December 1, 2013 (Vol. 33, No. 21)

Richard A. A. Stein M.D., Ph.D.

“Science is more than a body of knowledge, it’s a way of thinking,” remarked Carl Sagan, and probably his words were never more powerfully relevant than for portraying one of the newest biomedical fields, systems biology.

A recent symposium inaugurating the department of systems biology at Columbia University Medical Center comes at a very auspicious time, one in which biomedical sciences, chemistry, physics, engineering, bioinformatics, and computer sciences are converging to shape a vibrant new discipline.

As Lee Goldman, M.D., M.P.H., executive vice president for health and biomedical sciences at the Columbia University College of Physicians and Surgeons pointed out during the opening remarks, this new field “represents so much about our future.”

Within a relatively short time, we have realized the possibility of sequencing and mapping the genome of virtually any organism. Concomitantly, as increasingly sophisticated technologies allow whole-genome sequences to be completed within hours to days, navigating the vast datasets has become the foremost challenge, opening a gap in our ability to understand and interpret their significance.

“While elucidating the genome has given us a tremendous amount of information, it has revealed very little about how those parts work together,” says Andrea Califano, Ph.D., professor and chair of the department of systems biology at Columbia. At the recent systems biology inaugural symposium, several investigators affiliated with the department unveiled research programs and interdisciplinary approaches that promise to integrate a plethora of knowledge on cellular and molecular processes into cross-disciplinary frameworks.


The homodimer of the ligand binding domain of the nuclear hormone receptor, LXR, is used to predict an interaction between another hormone receptor, PPARγ, and the homeodomain of PAX7. One subunit of the LXR homodimer is shown in red and the other in blue. A structure of the ligand binding domain of PPARγ, shown in green, is superimposed on the blue subunit, and a homology model of the homeodomain of PAX7 is shown in yellow, superimposed on the other. [Cliff Zhang, Donald Petrey, and Barry Honig—Howard Hughes Medical Institute/Columbia University]

Protein-Protein Interactions

“We wanted to put a structural face on protein-protein interactions that occur in cells over time,” said Barry Honig, Ph.D., professor of biochemistry and molecular biophysics. In parallel with emerging knowledge about the cellular ensemble of protein-protein interactions, the possibility of using structural information to describe protein-protein interfaces on a large scale has attracted increasing attention. “But there are challenges associated with this, and the main challenge is the existence of many more genes than structures,” Dr. Honig added.

Thousands of human proteins have at least one known domain structure, yet only approximately 500 protein-protein complex structures are available in the protein data bank. At the same time, approximately 80,000 interactions are described in publicly available databases, and many more are anticipated to occur in cells. “What we want to do is mine structural information about those interactions,” explained Dr. Honig.

One of the key approaches used by Dr. Honig and colleagues, homology modeling, takes advantage of the possibility of finding, given a query sequence of unknown structure, another sequence that is related through alignment and whose structure is known. Models of proteins or protein fragments can be analyzed by three-dimensional superimposition, in an approach that Dr. Honig coined as structural BLAST.

Building on this strategy, Dr. Honig and colleagues developed and validated an algorithm known as PrePPI (predicting protein-protein interactions), and used it to generate a database of over 300,000 predictions for protein interaction pairs. PrePPI is based on Bayesian statistics and relies on the input of structural and nonstructural information.

For a pair of query proteins thought to interact, structures for subunits from each protein are obtained from databases or are generated by homology modeling. After finding structural neighbors for each subunit, a template is identified for a pair of structural neighboring chains, and a model is built by superimposing individual subunits on their corresponding structural neighbors.

A single structure-derived score, combined from five different structure-based scores, generated to evaluate the model, reflects the likelihood that the interaction represents a true complex. Nonstructural information, such as functional and evolutionary similarities, is subsequently incorporated into the model. “This could not have been possible without applying systems biology thinking to structural biology,” said Dr. Honig.

Perturbations and Pathways

“I was fascinated by processes that have evolved to create such amazingly precise spatiotemporal gene expression patterns, and I was even more astounded to realize that information underlying these patterns is encoded in the sequence,” explained Saeed Tavazoie, Ph.D., professor of biochemistry and molecular biophysics. A major effort in Dr. Tavazoie’s lab has focused on the concept of predicting gene expression dynamics from information encoded in the sequence.

Decades ago, Dr. Tavazoie’s approach would have been difficult to implement. However, the advent of microarray technology, which allows the expression of thousands of genes to be surveyed simultaneously under a broad range of conditions, has been instrumental toward making this approach become reality.

Previously, Dr. Tavazoie and colleagues developed a computational pipeline to profile a number of gene expression perturbations occurring in a large number of different cellular states, and the genes were subsequently clustered into co-expression modules. “We then looked at the regulatory regions for the occurrence of de novo motifs that are enriched in these genes with regard to the background of the genome,” said Dr. Tavazoie.

These motifs were predicted to function as transcriptional regulator binding sites, and the strategy provided a powerful approach to perform reverse engineering in simple organisms. “However, going from yeast to humans has been more challenging, due to the scale and the complexity of the human genome,” reported Dr. Tavazoie.

To address this challenge, Dr. Tavazoie’s lab developed new algorithms based on information theory that enable sensitive and specific detection of transcriptional and post-transcriptional regulatory elements within the human genome. In particular, TEISER (Tool for Eliciting Informative Structural Elements in RNA) is a new framework that enables the discovery of structural RNA elements by using context-free grammars and mutual information. “The application of TEISER to mammalian datasets is revealing a rich picture of mRNA stability regulation by these elements and the RNA-binding proteins that bind them,” said Dr. Tavazoie.

In a recent genome-wide, systems-level analysis of 46 different cancers, Dr. Tavazoie and colleagues used cancer gene expression datasets to identify known pathways and processes that are perturbed. “We have identified some of the most commonly recurring elements in several cancers, and this allowed us to dig deeper into individual functional categories and pathways, such as those involved in regulating apoptosis or the mitotic cell cycle,” Dr. Tavazoie added.

Quantitative gene expression measurements and pathway analyses on these data also allowed causality between perturbations and the pathways that are changed to be explored. This approach revealed that, as opposed to a “universal” signature for tumor pathways, perturbed pathways are very diverse, and cellular modifications underlying the malignant state are broadly heterogeneous.

Additionally, the systematic analysis of cis-regulatory elements showed that only approximately 25% of the newly discovered motifs corresponded to known binding sites, illustrating the complexity of the malignant state and the limited amount of information that we currently have on describing cellular perturbations that characterize it.

Co-opting Biological Systems

“We developed the ability to synthesize almost any molecule that we would like to, but one great challenge for the field is transitioning from individual molecules to networks that function together inside the living cells,” explained Virginia W. Cornish, Ph.D., professor of chemistry.

As part of early efforts to co-opt directed evolution to build new molecules inside cells, Dr. Cornish and colleagues linked the chemistry of enzyme catalysis to cell survival. “We envisioned that we could do this by including small molecule chemistry into the yeast two-hybrid assay. This involves selecting for enzymes that catalyze the synthesis or the cleavage of a bond based on the transcription of an essential reporter gene and coupling two small molecules,” Dr. Cornish said.

This concept, which initially took advantage of the high-affinity interaction between methotrexate and dihydrofolate reductase, provided the possibility of exploiting orthogonal chemistry in a robust way inside living cells, and is currently being widely used for many applications, such as assays for controlled protein degradation.

The proof of principle for linking enzyme catalysis to reporter genes exploited cephalosporin hydrolysis by the Enterobacter cloacae P99 cephalosporinase, which was incorporated into a three-hybrid system to measure cephalosporinase activity in vivo as a change in lacZ transcription, and this screen also helped isolate wild-type cephalosporinase from a pool of inactive mutants.

The promises offered by yeast genetics emerged as an attractive experimental possibility. “As we started taking advantage of the budding yeast as an organism in which to study directed evolution, we sought ways to perform mutagenesis in this setting,” said Dr. Cornish. While Dr. Cornish and colleagues initially exploited homologous recombination as a mutagenesis strategy, they saw the dependence on transformation emerge as a shortcoming.

“We addressed this by placing the DNA on a heritable cassette plasmid, embedding it in endonuclease sites, and performing endonuclease cleavage to create double-stranded breaks and recruit the homologous recombination machinery, and this helped generate reasonably large libraries,” Dr. Cornish explained.

Synthetic and systems biology promise to profoundly impact the field of biosynthetic engineering. “Increasingly, rather than make drugs by chemical synthesis, we can engineer yeast and other organisms to biosynthesize intermediates on which we can do further chemistry,” he continued.

In a current project, Dr. Cornish and colleagues are using engineered yeast as a low-tech biosensor for cholera. Receptors that recognize cholera were engineered on the surface of yeast, and the yeast were engineered to turn on a plant pigment. “We can, in this way, envision a very cheap and safe product that we can hand out into local communities for the detection of cholera by nontechnical people in the community,” Dr. Cornish added.

Single-Cell Analysis

“When we take a complex tissue and perform expression profiling, we do not necessarily know the cellular origin of the molecules that were quantified,” says assistant professor Peter A. Sims, Ph.D. In cellular populations, biological processes are unlikely to occur in a synchronous fashion, and this opens significant challenges for analyzing those processes and interpreting the data.

One approach to address this shortcoming is to perform analyses at the single-cell level. “We wanted to look at the transcriptomes of individual cells,” explained Dr. Sims. Examining many genes across large numbers of individual cells provides an ideal way to catch a glimpse of biological processes at the single-cell level, and concomitantly capture interindividual heterogeneity in the population.

Ideally, single-cell analysis should be performed by direct detection, to avoid the consequences of amplification bias, and the approach should be time-efficient and affordable. “At this time, we do not have a technology that does not compromise the number of cells analyzed in favor of the number of targets analyzed and vice versa, so it is very difficult to look at a lot of genes and a lot of cells with one tool,” said Dr. Sims.

Investigators in Dr. Sims’ lab rely on two approaches for single-cell analysis, microfluidics and microscopy. Microfluidics uses very small reagent volumes and consequently reduces contamination, whereas microscopy provides additional visual information about the system being examined at a specific time.

By using soft lithography microfluidics, a technology that allows single-cell behavior to be captured under a broad number of conditions, Dr. Sims and colleagues have developed tools for single-cell transcriptome analysis. One of these tools, targeted probe-based expression profiling, offers the possibility of simultaneously exploring several tens of genes within the same cell. “The advantage of this approach is that we are able to probe the transcriptome at the location where the cell is observed,” noted Dr. Sims.

With each cell in its own picoliter-sized chamber, the array can be sealed with a glass surface that is chemically functionalized to allow the capturing of the RNA, which can be reverse transcribed. “We are also developing another tool that uses microarray platforms for large-scale, genome-wide RNA sequencing, where we can generate thousands of cDNA libraries on the microarray well chip,” Dr. Sims said.

Systems biology has made it possible to test hypotheses that merely a few years ago were beyond the reach of experimental approaches. Nonetheless, by working at the juncture of biomedicine, biotechnology, and the clinic, scientists are extending their reach, unveiling mechanistic details, and hastening paradigm shifts. Central to these endeavors, and one of the fundamental teachings that has emerged thus far, is the power of science’s integrative nature.

To the Microbiome…and Beyond!

“We are focusing on bacterial and human systems, and we are beginning to study the interface of the two, the microbiome, from a synthetic biology perspective,” says James J. Collins, Ph.D., professor of biomedical engineering at Boston University. Dr. Collins served as keynote speaker at Columbia’s recent symposium.

Many advances in systems and synthetic biology have become reality thanks to landmark developments, and one of these was the construction of transcriptional switches that co-opted RNA regulatory functions. A significant stride came in 2004, when Dr. Collins and colleagues reported the development of a post-transcriptional Escherichia coli RNA-based regulator that was able to either suppress or activate gene expression.

In this model, a cis-repressive sequence inserted upstream of a ribosome binding site formed a stem-and-loop structure at the 5′-untranslated region of the mRNA, interfering with gene expression. Also, a small noncoding RNA molecule expressed in trans was able to target the repressed RNA, activating gene expression. Subsequently, Dr. Collins and colleagues developed genetic counters. With this approach, several RNA switches are serially activated by output from the previous switch, allowing user-defined induction transcriptional output events to be counted.

These advances opened the need to generate synthetic circuits that can be programmed to also kill a cell that has been rewired with a new function. By using two lytic proteins, Dr. Collins’ lab developed a programmable kill switch. “After three public hearings, the Presidential Commission for the Study of Bioethical Issues highlighted the kill switch as a much-needed safeguard,” said Dr. Collins.

In addition to safety, this system offered other advantages. “After we published this work, several biotech companies showed interest in using this technology for combating corporate espionage,” explained Dr. Collins. They feared that competitors could obtain engineered microbial strains. This concern is alleviated if a microorganism is endowed with the property to self-destruct when programmed to do so.

“What really has been driving my lab is going after antibiotics,” remarked Dr. Collins. The need to develop new antimicrobial agents, or to devise new approaches to address antimicrobial resistance, is fueled by the increasing number of resistant strains, a trend that is compounded by the decreasing number of newly developed and approved antimicrobial agents.

“We collected hundreds of E. coli expression profiles, and used systems biology to develop whole-scale genome models to gain new insights into how antibiotics work,” said Dr. Collins. Experiments on quinolones revealed that the DNA damage response network was a key participant in bacterial response. The oxidative damage response network was also induced, however, and studies on additional antibiotics revealed that they can also alter cellular metabolism and affect cellular respiration as part of their mechanisms to cause bacterial cell death.

“We tried to use these findings to boost the killing efficacy of antibiotics,” added Dr. Collins. In a high-throughput assay that examined over 2,000 compounds, Dr. Collins and colleagues showed that the bactericidal activity of antibiotics was improved with RecA inhibitors. Subsequently, the use of bacteriophages overexpressing LexA3, a noncleavable version of the SOS system repressor LexA, boosted from 100- to 10,000-fold the in vitro killing efficacy of bactericidal antibiotics, and illustrated the possibility of using engineered bacteriophages as antibacterial adjuvants.

Moreover, in the first animal study in systems biology, the antibiotic-enhanced bacteriophage in combination with quinolone treatment rescued 80% of the mice from death. “Much of what is happening in synthetic biology is still at the microbial stage, and we would like to move this field toward higher organisms,” concluded Dr. Collins.


A team led by Boston University’s Dr. James Collins collected numerous E. coli expression profiles and then used systems biology to develop whole-scale genome models to gain novel insights on how antibiotics work. [fusebulb/Fotolia]

Previous articleMale Presence Hastens Demise of Female Nematodes
Next articleThe Best Science Apps: 12 December Picks