This page contains the whole text of the
following article:
Athel Cornish-Bowden and María Luz Cárdenas (2001)
Complex networks of interactions connect genes to
phenotypes
Trends in Biochemical Sciences
26, 463–465, which was published as a
Research Update commentary on T. Ideker
et
al. (2001) Integrated genomic and
proteomic analyses of a systematically perturbed metabolic
network
Science 292, 929–934.
An Acrobat file of the paper as printed
in Trends in Biochemical Sciences (© 2001) is also available, with permission from
Elsevier
Science. Readers may download and print single copies for personal research and study.
Recent work on the galactose-utilization pathway of yeast has shown how transcriptome and proteome data can be combined to deduce a network of hundreds of genes involved in protein–protein and protein–DNA interactions, leading to a picture of how the pathway is regulated that is clearer and more complete than what was previously known.
According to William Bains,
The genome has turned out to be a relatively poor
source of explanation for the differences between cells or
between people
1. Negative
statements of this sort contrast wildly with the
triumphalism that characterized reports on genome projects
only a few years ago. However, they are beginning to be classed as no more than
realistic; hardly anyone can have failed to notice that
the capacity of genome sequencing to generate mountains of
new data has far outstripped our capacity to make sense of
it all. For example, the WIT
integrated pathway-genome database now contains
information about more than 80 genomes; one year ago there
were a few more than 40. Clearly there is an urgent need
for more efficient and powerful methods of data analysis,
because without these a fully sequenced genome is hardly
more useful than a complete list of telephone numbers
witout all the associated names. The genomic problem is
much more complex, as assigning a molecular function to
every gene is only a beginning: afterwards the network of
all the interactions between genes has to be established to
see how effects on the whole system are produced; this
requires integration of many different kinds of methods.
DNA microarrays already enable the mRNA expression levels of almost all the genes in an organism to be measured2; corresponding techniques to measure protein expression are being developed3. Applied separately, each of these is insufficient to predict new system properties. However, in a recent paper4 Leroy Hood and his colleagues have shown how these two measurements can be usefully combined with one another to validate and extend the existing model of how galactose utilization is regulated in yeast5. Their integration of mRNA and protein expression responses with the global set of protein–protein and protein–DNA interactions has allowed them to deduce a network involving hundreds of genes.
| Gene | Type of product | Function of product |
|---|---|---|
| GAL1 | Enzyme | Galactokinase |
| GAL2 | Transporter | Galactose transporter |
| GAL3 | Regulator | Relieves effect of Gal80p |
| GAL4 | Regulator | Activates expression of all other GAL genes |
| GAL5 | Enzyme | Phosphoglucomutase |
| GAL6 | Regulator | Acts in a drug-resistance pathway |
| GAL7 | Enzyme | Gal-1-P Glc-1-P uridyltransferase |
| GAL10 | Enzyme | UDPgalactose epimerase |
| GAL80 | Regulator | Represses effect of Gal4p |
Combining deletions of each of nine galactose-utilization genes (Table 1) with two dietary states, with and without galactose, the authors compared the effects of 20 perturbations on cellular gene expression. They found that the levels of 997 mRNAsvaried significantly with one or more of these perturbations — a far cry from the naive expectation, still common, that deletion (or overexpression) of one gene should affect expression of just that gene. Although widespread pleiotropy of this kind was predicted and understood many years ago6 by the few who had the vision to realize that systems need to be studied as systems, and not as mere collections of parts, it has only become a matter of common observation much more recently7. Underlining the need to look at both mRNA and protein expression patterns and not to rely on just one or the other to provide the whole story, the correlation between these two patterns in the galactose utilization study proved to be far from perfect: some protein expression levels remained virtually unchanged despite large changes in the expression of mRNA; others changed substantially with negligible changes in mRNA levels. By and large, however, the plot of one against the other yielded an intelligible scatter, with genes of functionally related roles appearing in clusters: one such cluster is provided by GAL1, GAL7 and GAL10, three genes that code for three enzymes specific to galactose metabolism.
Authors of a proof-of-principle
paper, designed to show
that a proposed new method is useful, need to steer a careful
course between the Scylla of too little novelty and the
Charybdis of too much. Just confirming the validity of a
previous model suggests that the new approach might be valid
while satisfying no obvious need; large amounts of new
information with little confirmation of what was known
before raises doubts about the reliability of the new approach. Hood and his colleagues4 have the balance about right. They
used the existing biochemical model of regulation in the
galactose-utilization pathway to predict the effects
of their 20 perturbations on the expression levels of the
GAL genes. Many of these predictions
would have been obvious without a model, for example that
deleting GAL1 would lower the expression of
GAL1, but the majority were less obvious and were, in
general, confirmed by the measurements. For example, when
galactose is available the product of GAL4 is
believed to increase the expression of all the genes
that code for enzymes that catalyse the transformation of
galactose into glucose 6-phosphate, so deletion of
GAL4 should decrease the amounts of these enzymes
synthesized during growth in the presence of galactose, but
have little or no effect in the absence of galactose. This is
in general what was observed, although expression of
GAL5 did not decrease when GAL4 was deleted
(if anything it increased slightly). This makes sense when
one recognizes that the product of
GAL5, phosphoglucomutase, has
important metabolic roles to fulfil regardless of galactose
utilization (so much so that its
classification as a GAL gene is surprising). The enzymes
that behaved as expected, such as galactokinase (Gal1p), were
specific to galactose metabolism.
The preliminary model contained a mechanism to suggest how the presence of galactose in the cell might stimulate its metabolism — intracellular galactose activates the Gal3p-mediated inhibition of the action of Gal80p on Gal4p, with a net result being activaton of the genes involved in the galactose pathway. In general, the GAL gene expression results agreed with this, but, of course, the model does not predict the changes in expression observed for the hundreds of other genes. Moreover, deletion of the gene encoding either uridyltransferase (Gal7p) or UDPgalactose epimerase (Gal10p), two enzymes needed to convert galactose 1-phosphate into glucose 1-phosphate, also resulted in decreased expression of the other galactose enzymes. Although the original model did not predict this observation, it did not require radical revision to accommodate it. The immediate effect of inactivation of either enzyme would be accumulation of galactose 1-phosphate: if this had an intrinsic regulatory effect to repress synthesis of other pathway enzymes this would provide an indirect way for deletion of GAL7 or GAL10 to repress synthesis of these enzymes. This hypothesis was tested by deleting the gene for galactokinase (Gal1p) at the same time as deleting GAL7 or GAL10, thus avoiding accumulation of galactose 1-phosphate. Sure enough, the profile over all genes of a gal1Δgal10Δ mutant resembled that of a gal1Δ mutant in the presence of galactose, not that of a gal10Δ mutant, confirming that the properties of the gal10Δ and gal7Δ mutants were caused by accumulation of galactose 1-phosphate, and not by a purely genetic effect such as interaction between chromosomes.
Fig. 1. Growth rates of different deletion mutants in the presence and absence of galactose. The wild type is shown by a black symbol, deletions of genes for transporters or enzymes by red symbols, and deletions of regulatory genes by green symbols. The green arrows show the regulatory effect of Gal4p on the transporter and metabolic enzymes, and the red arrows show inhibitory effects of other genes. The pathway from galactose via galactose 1-phosphate to glucose 6-phosphate is superimposed on the red symbols. Note the clear separation between those that act before and after galactose 1-phosphate in the pathway. The vertical dotted line maps the growth rate of the wild type in the absence of galactose onto the scale for growth in the presence of galactose.
The potentially harmful effects of galactose 1-phosphate accumulation are also evident in the growth rates of the different mutants (Fig. 1). In the absence of galactose none of the deletions apart from gal80Δ had very much effect on growth, but in the presence of galactose a clear separation between two groups of genes became evident: any of the regulatory genes or those for processes leading to formation of galactose 1-phosphate could be deleted without producing a growth rate much less than that of the wild type in the absence of galactose; by contrast, deleting genes for enzymes needed for metabolism of galactose 1-phosphate resulted in very slow growth in the presence of galactose. We see here an example of the large effect that loss of an enzyme activity can have on metabolite concentrations, typically much larger than its effects on fluxes. This in part accounts for the growing interest in the metabolome (i.e. the complete set of metabolite concentrations) in studies of functional genomics, because measurements of metabolite concentrations can often reveal the effects of genes that otherwise appear to be silent8, 9.
The observed pleiotropy underlines the crucial point that genes act in concert with one another and with the environment. The more complex the level at which one seeks to explain a living system, the greater the need to examine the network of interactions that lie behind the genome. This investigation of galactose metabolism4 represents a clear step in this direction, opening the door to studies of systems with much less prior information about the genes involved and their regulation than was available in this case.