purushottam dixit recent advances in -omics technologies allow collection of diverse...
TRANSCRIPT
Research Statement Purushottam Dixit
Introduction: Recent advances in -omics technologies allow collection of diverse in-
formation about biological systems. The data avalanche has led of several descriptive ap-
proaches. These statistical efforts need to be complemented with a mechanistic models of
the underlying biological processes. This task demands a comprehensive knowledge of the
physical and chemical phenomena that constrain biology from the nano- to the macro-scale.
In my research career, I have amassed an array of quantitative skills ranging from molecular
dynamics simulation to computational population genetics. I have an extensive experience
in analyzing -omics experimental data such as metabolomics, genomics, and phosphopro-
teomics. My skills place me in an ideal position to quantitatively answer interesting biological
questions. My future research will pursue two distinct directions: (I) Gaining mechanistic
insights into cell-to-cell variability in mammalian signaling networks and (II) Identifying
the forces that shape bacterial genome evolution.
Aim I: A key challenge in the mechanistic understanding of mammalian signaling is
cell population heterogeneity of signaling dynamics. This heterogeneity has important func-
tional consequences, for example, in emergence of chemotherapy resistant subpopulations
in tumors. I am interested in identifying biochemical network parameters that dictate
heterogeneity in mammalian signaling networks and the downstream cellular phenotypes.
FIG. 1: Time evolution of the distribution ofpAkt levels upon EGF stimulation (measured,filled circles) and the corresponding predictions(dashed lines).
Recent advances in single cell measure-
ments allow us to measure the variability in
signaling network components. We have de-
veloped a computational framework to infer
distribution of network parameters from sin-
gle cell data based on the maximum entropy
(ME) principle1,2. Among all possible dis-
tributions that agree with experiments, ME
chooses the one with the least amount of
overfitting. Using the framework, we have
studied the heterogeneity in the epidermal
growth factor (EGF)/Protein kinase B (Akt) pathway. Aberrations in the EGF/Akt path-
way are implicated in many cancers. In a combined computational and experimental study,
we measured heterogeneity in Akt phosphorylation levels (pAkt) upon stimulation with
EGF (see Fig. 1). From the inferred parameter distribution, we could identify key factors
that governed the heterogeneity in Akt activity. For example, we showed that the cell to
cell variability in the abundance of surface EGF receptors (EGFRs) largely explained the
variability in pAkt levels at steady state.
Chemotherapy drugs, for example, gefitinib and lapatinib, bind to EGFRs and downreg-
ulate their kinase activity. However, patients that benefit from EGFR inhibitors eventually
develop drug resistance. Similarly, while a majority of ‘sensitive’ cells in a cultured cell pop-
ulation deactivate the EGF/Akt pathway upon drug action, a small fraction of ‘resistant’
cells show sustained Akt phosphorylation in the presence of the drug. Notably, the role of
non-genetic heterogeneity in the emergence of tumors resistance is now well recognized. I
1
Research Statement Purushottam Dixit
want to understand the biochemical origins of drug resistance in the EGF/Akt pathway.
Towards that end, single cell phosphoproteomics data can first be collected across multi-
ple doses of EGF and drugs for several cancer cell lines. Next, using the ME framework, the
joint distribution over network parameters can be inferred. The inferred distribution will al-
low us to identify in silico the cells with phenotypes of interest, for example, ‘resistant’ cells
that exhibit high levels of pAkt even in the presence of EGFR inhibitors. Next, we can com-
putationally identify biochemical parameters (intracellular protein abundances and reaction
rates) that characterize the ‘resistant’ subpopulation. These predicted subpopulations can
lead to synergistic treatments. For example, if we predict that Phosphoinositide 3-kinase
(PI3K) levels correlate with high Akt activity in resistant cells, we can experimentally verify
that simultaneously targetting PI3K and EGFR overcomes gefitinib resistance.
More broadly, I envision studying the heterogeneity in the transcriptional network down-
stream of the EGF/Akt pathway3, as well as other signaling pathways. Examples include
the hepatocyte growth factor cascade responsible in organ regeneration and wound healing,
as well as the insulin-dependent cascade implicated in diabetes and several cancers.
Aim II. Bacteria are the most ubiquitous organisms on Earth. They play a significant
role in human health as well as the environment. Bacteria are capable of rapid genetic evo-
lution with new traits spreading in the population within days to months. Notably, multiple
evolutionary forces such as mutations, recombination (swapping of DNA in the population),
and adaptation simultaneously affect the genetic diversity in bacterial populations4,5. I am
interested in quantifying the relative importance of evolutionary forces in bacteria in order
to better understand how environmental conditions, such as host diet, and natural selection
dictate the innovation and spread of phenotypes in bacterial populations.
Evolutionary simulations of realistic bacterial populations are numerically infeasible ow-
ing to large effective population sizes (Ne > 109)4,5. As a result, several fundamental ques-
tions about bacterial evolution remain unresolved. For example, it remains unknown whether
bacteria can retain clonal ancestry in the presence of recombination or whether recombina-
tion can overcome geographical isolation and prevent formation of new species. We have
developed4,5 a computational evolutionary framework to model the evolution of the diver-
gence between pairs of genomes using the coalescent theory. The computationally efficient
framework provides a computationally feasible alternative to detailed simulations of large
bacterial populations.
Using the framework, we studied the competition between clonality and recombination4,5.
While asexual reproduction imposes a clonal structure in bacteria, recombination of small
genetic fragments locally destroys clonal ancestry. We showed that depending on the popula-
tion genetic diversity and the rate of recombinations relative to mutations, the competition
between clonality and recombination results in two possible ‘phases’ in the evolutionary
dynamics that corresponded to clonal and sexual bacteria (see Fig. 2a)4. Our analysis
also suggested computational metrics to quantify the relative strength of mutation and re-
combination. For example, we predicted that the histogram of local density of nucleotide
differences (SNPs) between a pair of genomes is a combination of a Poisson and an expo-
2
Research Statement Purushottam Dixit
nential distribution4 corresponding to de novo mutations in clonal regions and recombined
regions respectively. With that insight, in a population genomics study of E. coli, we iden-
tified clonally inherited and recombined regions in wild E. coli strains (see Fig. 2b for a
comparison between strains K12 and B)5.
FIG. 2: a) Classification of bacterial species as clonal and
sexual. b) A histogram of SNPs in small genomic segments
distinguishes clonally inherited and recombined segments in
E. coli strains K12 and B.
Using the framework, I want
to address fundamental questions
about the interaction between
evolutionary forces. For exam-
ple, recombination as opposed to
de novo innovation is responsi-
ble for the rapid spread of antibi-
otic resistance. However, ecolog-
ical isolation suppresses recom-
bination. Thus, it is crucial to
identify the conditions in which
recombination can overcome eco-
logical isolation. Another example is the competition between recombination and selection.
While adaptive mutations allow strains to proliferate rapidly, frequent exchange of these
mutations nullifies their relative advantage. At the same time, low genetic exchange results
in lowered overall population fitness (also known as the Muller’s ratchet). However, the
optimal recombination rates in the presence of selection are not systematically explored. I
can address these questions using our coalescent framework. For example, I will incorporate
the distribution of coalescent times among recombining regions to represent adaptation and
ecological isolation. I will then study how relative strengths of these forces affect genetic
diversity by simulating the divergence in bacterial populations. This work will have broad ap-
plications in understanding bacterial evolution, for example, in the study of host-associated
microbial communities6,7, determinants of host-pathogen interactions, and stability of crucial
phenotypes in oceanic bacteria.
Conclusion: Our ability to collect -omics data is ever increasing. The data deluge has
led to a surge in descriptive models in quantitative biology. I believe that an essential task for
the next generation of scientists is to complement the descriptive approach with mechanistic
models that integrate the data with the constraints that govern biology. As evidenced in this
statement, this undertaking will require tools from diverse fields such as thermodynamics,
reaction chemistry, and population genetics. Given my research experience, I believe that I
am an ideal candidate for this task.
Reference: [1] Dixit, et al., 10.1101/137513. [2] Dixit et al. J. Chem. Phys., 2018.
[3] Lyashenko*, Niepel*, Dixit*, Lim, Sorger, and Vitkup, 10.1101/158774. [4] Dixit, et
al. Genetics, 2017. [5] Dixit et al. PNAS, 2015. [6] Ji*, Sheth*, Dixit, and Vitkup,
10.1101/370676. [7] Ji*, Sheth*, Dixit, Wang, and Vitkup 10.1101/310649.
3