By examining orthology relationships, we found that orthologues of S. cerevisiae HI genes are highly retained among the yeast species; the probability of their having an orthologue in another yeast species is significantly higher than the average for the S. cerevisiae genome (Fisher's exact test; P = 10^-3). We found that this is true with respect to both those species which diverged prior to, and those diverging after, the whole-genome duplication (WGD) that occurred approximately 100 million years ago 15 in the Saccharomycetales lineage. Moreover, HI genes are also more likely to have subsequently been retained in duplicate following the WGD (see P values below), indicating that, at least through the approximately 20 million year period of rapid post-WGD gene shedding (between the divergence of the S. cerevisiae and Kluveromyces polysporus lineages 16 ), selective pressure existed toward maintaining a higher dosage of the proteins they encode.

Since ribosomal complexes are acutely sensitive to protein dosage imbalance, and most genes encoding ribosomal proteins (r-proteins) are over-represented among the gene pairs retained following the WGD 17 , we excluded r-protein genes from the analysis and still found the S. cerevisiae genome to be enriched for pairs containing at least one (P < 10^-5), or two (P = 0.007) HI members (see Additional file 1). A similar pattern is observed in other post-WGD yeast species, with orthologues of HI genes being over-represented among surviving WGD paralogous pairs (P = 10^-7, 0.05, 0.01 for Saccharomyces bayanus, K. polysporus and Saccharomyces castellii, respectively). The K. polysporus result is particularly significant, since post-WGD gene loss in this species occurred essentially independently of S. cerevisiae evolution 16 . All of this suggests that the haploinsufficient phenotype predated the WGD.

The retention of orthologues of yeast HI genes appears to have acted over a still greater evolutionary timeframe: we found that yeast HI orthologues are highly retained across the entire Ascomycete phylum, an evolutionary span of at least 900 million years 18 , and that the orthologues of yeast HI genes are retained significantly more frequently than the genomic average in 75% of eukaryotic species with published sequenced genomes (see P values in Tables 1 and 2). We considered the possibility that this over-representation is driven by a subset of the HI genes coding for proteins involved in some important biological process that is, itself, highly conserved among eukaryotes. In particular, the set of HI genes in S. cerevisiae is enriched for genes encoding r-proteins. In order to rule out this possibility, the HI genes were first classified into the broad functional categories represented by the generic S. cerevisiae GO_slim terms provided by the Gene Ontology Consortium 19 . Genes within each of these discrete sets were then sequentially removed and the orthology analysis repeated. In each case, the statistical significances of the orthology comparisons remained at P < 0.05; for those species in which the full HI set is more highly retained, the significant over-representation is preserved and, for those few species not exhibiting HI-orthologue over-representation, deletion of functional categories does not give rise to significance. This suggests that the over-representation of orthologues is indeed due to haploinsufficiency, rather than any particular functional subset.

Similarly, we also sought to rule out the possibility that conservation is driven by the homozygous deletant phenotype, rather than the HI phenotype. Accordingly, we first confirmed that, congruent with our previous findings for chemostat culture 8 , there exists no significant correlation between our turbidostat haploinsufficient set and essentiality/significant fitness defect in the homozygous deletant (P value of an overlap between the phenotypes = 0.7). Secondly, those HI genes for which the null mutant is fitness compromised are no more likely than the remainder of the HI gene set to be retained across the species (the P value across the yeast species for preferential retention = 0.3). Moreover, there is no significant correlation between the null mutant fitness for a given HI gene, and the fraction of species in which its orthologues are retained (correlation coefficient = 0.06).

We also found that haploinsufficiency in yeast is a good predictor for the phenotype of orthologues in higher eukaryotes. Members of a set of 299 published human haploinsufficient genes, assembled via text mining (following the method described by Dang et al. 20 ), are significantly more likely than other human genes to have a yeast orthologue that is HI (P < 10^-4). A total of 83 of these human haploinsufficient genes have a BLAST best-hit orthologue in S. cerevisiae (consistent with the approximately 25% likelihood of a human gene being associated with a yeast orthologue by this method), and 40 of these (unique) orthologues are also HI in yeast.

We therefore defined a set of candidate human HI genes based upon the phenotype of their yeast orthologues (using unique yeast-human orthologue matches), listed in Additional file 2, and examined their gene ontology and disease associations. Of our 563 human gene candidates, 104 are associated in the Online Mendelian Inheritance in Man (OMIM) database 21 with a genetic disease (see Additional file 3). A GO enrichment search of the candidate HI/disease-associated set suggests an over-representation of gene functions essential in tumourigenesis and metastasis. DNA damage response and repair, apoptosis, and the oxidative stress response, along with cell migration and cell-cell adhesion terms are all enriched with a P value of less than 0.0005 (corrected for multiple testing).

Our predicted HI/OMIM-associated set includes the known HI tumour suppressor NF1, along with a cluster of four genes, RPS19, RPL1, RPS7 and RPS24, associated with Diamond-Blackfan anaemia (the latter three of which are novel HI predictions), and three connected with Williams syndrome: RFC2, STX1A and PSPH (the latter two being novel HI predictions). Since both of these conditions have independently been associated with haploinsufficiency 5 22 , their presence among our predicted set provides encouraging support for our yeast-based method. Further clusters include four genes related to colorectal cancer (a GO enrichment P value of 0.0006). None of these four genes has previously been annotated as HI according to our text-mining search. In addition, we make novel HI predictions for several genes associated with Xeroderma pigmentosum and, by extension, defective nucleotide excision repair. Genes involved in the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways 23 of Huntingdon's disease, Parkinson's disease and Alzheimer's disease are also over-represented (P < 0.02 for all; see Additional file 3).

Aside from disease, in heterogametic eukaryotes, haploinsufficiency is obviously relevant with respect to the sex chromosomes since the dosage of the sex-linked genes can vary by a factor of two between the sexes. We previously noted 8 that chromosome III of S. cerevisiae, which carries the mating locus (MAT), is enriched for genes that are HI under different nutrient-limited conditions (P < 10^-50). We argue that this accumulation was selectively advantageous upon its occurrence in an ancestral yeast, and has been subsequently retained across the lineage.

As to the selective advantage, we have previously proposed that the accumulation of HI genes on this chromosome is a mechanism that leads to cells that have lost one copy of chromosome III being eliminated from the population. Generally, mating ability is repressed in diploid cells due to the presence of both mating-type alleles, but a diploid bearing only one copy of chromosome III can express the haploid mating programme, becoming a so-called 'diploid mater'. Such aneuploids would be unable to sporulate and would give rise to (fitness-compromised) triploid progeny on mating with a haploid partner (Figure 1a); thus, chromosome III monosomy is a serious detriment to the population.

Here, we have found that the over-representation of HI genes on S. cerevisiae chromosome III also holds when the phenotype was defined in turbidostat culture (P = 10^-5). Thus the effect is not nutrient specific and holds true in yeast growing in near-optimal conditions. To test the loss-of-chromosome III hypothesis, we examined whether the observed MAT linkage also holds in other fungi having the same mating-type determination system. Indeed, orthologues of HI genes from S. cerevisiae chromosome III are found to be over-represented on the chromosomes bearing the mating-type locus in species throughout the Ascomycota (with P values ranging from 10^-2 to < 10^-19). By constructing the relevant heterozygous deletion mutants, we have confirmed, in a subset of these MAT-linked HI orthologues, that the haploinsufficient phenotype does indeed carry over to two other yeast species. Mutants for 4 of 5 Saccharomyces kluyveri genes, and 9 of 10 deletions in a Kluyveromyces lactis diploid, grow significantly slower than the wild type, corresponding to P values of 7 × 10^-3 and 4 × 10^-6, respectively (see Table 3). It should be noted the batch growth assays performed for the non-cerevisiae yeasts are significantly less sensitive than continuous culture competition experiments 8 , and hence small (but significant) differences in growth rate will not be detected. That a growth defect (much less a significantly lower growth rate) is observed at all in batch culture is indicative of a strongly haploinsufficient phenotype.

The antiquity of the relationship between HI genes and the MAT chromosome is further suggested by the fact that the association holds not only in diploid species, but also in predominantly haploid yeasts such as K. lactis and Ashbya (Eremothecium) gossypii (Table 4). Since the loss of any chromosome from the haploid genome is lethal, additional selective mechanisms against the loss of the MAT chromosome should be unnecessary. Thus, HI genes should not be significantly clustered near the MAT locus in such species unless the arrangement is inherited from a common ancestor of the diploid/haploid yeasts.

The only species not exhibiting linkage of HI genes to MAT are the closely related pathogens Candida albicans and Candida dubliniensis (see Table 4). These species possess only a parasexual cycle and are believed to be incapable of undergoing meiosis 24 , in contrast to the other Ascomycetes studied (including Candida glabrata, for which a putative sexual cycle has been observed; 25 ).

In the higher eukaryotes that have a heterogametic sex-determination system, the situation is the reverse of that in the Fungi. In the XY, X0, and ZW systems, all chromosomes are maintained in duplicate in both sexes, except for the sex chromosomes, for which there is only a single copy in one of the sexes. The haploid status of the sex chromosome means that the presence of HI genes on these chromosomes would detrimental to the fitness of the heterogametic sex. Indeed, we have found that the orthologues of yeast HI genes are significantly under-represented on the X chromosome of the human, mouse, rat, chimp, dog, cow and horse (XY species) and the X chromosome of the nematode worm (X0) (see Table 4). In general, the Y chromosome is too small, and carries too few yeast orthologues, to perform significance tests.

Genes on the X chromosome are also under-represented among the set of curated human haploinsufficient genes (6 of the 299 are on the X chromosome, a P value for under-representation of 0.005). Furthermore, 2 of the 10 HI orthologues (that is, genes in our predicted human haploinsufficient set) found on the human (and chimp) X chromosome (namely NLGN4X and DDX3X) are among the 12 human genes that retain a functional homologue on the Y chromosome (P = 0.003, hypergeometric test). This perhaps indicates selection for the maintenance of their two-copy status. The remaining 10 human X chromosome genes possessing a Y chromosome gametologue do not possess yeast orthologues.

Orthology assignments were made using the InParanoid algorithm 36 , and compared with the results of a BLAST 37 reciprocal best-hits search. GO_slim annotations for S. cerevisiae genes were retrieved from the Gene Ontology Consortium 20 . GO enrichment searches were performed using the Babelomics 4 FatiGO tool 38 . Associations of the human candidate set with inherited diseases were obtained from the Online Mendelian Inheritance in Man (OMIM) database 19 .

Paralogous WGD pairs for the Saccharomycetales were obtained from Byrne and Wolfe 39 . S. cerevisiae over-expression phenotypes were from Sopko et al. 40 and homozygous deletion phenotypes from Winzeler et al. 41 . The sequence of the human X chromosome and the pseudoautosomal regions were obtained from Ross et al. 42 and Graves et al. 43 , respectively.

To assess the significance of HI gene conservation, the number of HI genes having orthologues in a given species, given the number of S. cerevisiae HI genes, was compared against the whole-genome conserved proportion using a χ² or Fisher exact test (depending on sample size), with the null hypothesis of identical distribution. To assess the significance of HI gene numbers on a given chromosome, we calculated the hypergeometric probability, given the number of genes on the chromosome, and the number of HI genes in the genome.

A similar approach was taken in order to determine the significance of the distribution of HI orthologues across the chromosomes. The number of HI orthologues on the particular chromosome, given the total number of orthologues on the chromosome, was compared against the total number of HI orthologues in the species. Fisher's exact test was used to determine significance, using the null hypothesis that the distribution of HI orthologues would not differ from that of orthologues in general. We applied a similar methodology to confirm that the distribution of HI orthologues did indeed differ from the set of non-HI orthologues (data not shown). All findings of significance were reiterated using a z test for difference of proportions. Where necessary, P values were corrected for multiple testing.

Candidate K. lactis and S. kluyveri genes were those on their chromosomes bearing the mating-type locus, which have a reciprocal best-hit orthologue on S. cerevisiae chromosome III which showed a haploinsufficient phenotype in selections in turbidostat culture. Of these, 10 K. lactis, and 15 S. kluyveri genes (see Table 3) were selected for targeting.

Strains were cultured in YPD rich medium (1% yeast extract, 2% peptone, 2% glucose), and Escherichia coli strains in LB medium (1% tryptone, 0.5% yeast extract, 1% NaCl). The pFA6a-KanMX and pAG25-NatMX plasmids were harvested from, respectively, E. coli transformants pFA6 and pAG25 by miniprep and digested with NotI. The resulting KanMX/NatMX gene was purified using a QIAquick kit (Qiagen, West Sussex UK).

Though amenable to transformation using techniques and plasmids designed for S. cerevisiae, the efficiency of homologous recombination in both K. lactis and S. kluyveri is much lower than in S. cerevisiae. Efficient gene targeting therefore requires at least 500 bases of homology between the regions flanking the target gene, and the replacement cassette. Such cassettes, containing either the KanMX or NatMX 44 gene (conferring resistance to geneticin 418 or nourseothricin, respectively), were constructed using the two-step polymerase chain reaction (PCR) method described by Wach et al. 45 with modifications. For the first step, two standard PCR reactions were set up to amplify the 500-nucleotide (nt) regions immediately upstream (or downstream) of the target gene, with genomic K. lactis CK11 or S. kluyveri CBS 3082 DNA as a template. Oligonucleotides were designed so as to concatenate an additional 26 bases homologous to the 5' (or 3') end of the marker gene onto the upstream (or downstream) amplified region.

These approximately 550-base upstream and downstream fragments were then used as primers in a second PCR reaction. The Roche Expand Long Template PCR system (Roche Diagnostics, West Sussex UK) containing a mixture of Taq and Tgo DNA polymerases was used, a typical reaction containing 0.5 μl (approximately 50 ng) of each of the first PCR products, 0.3-0.5 μg of the NotI-digested plasmid. PCR was carried out in a reaction volume of 50 μl, containing 200 μM of each dNTP, 1.75 mM MgCl₂, 50 mM Tris/HCl pH 9.2, 16 mM (NH₄)SO₄, and 0.75 μl of enzyme mix. A Bio-Rad Thermo-Cycler 1000 (Bio-Rad laboratories, Hertfordshire UK) was programmed to initially denature the samples at 92°C for 120 s, then perform 10 cycles of 10 s at 92°C, 30 s at 50°C, 120 s at 68°C, followed by 25 cycles of 15 s at 92°C, 30 s at 50°C, and 120 s at 68°C, with an additional 20 s elongation for each successive cycle 46 .

The resulting PCR product was used to transform K. lactis CK11 or S. kluyveri CBS 3082 following the method of Hill et al. 47 . Cells were grown in rich medium to a density of approximately 10⁷ cells/l, harvested by centrifugation, washed in 10 ml 100 μM lithium acetate, and finally resuspended in 1 ml of 100 μM lithium acetate. A total of 100 μl of cell suspension was added per 10 μl of PCR product; 240 μl of 50% (w/v) poly(ethylene glycol)-3000; 30 μl 1 M lithium acetate; and 25 μl 2 mg/ml single-stranded (ss)-DNA. The suspension was incubated at 30°C with shaking for 30 min; 43 μl of dimethylsulfoxide added; and subsequently heat shocked at 42°C for 10 min. Cells were incubated in YPD for 3 h at 30°C, and G418 and nourseothricin-resistant transformants selected on YPD containing 0.2 mg/ml G418 and 0.1 mg/ml clonNAT (Werner Bioagents, Jena Germany), respectively. Correct marker integration was confirmed by colony PCR amplification of both the marker gene, and of the remaining copy of the target locus.

Stationary-phase cultures were diluted 1:100 in YPD, and 200 μl transferred into 1 well of a 96-well microtitre plate. A BMG Fluostar Optima microplate reader (BMG Labtech, Aylesbury UK) was used to take 600 nm absorbance readings at 10-min intervals over a period of 24-35 h. A curve-fitting algorithm, implemented in R, calculated the maximum growth rate from the measured growth curve, and the median was taken across all replicates of the strain in the plate. This value was compared with the median for each of the other strains in the same microplate, with significant growth-rate variation identified using a two-tailed t test at 95% significance. Measurements were repeated at least five times for each strain. The standard deviation of the repeated measurements was calculated, and the maximum among all strains (9%) was taken as the uncertainty in each strain.

The P value for the overlap between the sets of haploinsufficient genes in S. kluyveri or K. lactis and S. cerevisiae was calculated by assuming that the frequency of haploinsufficiency within the genome of these two species was equal to that in S. cerevisiae (approximately 20% in YPD). A P value can be ascribed to each orthologue set individually. With the null hypothesis that haploinsufficiency in S. cerevisiae and S. kluyveri/K. lactis is independent, the probability that both members of a given orthologue pair are HI is P = 0.2, and if the gene is HI in all three species, P = 0.2² = 0.04. The probability of observing n orthologue pairs in which both genes are HI, out of N pairs tested, is then given by the binomial probability of n successes, and (N-n) failures, with probability of success = 0.2. Thus the P value for 9/10 K. lactis/S. cerevisiae pairs having both orthologues HI is 4 × 10^-6, and for 4/5 S. kluyveri/S. cerevisiae pairs both being HI is 7 × 10^-3.