Leaders in Pharmaceutical Business Intelligence (LPBI) Group

Metabolomic analysis of two leukemia cell lines. I.

Larry H. Bernstein, MD, FCAP, Reviewer and Curator

Leaders in Pharmaceutical Intelligence

 

I have just posted a review of metabolomics.  In the last few weeks, the Human Metabolome was published.  I am hopeful that my decision has taken the right path to prepare my readers adequately if they will have read the articles that preceded this.  I pondered how I would present this massive piece of work, a study using two leukemia cell lines and mapping the features and differences that drive the carcinogenesis pathways, and identify key metabolic signatures in these differentiated cell types and subtypes.  It is a culmination of a large collaborative effort that required cell culture, enzymatic assays, mass spectrometry, the full measure of which I need not present here, and a very superb validation of the model with a description of method limitations or conflicts.  This is a beautiful piece of work carried out by a small group by today’s standards.

I shall begin this by asking a few questions that will be addressed in the article, which I need to beak up into parts, to draw the readers in more effectively.

Q 1. What metabolic pathways do you expect to have the largest role in the study about to be presented?

Q2. What are the largest metabolic differences that one expects to see in compairing the two lymphoblastic cell lines?

Q3. What methods would be used to extract the information based on external metabolites, enzymes, substrates, etc., to create the model for the cell internal metabolome?

 

 

Abstract

Metabolic models can provide a mechanistic framework to analyze information-rich omics data sets, and are increasingly being used

• to investigate metabolic alternations in human diseases.

An expression of the altered metabolic pathway utilization is

• the selection of metabolites consumed and released by cells.

However, methods for the inference of intracellular metabolic states from extracellular measurements in the context of metabolic models

• remain underdeveloped compared to methods for other omics data.

Herein, we describe a workflow for such an integrative analysis

• extracting the information from extracellular metabolomics data.

We demonstrate, using the lymphoblastic leukemia cell lines Molt-4 and CCRF-CEM, how

• our methods can reveal differences in cell metabolism.

Our models explain metabolite uptake and secretion by

• predicting a more glycolytic phenotype for the CCRF-CEM model and
• a more oxidative phenotype for the Molt-4 model, which
• was supported by our experimental data.

Gene expression analysis revealed altered expression of gene products at

• key regulatory steps in those central metabolic pathways,

and literature query emphasized

• the role of these genes in cancer metabolism.

Moreover, in silico gene knock-outs identified

• unique control points for each cell line model, e.g., phosphoglycerate dehydrogenase for the Molt-4 model.

Thus, our workflow is well suited to the characterization of cellular metabolic traits based on

• extracellular metabolomic data, and
• it allows the integration of multiple omics data sets into a cohesive picture based on a defined model context.

Keywords Constraint-based modeling _ Metabolomics _Multi-omics _ Metabolic network _ Transcriptomics

 

Reviewer Summary:

1. A model is introduced to demonstrate a lymphocytic integrated data set using to cell lines.
2. The method is required to integrate extracted data sets from extracellular metabolites to an intracellular picture of cellular metabolism for each cell line.
3. The method predicts a more glycolytic or a more oxidative metabolic framework for one or the othe cell line.
4. The genetic phenotypes differ with a unique control point for each cell line.
5. The model presents an integration of omics data sets into a cohesive picture based on the model context.

Without having seen the full presentation –

1. Is the method a snapshot of the neoplastic processes described?
2. Does the model give insight into the cellular metabolism of an initial cell state for either one or both cell lines?
3. Would one be able to predict a therapeutic strategy based on the model for either or both cell lines?

Before proceeding further into the study, I would conjecture that there is no way of knowing the initial state ( consistent with what is described by Ilya Prigogine for a self-organizing system) because the model is based on the study of cultured cells that had an unknown metabolic control profile in a host proliferating bone marrow that is likely B-cell origin.  So this is a snapshot of a stable state of two incubated cell lines.  Then the question that is raised is whether there is not only a genetic-phenotypic relationship between the cells in culture and the external metabolites produced, but also whether differences can be discerned between the  internal metabolic constructions that would fit into a family tree.

 

Introduction

Modern high-throughput techniques

• have increased the pace of biological data generation.

Also referred to as the ‘‘omics avalanche’’, this wealth of data

• provides great opportunities for metabolic discovery.

Omics data sets contain a snapshot of almost the entire repertoire of

• mRNA, protein, or metabolites at a given time point or
• under a particular set of experimental conditions.

Because of the high complexity of the data sets,

• computational modeling is essential for their integrative analysis.

Currently, such data analysis

• is a bottleneck in the research process and
• methods are needed to facilitate the use of these data sets, e.g.,

1. through meta-analysis of data available in public databases
   [e.g., the human protein atlas (Uhlen et al. 2010)
2. or the gene expression omnibus (Barrett  et al.  2011)], and
3. to increase the accessibility of valuable information
   for the biomedical research community.

Constraint-based modeling and analysis (COBRA) is

• a computational approach that has been successfully used
• to investigate and engineer microbial metabolism through
  the prediction of steady-states (Durot et al.2009).

The basis of COBRA is network reconstruction: networks are assembled

1. in a bottom-up fashion based on genomic data and
2. extensive organism-specific information from the literature.

Metabolic reconstructions

1. capture information on the known biochemical transformations
   taking place in a target organism
2. to generate a biochemical, genetic and genomic knowledge base
   (Reed et al. 2006).

Once assembled, a metabolic reconstruction

• can be converted into a mathematical model
  (Thiele and Palsson 2010), and
• model properties can be interrogated using a great variety of methods
  (Schellenberger et al. 2011).

The ability of COBRA models to represent

• genotype–phenotype and environment–phenotype relationships
• arises through the imposition of constraints,
• which limit the system to a subset of possible network states
  (Lewis et al. 2012).

Currently, COBRA models exist for more than 100 organisms, including humans
(Duarte et al. 2007; Thiele et al. 2013).

Since the first human metabolic reconstruction was described
[Recon 1 (Duarte et al. 2007)],

• biomedical applications of COBRA have increased
  (Bordbar and Palsson 2012).

One way to contextualize networks is to

• define their system boundaries
• according to the metabolic states of the system,
  e.g., disease or dietary regimes.

The consequences of the applied constraints

• can then be assessed for the entire network
  (Sahoo and Thiele 2013).

Additionally, omics data sets have frequently been used

• to generate cell-type or condition-specific metabolic models.

Models exist for specific cell types, such as

• enterocytes (Sahoo and Thiele2013),
• macrophages (Bordbar et al. 2010), and
• adipocytes (Mardinoglu et al. 2013), and
• even multi-cell assemblies that represent
  the interactions of brain cells (Lewis et al. 2010).

All of these cell type specific models,

• except the enterocyte reconstruction
• were generated based on omics data sets.

Cell-type-specific models have been used

• to study diverse human disease conditions.

For example, an adipocyte model was generated using

• transcriptomic,
• proteomic, and
• metabolomics data.

This model was subsequently used to investigate

• metabolic alternations in adipocytes
• that would allow for the stratification of obese patients
  (Mardinoglu et al. 2013).

One highly active field within the biomedical applications of COBRA is

• cancer metabolism (Jerby and Ruppin, 2012).

Omics-driven large-scale models have been used

• to predict drug targets (Folger et al. 2011; Jerby et al. 2012).

A cancer model was generated using

• multiple gene expression data sets and
• subsequently used to predict synthetic lethal gene pairs
• as potential drug targets selective for the cancer model,
• but non-toxic to the global model (Recon 1),
• a consequence of the reduced redundancy in the
  cancer specific model (Folger et al. 2011).

In a follow up study, lethal synergy between

• FH and enzymes of the heme metabolic pathway
  were experimentally validated and
• resolved the mechanism by which FH deficient cells,
  e.g., in renal-cell cancer cells
• survive a non-functional TCA cycle (Frezza et al. 2011).

Contextualized models, which contain only 

• the subset of reactions active in 
• a particular tissue (or cell-) type,
• can be generated in different ways
  (Becker and Palsson, 2008; Jerby et al. 2010).

However, the existing algorithms mainly consider

• gene expression and proteomic data to define the reaction sets
• that comprise the contextualized metabolic models.

These subset of reactions are usually defined based on

• the expression or absence of expression of the genes or proteins
  (present and absent calls), or
• inferred from expression values or differential gene expression.

Comprehensive reviews of the methods are available
(Blazier and Papin, 2012; Hyduke et al. 2013).

Only the compilation of a large set of omics data sets

• can result in a tissue (or cell-type) specific metabolic model, whereas

the representation of one particular experimental condition is achieved through

• the integration of omics data set generated from one experiment only
  (condition-specific cell line model).

Recently, metabolomic data sets

• have become more comprehensive and using these data sets allow
• direct determination of the metabolic network components (the metabolites).

Additionally, metabolomics has proven to be

1. stable,
2. relatively inexpensive, and
3. highly reproducible
   (Antonucci et al. 2012).

These factors make metabolomic data sets

•  particularly valuable for interrogation of metabolic phenotypes. 

Thus, the integration of these data sets is now an active field of research
(Li et al. 2013; Mo et al. 2009; Paglia et al. 2012b; Schmidt et al. 2013).

Generally, metabolomic data can be incorporated into metabolic networks as

1. qualitative,
2. quantitative, and
3. thermodynamic constraints
   (Fleming et al. 2009; Mo et al. 2009).

Mo et al. used metabolites detected in the spent medium
of yeast cells to determine

• intracellular flux states through a sampling analysis (Mo et al. 2009),
• which allowed unbiased interrogation of the possible network states
  (Schellenberger and Palsson 2009)
• and prediction of internal pathway use.

Such analyses have also been used

• to reveal the effects of enzymopathies on red blood cells (Price et al. 2004),
• to study effects of diet on diabetes (Thiele et al. 2005) and
• to define macrophage metabolic states (Bordbar et al. 2010).

This type of analysis is available as a function in the COBRA toolbox
(Schellenberger et al. 2011).

 

 

 

In this study, we established a workflow for the generation and analysis of

• condition-specific metabolic cell line models that
• can facilitate the interpretation of metabolomic data.

Our modeling yields meaningful predictions regarding

• metabolic differences between two lymphoblastic leukemia cell lines
  (Fig. 1A).

 

 

 

http://link.springer.com/static-content/images/404/art%253A10.1007%252
Fs11306-014-0721-3/MediaObjects/11306_2014_721_Fig1_HTML.gif

Fig. 1

A  Combined experimental and computational pipeline to study human metabolism.
Experimental work and omics data analysis steps precede computational modeling. Model

• predictions are validated based on targeted experimental data.

Metabolomic and transcriptomic data are used for

• model refinement and submodel extraction.

Functional analysis methods are used to characterize

• the metabolism of the cell-line models and compare it to additional experimental
  data.

The validated models are subsequently 

• used for the prediction of drug targets.

B Uptake and secretion pattern of model.
All metabolite uptakes and secretions that were mapped during model
generation are shown.
Metabolite uptakes are depicted on the left, and

• secreted metabolites are shown on the right.

A number of metabolite exchanges mapped to the model

• were unique to one cell line.

Differences between cell lines were used to set

• quantitative constraints for the sampling analysis.

C Statistics about the cell line-specific network generation.

D  Quantitative constraints.
For the sampling analysis, an additional

• set of constraints was imposed on the cell line specific models,
• emphasizing the differences in metabolite uptake and secretion between cell lines.

Higher uptake of a metabolite was allowed in the model of the cell line

• that consumed more of the metabolite in vitro, whereas
• the supply was restricted for the model with lower in vitro uptake.

This was done by establishing the same ratio between the models bounds as detected in vitro.
X denotes the factor(slope ratio) that

1. distinguishes the bounds, and
2. which was individual for each metabolite.

• (a) The uptake of a metabolite could be x times higher in CCRF-CEM cells,
  (b) the metabolite uptake could be x times higher in Molt-4,
  (c) metabolite secretion could be x times higher in CCRF-CEM, or
  (d) metabolite secretion could be x times higher in Molt-4 cells. LOD limit of detection.

The consequence of the adjustment was, in case of uptake, that  one model

1. was constrained to a lower metabolite uptake (A, B), and the difference
2. depended on the ratio detected in vitro.

In case of secretion,

• one model had to secrete more of the metabolite, and again

the difference depended on

• the experimental difference detected between the cell lines.

Q5. What is your expectation that this type of integrative approach could be used for facilitating medical data interpretations?

The most inventive approach was made years ago by using data constructions from the medical literature by a pioneer in the medical record development, but the technology was  not what it is today, and the cost of data input was high.  Nevertheless, the data acquisition would not be uniform across institutions, except for those that belong to a consolidated network with all of the data in the cloud, and the calculations would be carried out with a separate engine.  However, whether the uniform capture of the massive amount of data needed is not possible in the near foreseeable future.  There is no accurate way of assessing the system cost, and predicting the benefits.  In carrying this model forward there has to be a minimal amount of insufficient data.  The developments in the regulatory sphere have created a high barrier.

This concludes a first portion of this presentation.