag-specific biases and further normalized the data. The log.-ratios were expressed as change (in relative strain abundance) per number of cell generations; the latter correction accounted for differences in the generations elapsed between control and Cr-treated cultures (31 and 24 generations respectively). For each strain, the differences in the mean log.-ratios between the control incubations and incubations with Cr indicated the size of the growth effect of Cr. These growth effects were assessed for significance using the p value obtained from an independent-samples t-test. To account for multiple testing, false discovery rates (q-values) were estimated 47, using the Qvalue v1.0 library implemented in the statistical package R, version 2.4.1. Differences yielding a q-value < 0.05 were considered as statistically significant and the corresponding open reading frames selected for further analysis with GoMiner.

Intensity values from tags that do not correspond to deletion mutants were taken as being representative of the background intensity. This was used as a baseline, enabling determination of the presence or absence of individual deletion strains in the experiments. Several strains giving a median signal that was not significantly different to this background were strains that characteristically yield poor hybridization signals 48 (termed 'PH'; Additional data file 1). Data for these were removed from subsequent analyses. Some other strains were lost (out-competed) during competitive culture. Such strains that were not detected in the Cr condition but were in the control condition (termed 'absent'; Additional data file 1) were considered haploinsufficient. Strains that were not detected in the control condition but were in the Cr condition (termed 'present') were considered haploproficient. These strains were included with the relevant haploinsufficient or haploproficient datasets, although the absence of a hybridization signal under either the stressed or the control condition precluded assignment of a q-value.

Protein extraction (total, and the aggregated fraction) was as described in Rand and Grant 49, with the modification of an additional final wash in lysis buffer (minus Igepal), prior to protein quantification with the Bradford assay (Bio-Rad Laboratories, Hemel Hempstead, Hertfordshire, UK). The technique for isolation of aggregates involves solubilization and separation of membrane proteins, so reducing the background of insoluble proteins in aggregate fractions. The term 'aggregated protein', as used in this paper, refers to those fractions that include residual insoluble protein, separated from total protein 49. For radio-labeling, exponential-phase cells (OD₆₀₀ ~0.5) in 25 ml YNB medium were incubated with 1 μl (10 μCi) [³⁵S]methionine (MP Biomedicals, Cambridge, Cambridgeshire, UK) ± CrO₃ for 1 h at 30°C with shaking. Cells were washed twice in chase medium (YNB plus 1 mg ml^-1 unlabeled methionine) and protein was extracted and quantified, as above, either immediately or after 1 h incubation in chase medium + CrO₃. Incorporated isotope was quantified in 5 ml scintillation fluid (Emulsifier Safe, Perkin Elmer, Beaconsfield, Buckinghamshire, UK) using a Packard Tri-Carb 2100TR liquid scintillation analyzer. Incorporation of [³⁵S]methionine was expressed as counts-per-minute (cpm) per μg protein.

Cultures (5 ml) of cells transformed with the dual-luciferase plasmid 50 (a kind gift from Dr David Bedwell, University of Alabama), were grown to OD₆₀₀ ~0.5 and treated with Cu(NO₃)₂, CrO₃ or paromomycin as specified. At intervals, 5 ml of cells were pelleted by centrifugation and resuspended in 60 μl of Passive Lysis Buffer (Promega, Southampton, Hampshire, UK) before vortexing with 40 μl glass beads (0.5 mm diameter, Biospec Products, Bartlesville, OK, USA)) for 10 × 30 s, with a 30 s incubation on ice between each disruption. The subsequent assay was with the Dual Luciferase Assay system (Promega). Extracts were centrifuged at 15,000 g, 30 s and 5 μl of supernatant added to 20 μl Luciferase Assay Reagent II. Samples were read in a Berthold Lumat LB9507 luminometer for 10 s, 20 μl of Stop and Glo reagent was added and the luminescence was again read for 10 s. Background measurements obtained for cells that lacked the plasmid were subtracted from test measurements. The derived ratio of luminescence attributable to the firefly versus Renilla luciferases indicated the level of UAA mis-translation.