Biology Reference
In-Depth Information
the genes most required for resistance to the molecule. The
biological 'richness' of each profile is readily appreciated
in the GO enrichment maps generated for each small
molecule screened (e.g., Figure 8.2B). Provided the
profiled small molecules inhibit growth, HIPHOP profiles
that are highly correlated indicate that the corresponding
small molecules induce similar cellular responses, sug-
gesting that they may have similar MOAs. We previously
defined the quantitative term 'co-inhibition', a metric
defined by the pairwise correlation between HIPHOP
profiles
[54]
. From a global perspective, a weak but
significant correlation between structural similarity of the
profiled small molecules and co-inhibition was observed,
suggesting that chemical structure influences patterns of
inhibition, as has often been observed. Perhaps more
interestingly, we observed a more significant association
between co-inhibition and small molecules that belong to
the same therapeutic class, compared to those that are
structurally similar
[54]
. For example, clozapine and pro-
piomazine clustered together based on co-inhibition and
both are annotated as neuroleptics. Furthermore, of the
pairs of small molecules that were co-inhibited and shared
therapeutic class, more than 70% did not have significant
structural similarity. This indicates that molecules with
very different structures can result in similar biological
responses. This finding was surprising, particularly as most
of the therapeutic targets do not have a yeast homolog.
However, the observed co-inhibition may arise from
a variety of sources, for example structurally diverse
molecules that inhibit different proteins within the same
pathway, or different structures that inhibit the same target,
such as those that may arise from 'me-too' targets, where
different pharmaceutical companies search for molecules
that inhibit the same target but must be structurally distinct
(to avoid intellectual property issues), and strategic
medicinal chemistry tactics may be in play. Whatever the
cause, the observed strong co-inhibition of drugs that
belong to the same therapeutic class suggests that yeast
HIPHOP assays may be useful for classifying therapies
and/or possibly for proposing alternatives.
acid metabolism, compared to measures of similar gene
signatures based on physical interactions, genetic interactions
or gene expression
[54]
. In our more recent dataset of ~3500
HIPHOP profiles, hierarchical clustering of all genes by co-
fitness defined a subcluster revealing the detailed pathway
required for each step on the road to translation (
Figure 8.3
,
discussed in further detail below;
[70]
).
We also found that essential genes are co-fit with other
essential genes more frequently than expected
[54]
. This
observation suggests that essential genes tend to work
together in 'essential processes'. Indeed, if we define
a complex as essential if more than 80% of its members are
essential, a significantly greater number of complexes are
essential in rich medium than expected by chance
[71]
.By
the same token, we exploited our HOP results to identify
conditionally essential complexes, that is, complexes where
more than 80% of the (non-essential) members are signif-
icantly sensitive in a given condition. This analysis
revealed that there are significantly more conditionally
essential complexes than expected by chance, in 40% of the
tested conditions
[54]
. In fact, vesicle transport genes
involved in complexes are, in general, sensitive to a large
number of diverse compounds, suggesting that these
complexes are required for the cellular response to chem-
ical stress. This finding supports and extends previous
findings on multi-drug resistance (MDR) (e.g.,
[21,26]
).
Significant co-fitness between two genes can be thought
of as a functional linkage in the context of chemical stress,
and as such, it is possible to construct a co-fitness network
where nodes represent genes and edges represent co-fitness
linkages. One could characterize such a network by, for
example: (1) identifying sets of genes that are highly co-fit
with one another to define functional modules relevant to
chemical stress; (2) identifying relationships between the
modules; and (3) defining global properties of the network,
e.g., by examining distributions of gene centrality, or the
shortest path length between genes. Importantly, the char-
acterization of a co-fitness network may provide novel
insights into the (re-)organization of genes, the dependen-
cies between genes, and the importance of particular genes
when the cell is under chemical stress conditions. Once
integrated with other types of biological networks derived
under chemical stress conditions as they become available,
an even richer system-wide picture of a chemically per-
turbed cell should be possible.
Co-Fitness Predicts Gene Function
Many systems-level datasets can be used to define
gene/protein signatures. For example, the genetic or pro-
tein
protein interactions of a gene/protein define a signature
for that gene/protein. Pairs of genes (as opposed to
compounds for co-inhibition) that are highly correlated in
a HIPHOP dataset can be described as being co-fit. Pairs of
co-fit genes often share function, and co-fitness predicts gene
function on par with or better than other genomic platforms.
Furthermore, when considering individual biological
processes, co-fitness performed particularly well in predict-
ing genes involved in certain processes, for example amino
e
The Multi-Drug Resistance Network
Global View of the Mechanisms Involved
in Cellular Resistance to Small Molecules
Classically, MDR genes are defined by genes involved in
xenobiotic metabolism, a process designed by the cell for
the detoxification of foreign substances. The proteins
