Balázs Papp, Csaba Pál & Laurence D

Metabolic network analysis of the causes and evolution of enzyme dispensability in yeast

Balázs Papp, Csaba Pál & Laurence D. Hurst

Supplementary Information

Contents

· Table S1a. Comparison of experimentally estimated flux values with predicted ones under four growth conditions. Page 3-4

· Table S1b. The model predicts presence or absence of fluxes with 91-95% success in central carbon metabolism. Page 4

· Table S2a. In silico flux values of selected metabolic reactions. Page 5

· Table S2b. Average in silico flux values of major metabolic units. Page 6

· Table S3. Comparison of predicted and measured gene dispensability on rich medium with glucose (YPD). Page 7

· Table S4. Causes of metabolic gene dispensability. Page 8

· Table S5. What explains failure of compensation by a duplicate? Page 9

· Figure S1. The extent of flux reorganisation correlates with predicted fitness effect of the knock-out. Page 10

· Table S6. No association between importance of reactions and presence of isoenzymes. Page 11

· Table S7. Reactions catalysed by isoenzymes have larger fluxes than those catalysed by solo copy enzymes. Page 12

· Figure S2. Distribution of environmental specificity of fitness defects of Escherichia coli mutant strains. Page 13

· Figure S3. Phylogenetic distribution of E. coli genes correlates with environmental specificity of mutant phenotypes. Page 14

· Table S8. Calculation of phylogenetic distribution of E. coli genes. Page 15-16

· Equation S1. Biomass equation used for metabolic simulations. Page 16

· Figure S4. Distribution of predicted growth rates for knock-out strains. Page 17

· References Page 18

Table S1a. Comparison of experimentally estimated flux values with predicted ones under four growth conditions.

Daran-Lapujade et al.1 recently measured in vivo flux levels for enzymatic reactions of central carbon metabolism (maltose utilization, glycolysis, pyruvate branch-point, TCA cycle, glyoxylate cycle, gluconeogenesis, and pentose-phosphate pathway) in chemostat cultures of S. cerevisiae grown in aerobic, carbon-limited defined mineral medium. Four different carbon sources were used in this study: glucose, maltose, ethanol and acetate, respectively. We set up the FBA model to mimic these growth conditions and calculated Spearman rank correlation coefficients between measured and predicted fluxes. For most enzymes the model's prediction fits well in vivo measurements, however fluxes carried by the cytosolic and mitochondrial malate dehydrogenases (MDH2 and MDH1) seem to be inappropriately predicted. Excluding these two genes from the analysis leads to higher correlation coefficients.

Carbon source used in aerobic, minimal medium / Spearman rank correlation between measured and predicted fluxes (N=44, P<10-6) / MDH1 and MDH2 excluded from analysis. Spearman rank correlation between measured and predicted fluxes (N=42, P<10-15)
glucose / 0.69 / 0.91
maltose / 0.92 / 0.91
ethanol / 0.79 / 0.94
acetate / 0.92 / 0.97

After excluding enzymes with zero fluxes:

Carbon source used in aerobic, minimal medium / Spearman rank correlation between measured and predicted fluxes (P<10-3) / MDH1 and MDH2 excluded from analysis. Spearman rank correlation between measured and predicted fluxes (P<10-10)
glucose / 0.61
(N=33) / 0.88
(N=31)
maltose / 0.89
(N=34) / 0.89
(N=32)
ethanol / 0.74
(N=31) / 0.96
(N=29)
acetate / 0.94
(N=30) / 0.98
(N=28)

Table S1b. The model predicts presence or absence of fluxes with 91-95% success in central carbon metabolism.

We used in vivo flux measurements by Daran-Lapujade et al.1 to assess the model’s ability to correctly predict reactions with zero and non-zero fluxes (as above, flux levels were measured for reactions of central carbon metabolism under four different carbon sources). The table gives the number of true and false predictions of the model under all four conditions.

Carbon source used in aerobic, minimal medium / Correctly predicted non-zero flux / Wrongly predicted non-zero flux / Correctly predicted zero flux / Wrongly
predicted zero flux / Overall prediction
glucose / 33 / 3 / 7 / 1 / 90.9%
maltose / 34 / 3 / 6 / 1 / 90.9%
ethanol / 31 / 1 / 11 / 1 / 95.45%
acetate / 30 / 1 / 12 / 1 / 95.45%

Table S2a. In silico flux values of selected metabolic reactions.

The table lists predicted flux levels of some selected metabolic reactions under 9 simulated growth conditions. Fluxes were normalised by the biomass production rate. As expected, glycolysis is highly active when glucose is the sole carbon source, while gluconeogenesis becomes active under ethanol and acetate growth condition (see also ref 1). The model also correctly predicts that fermentation dominates during growth on glucose with limited or no O2 uptake. For comparison of model properties with experimental observations in aerobic and anaerobic cultures see also Famili et al.2. Although, S. cerevisiae ferments sugars under aerobic conditions at high growth rates (Crabtree effect3), this behaviour is not predicted by the metabolic model without incorporating further regulatory rules2. As no such regulatory rules are included in our model, its validity under fully aerobic conditions is confined to moderate growth rates (<0.3 h-1).

CO2 production

/ 59.44 / 67.39 / 71.65 / 11.4 / 76.53 / 26.47 / 154.45 / 73.88 / 27.59

* Minus sign indicates ethanol uptake during growth on medium with ethanol as carbon source.

Table S2b. Average in silico flux values of major metabolic units.

Average of absolute flux values were calculated for 10 major metabolic units / pathways under 9 simulated growth conditions. For comparisons across environmental conditions, flux values were normalised by the biomass production rate.

Simulated growth medium / rich with glucose (YPD), low O2 / minimal with glucose, low O2 / minimal with glucose, anaerob / minimal with ethanol, low O2 / minimal with acetate, low O2 / minimal with glucose, limited by C / minimal with glucose, limited by N / minimal with glucose, limited by P / minimal with glucose, limited by S
Metabolic unit
¯
Glycolysis / gluconeogenesis / 21.98 / 25.97 / 33.1 / 1.72 / 1.73 / 5.47 / 19.69 / 26.69 / 6.27
Citrate cycle / 0.5 / 0.51 / 1.44 / 4.13 / 24.7 / 5.79 / 37.42 / 6.41 / 6.07
Energy metabolism * / 2.13 / 2.55 / 1.13 / 36.03 / 49.97 / 15.39 / 27.19 / 2.57 / 16.07
Anaplerotic reactions / 0 / 0.45 / 1.54 / 5.33 / 5.53 / 0.34 / 0.34 / 0.32 / 0.34
Pentose phosphate cycle / 0.19 / 0.96 / 0.7 / 0.21 / 0.21 / 1.1 / 7.62 / 9.92 / 0.21
Pyruvate metabolism ** / 12.67 / 14.16 / 17.05 / 7.67 / 6.31 / 0.19 / 0.3 / 8.42 / 0.15
Lipid metabolism / 0.01 / 0.01 / 0.01 / 0.01 / 0.01 / 0.01 / 0.01 / 0.77 / 0.01
Nucleotide metabolism / 0.03 / 0.07 / 0.05 / 0.44 / 0.98 / 0.06 / 4.58 / 1.4 / 0.07
Amino acid metabolism / 0.04 / 0.15 / 0.24 / 0.4 / 0.2 / 0.13 / 2.03 / 1.19 / 0.24
Metabolism of cofactors and vitamins / 0 / 0.01 / 0.01 / 0.01 / 0.01 / 0.01 / 0.01 / 0.03 / 0.01

* Oxidative phosphorylation, electron transport system complex III and IV, ATP synthase.

** Alcohol dehydrogenases, pyruvate decarboxylase, acetyl-CoA synthetase, acetyl-CoA hydrolase, formaldehyde dehydrogenase, homocitrate synthase, S-formylglutathione hydrolase.

Table S3. Comparison of predicted and measured gene dispensability on rich medium with glucose (YPD).

We simulated single gene knock-outs of yeast metabolic enzymes using the MOMA4 method (using FBA to predict knock-out phenotypes gives identical results). Data on measured fitness effects were obtained from a large-scale study by Giaever et al.5 The table lists the number of true and false predictions of the model. To minimize confounding factors in designation of dispensability, multienzyme polypeptides, genes participating in protein complexes (according to the MIPS CYGD catalogue of annotated complexes) and genes with overlapping reading frames were excluded from the comparison. Knock-out phenotype predictions were not attempted for genes located on dead end pathways or for genes with functions not represented in the biomass equation6.

Number of genes
Correctly predicted lethal phenotype / 28
Wrongly predicted lethal phenotype / 18
Correctly predicted viable phenotype / 279
Wrongly predicted viable phenotype / 23

Overall prediction: 88.47%

Table S4. Causes of metabolic gene dispensability.

Predictions of the metabolic model is presented for 223 experimentally verified non-essential metabolic genes on rich medium with glucose (YPD). Based on the model's prediction and information on presence of isoenzymes, we estimated the contribution of three suggested mechanisms to explain gene dispensability: environmental specificity, compensation by gene duplication and compensation by flux reorganisation. For these estimations we considered only genes that have correctly predicted knock-out phenotypes under rich medium (N=212). For 28 dispensable genes both duplicate gene copy is present and network compensation is predicted to act. We used these cases to obtain an upper estimate on the contribution of isoenzymes and network compensation, respectively, by supposing that loss of these genes are compensated by either isoenzymes or flux reorganisation alone.

Model's prediction / Interpretation / Number of dispensable genes / Contribution to dispensability
(false predictions are not counted)
Genes with zero predicted flux under rich-glucose growth condition / genes having zero predicted flux under all conditions investigated / environmental specificity? / 66 / 31.1%
genes having non-zero predicted flux under at least one other growth condition / environmental specificity / 79 / 37.2%
Genes with non-zero predicted flux under rich-glucose growth condition / single copy genes predicted to catalyse essential reactions / false prediction / 11 / (not counted)
duplicate genes predicted to catalyse essential reactions / compensation by isoenzyme / 31 / lower estimate: 14.6%
upper estimate: 27.8%
single copy genes predicted to catalyse dispensable reactions / compensation by flux reorganisation / 8 / lower estimate: 3.8%
upper estimate: 17%
duplicate genes predicted to catalyse dispensable reactions / compensation by either isoenzyme or flux reorganisation / 28 / (counted in the above two categories to obtain upper estimates)

Table S5. What explains failure of compensation by a duplicate?

The table lists isoenzyme pairs predicted to be active under glucose rich conditions and one member being essential in vivo (denoted by †). Seemingly, genes localized to mitochondria and / or expressed at low levels on rich medium are not able to compensate the deletion of their cytoplasmic and / or highly expressed paralogs. Data on localization were collected from source of metabolic reconstruction7 and MIPS CYGD8. Expression level (mRNA molecules/cell) was measured on rich-glucose medium by Holstege et al.9 using microarray technology.

EC number / Enzymatic function / Duplicate isoenzyme gene pair / Subcellular localization / Expression level (mRNA molecules/cell) / Possible explanation for lack of compensation
1.6.4.5 / thioredoxin reductase / TRR1 † / cytoplasm / 14.7 / different cellular location and /or low activity of TRR2 under fermentative conditions
TRR2 / mitochondria / 1.1
3.6.1.1 / inorganic pyrophosphatase / IPP1 † / cytoplasm / 12.3 / different cellular location and /or low activity of IPP2 under fermentative conditions
IPP2 / mitochondria / 0.6

Figure S1. The extent of flux reorganisation correlates with predicted fitness effect of the knock-out.

The extent of flux reorganisation (D) of knock-out was measured as the square of the Euclidean distance between wild-type and knock-out flux configurations in flux space (we considered only knock-outs of enzymes predicted to catalyse dispensable reactions, N=47). The logarithm of D positively correlates with the logarithm of fitness effect (S) of knock-out (Pearson correlation: rlogD-logS = 0.823, P<10-11, see figure below). As the absolute flux value (Flux) of a given enzyme also correlates with these two variables, we performed a multiple regression analysis to see whether the correlation between D and S remains. We report that the extent of flux reorganisation positively correlates with both flux value (r logD - logFlux | logS =0.92, P<10-16) and fitness effect (r logD - logS | logFlux =0.47, P<10-3) of the knock-out.

Table S6. No association between importance of reactions and presence of isoenzymes.

Single enzyme knock-outs were simulated under 9 growth conditions. Reactions catalysed by isoenzymes were considered as a single flux, thus obtaining predictions on the dispensability of the underlying reaction. We investigated 213 EC numbers with appropriate information available on the presence or absence of duplicated isoenzyme (see Methods). Our analysis shows that important reactions are not more likely to be catalysed by duplicated isoenzymes irrespective of the growth medium.

Simulated growth medium

Percentage of isoenzymes among...

P-value

(chi2 test)

enzymes catalysing essential reactions / enzymes catalysing dispensable reactions
rich-glucose (YPD), low O2 / 32.1% / 27.5% / 0.523
minimal-glucose, low O2 / 28% / 29.2% / 0.846
minimal-glucose, anaerob / 29.5% / 28% / 0.81
minimal-ethanol, low O2 / 29.5% / 27.8 / 0.778
minimal-acetate, low O2 / 28.8% / 28.4% / 0.948
minimal-glucose, limited by C / 28.3% / 28.9% / 0.915
minimal-glucose, limited by N / 29.2% / 28% / 0.845
minimal-glucose, limited by P / 32.1% / 25% / 0.251
minimal-glucose, limited by S / 27.6% / 29.6% / 0.746

Table S7. Reactions catalysed by isoenzymes have larger fluxes than those catalysed by solo copy enzymes.

The table compares predicted absolute flux values of reactions catalysed by isoenzymes and single copy enzymes (flux level for each enzyme was defined as the sum of absolute flux values of reactions catalysed by the given enzyme). For comparisons across growth conditions, fluxes were normalised by the biomass production rate. For clarity, we considered here only reactions taking place in the cytoplasm and an enzyme was designated to have isoenzyme if a duplicate gene copy catalysing the same cytoplasmic reaction is reported. Only enzymes predicted to have non-zero fluxes were used for the comparisons.