There are many cancer types, but there are also a number of ways to tackle the problem of stopping the disease in its tracks. Some researchers are applying the latest cutting-edge tools, like those of synthetic biology, to advance knowledge on all of cancer's different incarnations, while others are busy working on next year's best new tool or method, like better labeling approaches. Others still are puzzling out how to move all this new knowledge into the clinic
In these next few pages, Genome Technology rounds up some — though certainly not all — of the advances in cancer research from the previous year.
Sequencing: Researchers Create Proof of Concept for Clinical Integrative High-Throughput Sequencing
The push toward personalized medicine is leading many researchers to work on ways of getting next-generation sequencing into the clinic. In cancer therapy, especially, targeted treatments for specific disease subtypes or gene mutations are being seen as the next logical step.
With that idea in mind, researchers at the University of Michigan published a pilot study in Science Translational Medicine in November describing their efforts to use integrative high-throughput sequencing for the treatment of patients in the clinic. The team enrolled two patients with advanced or refractory cancers, and performed whole-genome sequencing and transcriptome sequencing of their tumors and whole-exome sequencing of their tumors and normal DNA. "This was a pilot feasibility assessment to ask: 'Can we get this clinical protocol in place? Can we get it through the IRB? Can we return the results in a clinically relevant time frame? Can we handle the bioethical issues involved as well as what type of data would we generate from individual patients in terms of looking at their whole exome, transcriptomes, and low-pass genome?'" says the study's senior author Arul Chinnaiyan. "We call this approach integrative sequencing."
Using this method, the team was able to pinpoint that the patient with metastatic colorectal cancer had somatic point mutations in NRAS, TP53, AURKA, FAS, and MYH11, and amplification and over-expression of CDK8. In addition, the researchers found that the patient with malignant melanoma had a somatic point mutation in HRAS and a structural rearrangement affecting -CDKN2C. Based on these results, the researchers were able to recommend clinical trials in which the patients could enroll. These results were all available within a three- to four-week time frame that would be considered clinically actionable.
There are still challenges to getting high-throughput sequencing into the clinic, like cost and dealing with reams of data, Chinnaiyan says. Despite the problems, however, he says the use of such sequencing methods will lead to personalized oncology. "You end up getting a more comprehensive picture of the tumor landscape by carrying out this approach," Chinnaiyan says. "I think it'll take time to implement these types of approaches ... but we can do it with existing technologies and it'll just be a matter of time that these technologies will be improved and will become more commonplace."
In the end, he adds, comprehensive sequencing approaches will bring more value to patients than more focused approaches — integrative high-throughput sequencing rather than assays for individual genes or mutations. "At the end of the day you're going to want to know comprehensively what distinguishes your tumor from your normal DNA," Chinnaiyan says. "And, if the cost and turnaround time is reasonable, it seems that it would be the [integrative sequencing] approach that would win out over time rather than looking at a focused component of the genome, or a series of focused assays."
Synthetic biology: For Optimal Results in Fighting Cancer, Researchers Build Their Own Weapons
The advantage to building cancer-fighting tools from scratch is that researchers can innovate and create drug-delivery systems or apoptotic triggers that do not necessarily exist in nature. In addition, researchers can aim to deliver treatments more directly to cancer cells, sparing healthy cells.
To do so, researchers are using synthetic biology approaches to manipulate DNA. In February, Harvard Medical School's George Church and his colleagues published a study in Science describing DNA nanostructures that they built for the targeted delivery of drugs to cancer cells. "You take DNA — which is normally linear — and by choosing the sequences, you create branches, and, once you have branches, you can start building two- and three-dimensional objects rather than one-dimensional DNA. And then you can start decorating it with other molecules," Church says. "It's one of the first hybrid structures where you have not only structural DNA, but also aptamers and antibodies."
Much like picking the options for a car, Church and his team chose specific aptamers to add to the DNA nanobots that respond to molecules expressed by acute myeloid leukemia cells. The researchers also designed the nanobots in the shape of a clamshell so that they could hold a payload of cancer drugs. After releasing the bots into a mixture of healthy and leukemic blood cells, the team found that they destroyed most of the cancer cells while generally sparing the normal ones. "In principle, the delivery [of drugs] could be entirely cell surface-mediated, so you could have signaling molecules that stay on the surface," Church says. "Inevitably, they will be brought into the cell and degraded, or degraded on the outside of the cell, or eventually the cell itself will turn over. But since all the components are essentially natural proteins and nucleic acids, they get recycled just like anything else."
It's not just drug delivery systems that researchers are capable of building. A team from MIT and the Swiss Federal Institute of Technology in Zurich published a separate study in Science in September reporting a synthetic cell circuit that, when inserted into a cancer cell, can trigger apoptosis. "Our [gene circuit] systems are encoded with DNA — five genes to be precise — and what we have to do is deliver these five genes to the cell," says ETH's Yaakov Benenson. "This can be done using chemical reagents or electrical voltage. Once they're in the cell, each of these five genes have their own task — essentially like a little pathway which is engineered. The goal of the pathway is to activate the apoptotic gene hBax, but only when certain conditions occur in the specific cell."
Like a computer program designed to perform a certain action under specific conditions, the gene circuit's switches react individually when certain microRNAs are detected in the cell. Once all five switches are activated, the circuit is complete, and it induces apoptosis in the cell. "We sample from multiple indicators and we want all of them to be present at the same time and this should improve the specificity and selectivity of the process," Benenson says.
Church and Benenson both add that their approaches still need improvement, but may eventually be used in a variety of ways. Church's clamshell nanobot could be programmed to deliver drugs to different kinds of cancers, or could even be used preventatively. The trick is to have a smart piece of DNA-based technology that uses logic in its programming, he adds. With that, the synthetic bots and circuits could be made to fight cancer in any number of ways. Benenson's gene circuit, while currently designed to trigger apoptosis in cancer cells, could be customized to detect pre-tumor cells with a specific signature and cause them to commit suicide before they develop into a tumor.
Gene expression: Signatures Predicting Cancer Outcome May Be Influenced By Confounding Factors
Confounding factors like proliferation or differentiation may be behind a number of gene expression signatures' association with cancer outcomes, says Vincent Detours, a researcher at the Free University of Brussels. "Many genes are co-expressed together, and I think many people are being misled about the nature of the signal that dominates the gene expression data," he adds.
Detours was studying thyroid cancer — trying to determine whether there was a difference in gene expression between papillary thyroid cancer, a treatable disease, and anaplastic thyroid cancer, an aggressive disease with a poor prognosis — when he came across a surprising effect.
Based on this work, he thought that the p53 pathway may be involved in the differences between the two cancer types. However, the genes that were active when p53 was acting as a tumor suppressor were more active in anaplastic thyroid cancer. "Which is kind of odd because this kind of cancer is supposed to be extremely aggressive," Detours says. He followed up on other possible tumor suppressor mechanisms, like senescence, and got the same results.
Then he looked at proliferation. "I saw that the expression was very comparable to the senescence signature. I thought that what I have been observing is that both senescence signatures and p53 signature is connected with proliferation signatures," he says. "Anaplastic cancer, the most aggressive cancer, is proliferating much faster than the less aggressive cancers. So I thought that all the signatures — although they are different — are all proliferation."
Now, he adds, "the question is not to come up with a new signature. It is to come up with a signature that is independent of proliferation."
To do that, he and some of his colleagues turned to breast cancer signatures, and developed a proliferation index from gene expression data, which they reported in PLoS Computational Biology in October. "If the signature is still prognostic after you remove the proliferation signal, then it is independent of proliferation," Detours says, adding that in doing this, the team found that many breast cancer signatures depend on proliferation signatures to predict outcome.
In follow-up work that was published online in Oncogenes in January, Detours and his colleagues reported a similar effect for differentiation signals, and they developed a differentiation index from gene expression data and applied it to thyroid cancer.
Detours adds that while such gene expression signatures might not reveal biologically important information, that does not mean they are not useful in the clinic. "If there is a fire in the forest, you have some smoke," he says. "The smoke doesn't cause the fire, but it is a very useful marker of fire."
Pharmacogenomics: Genetic Variation and Response to Anticancer Therapies
Not every cancer patient who undergoes chemotherapy responds to it, though nearly all experience the side effects. Members of the Pharmacogenomics of Anti-cancer Agents Research group at the University of Chicago are working to understand how genetic variation influences patients' response to a variety of chemotherapeutic agents, to help physicians tailor patient treatment accordingly.
"Ultimately, what we're trying to do is identify patients at risk for severe toxicity associated with chemotherapy so they don't have to endure serious side effects, and also not respond," says Chicago's Eileen Dolan. "Our main objective is to identify genetic variants associated with sensitivity to chemotherapy, but we want to also understand the function of those to determine whether they are also associated with gene expression levels, protein levels, microRNA, or some other intermediate function within the cell."
In cell-based experiments, Dolan and her colleagues identified SNPs and copy-number variants associated with sensitivity to anticancer therapeutics. They deposited all data generated by these studies into the Pharmacogenetics and Cell database, or PACdb, a results resource they launched in 2010 that is now tied to both the PharmGKB and SCAN databases.
PACdb "allows you to evaluate genes, in terms of expression, that are correlated with drug sensitivity," Dolan says. Chicago's Eric Gamazon adds that the database can be queried for variants of interest, among other things.
"We've also had other groups participate by enabling us to make their results available through PACdb," Gamazon says. Going forward, the "plan is to expand the database to include studies of methylation, studies of the proteome, and [of] other molecular phenotypes," he adds.
Indeed, Dolan says her group has begun to interrogate miRNAs and proteins associated with chemotherapeutic sensitivity. "One of the more exciting projects we've embarked on is looking at baseline transcription factor levels [and] signaling proteins using a micro-western array," she says. "We've been able to look at about 400 different proteins and evaluate protein relationships, both with mRNA and with drug sensitivity, and with genetic variants."
With multiple datasets, Dolan and her colleagues can now evaluate how genetic variation affects expression and protein levels, among other things, and how that plays into drug sensitivity. "It's interesting because protein and gene expression don't always agree," Dolan says. "I think the protein is interesting because it's a little bit closer to what might be affecting [sensitivity], what might be good targets for new drugs."
Epigenetics: Fighting Cancer With Dietary 'Omics Data
It might not be news that eating fruits and vegetables is good for one's health, but researchers have found that sulforaphane — a compound found in cruciferous vegetables such as broccoli and kale — helps to prevent cancer through epigenetic functions.
A study published in Clinical Epigenetics by Oregon State University's Emily Ho found that DNA methylation and the inhibition of histone deacetylases — which can both be influenced by sulforaphane — work together to maintain normal cell function. When this balance is disrupted, the process of cancer development is initiated.
"For a long time, we've been interested in the health benefits of cruciferous vegetables — specifically how sulforaphane may act to help prevent cancer," Ho says. "What we found in this paper was that the compound sulforaphane may act through a novel epigenetic mechanism to help stop the growth of cancer cells, specifically the effect of DNA methylation, which is a relatively new finding."
Ho's study explored the effects of sulforaphane on DNA methylation of the cyclin D2 promoter gene and how changes in promoter methylation affect cyclin D2 expression in prostate cancer. She and her team found that sulforaphane decreases the expression of DNA methyltransferases, especially DNMT1 and DNMT3b, as well as methylation in the promoter regions of cyclin D2 containing c-Myc and multiple Sp1 binding sites. The results provided insight into how sulforaphane epigenetically modulates cyclin D2 expression and may regulate gene expression as a cancer chemopreventive agent.
"Cancer is classically thought to be a genetic disease where mutations or carcinogens damage DNA, kick-starting the cancer process," she says. "But now we know that epigenetic dysregulation — things like DNA methylation, histone changes, or changes in chromatin — can also be dysregulated and kick-start the cancer process."
Ho's current research is confined to cell culture study, so the next challenge is to see how these findings translate into an in vivo model and then in humans. One of the key pieces of data needed is how exactly vegetables like broccoli and kale are exerting effects in the human body. "With sulforaphane, we don't know a ton in terms of just basic questions, such as: We eat broccoli, and how much gets to our prostate? So there's a lot of research that needs to be done to really move it into real health application," she says. "But at the same time, it's not rocket science that eating more fruits and veggies is good thing to do. We're just getting more scientific data that really supports that."
Copy-number variation: Researchers Assess Computational Approaches to Analyze Cancer Genomes
Cancer genomes are marked by instability and, invariably, they contain a number of regions that are duplicated or deleted. Many computational methods to study such structural variants are developed with one cancer type in mind and often just for a single study — there is no one way to analyze cancer genomes, says Sampsa Hautaniemi at the University of Helsinki. "It has been a fragmented field, but still there are many good, strong computational methods for comparison," he says.
Hautaniemi and his team compared different software analysis packages using simulated data. The software packages he and his team looked at ranged from using simple averages to complex Bayesian models. The only criteria for inclusion in the team's study were that the software package had to be freely available and that it could be used on arrayCGH data.
The team evaluated each of the platforms' sensitivity and specificity using a simulated large dataset. "You have to have simulated data in order to say that this [finding] has nothing to do with copy-number variation," Hautaniemi says. "If you just test with tumor samples, they are so heterogeneous." That, he adds, makes it difficult to test specificity.
As they reported in an advance online Nature Methods article in February, most of the algorithms analyzed had good sensitivity, but a range of specificity. For the simulated large dataset of 100 samples, the most sensitive methods were, in order, statistical integration of microarrays, followed by double-layered mixture model and similarity-constrained probabilistic canonical correlation analysis.
However, the most specific methods for that same dataset were Pearson's correlation coefficient, sparse partial least squares regression, and S2N, which had 100 percent, 97 percent, and 94 percent specificity, respectively. Other methods, including the fairly sensitive double-layered mixture model, had much lower specificity. "It was surprising to us to that the specificity was so poor in some of the methods," Hautaniemi says.
He and his team also examined how the algorithms performed on real head and neck squamous cell carcinoma and lung squamous cell carcinoma data. "You can't say that this is a universally best algorithm. That you just can't do. It's also all this context, so we tried to make that context so that it is close as possible to reality to how researchers use these data," he says. For both cancer types, SIM had the highest sensitivity level.
That said, Hautaniemi adds that researchers need to be comfortable with the analysis software they choose and understand its parameters. "Perhaps the method that the researcher is comfortable with and belongs to the upper half of the best-performing algorithms, I think is the best," he says.
Bioinformatics: Assisting RNA Analysis for Non-Experts
To more efficiently analyze genomic data used to study cancer, powerful and efficient analysis pipelines are needed. While there are a slew of algorithms that support various forms of RNA-sequence data analysis, they are rarely coupled together in a way that would be inviting to the non-bioinformatics expert.
To address this limitation and allow cancer researchers and clinicians to compare RNA expression profiles of normal cells and tumors, researchers at the Georgia Institute of Technology developed an RNA-seq analysis pipeline called R-SAP that can analyze 100 million reads in roughly 90 seconds. The performance of the open source R-SAP can scale to multiple computers run simultaneously to further increase performance.
"Right now there are lots of individual tools for identifying variations of splicing patterns, biomarkers, and chimeric RNAs, but someone like me would find them difficult to use — they're different and separate," says Georgia Tech's John McDonald. "So what we're interested in is coming up with a new package or pipeline that would bring all of these functions together into one place and have it be user friendly so that the working biologist with RNA-seq data can have a tool to generate the kind of output information that we would like."
In January, McDonald and his colleagues published a paper in Nucleic Acids Research that described how R-SAP analyzed RNA-seq data from the Microarray Quality Control Consortium in about half the time it took Trans-ABySS, another RNA-seq analysis software tool. "Our algorithm stands up well performance-wise, so we think that's good, especially in the sense that we're bringing it together all in once package," McDonald says.
R-SAP uses a hierarchical decision-making approach to group transcripts into classes, to create gene-expression level data as well as data on biomarkers, chimeric RNAs, and splice variants. The software aligns input RNA sequences to reference genomes and searches for points that do not match up, which could indicate new isoforms, or fragmented alignments that indicate chimeric transcripts like fusion genes.
"The RNA-seq data allows you to quantitate the entire transcriptome, and we have been focusing on that as cancer biologists, but there's also converging evidence that things like alternative splicing are important parameters and that these change significantly in cancer versus normal," he says. "So I think RNA-seq data allows you to access all of that if you have the appropriate algorithms to do it."
Proteomics: Saturation Labeling Technique for Biomarker Detection
In some cancers, access to high-quality samples is scarce. For Gereon Poschmann and his colleagues working to detect protein biomarkers for cancer in Heinrich-Heine University's molecular proteomics lab, "our main concern was that we were very limited in the amount of material available," he says.
Finding that standard labeling approaches were limiting in terms of both time and effort, the group turned to protein saturation labeling using fluorescent dyes and 2D electrophoresis.
With this approach, Poschmann says "the sensitivity is much better." So, too, are the starting material requirements. Standard minimal-labeling techniques require about 50 micrograms of sample, while the saturation labeling approach only calls for three to five.
In a Methods in Molecular Biology chapter, Poschmann and his colleagues present a study on candidate biomarkers for lung cancer, in which they outline a complete protocol for differential proteome analysis on scarce sample amounts — from experimental design, to optimization of saturation labeling, to 2D electrophoresis, to protein identification and validation.
Poschmann says his team has also run similar experiments on pancreatic cancer samples, and has used this saturation labeling approach to monitor tumor progression in that disease.
Despite its low sample requirements and enhanced sensitivity, the saturation labeling technique does take more time to set up initially than a standard minimal labeling approach does. "To apply saturation labeling, first of all, you have to perform experiments to adapt the method to your sample," he says. "The second step is you have to adjust to different [target] types by doing the same experiments."
Poschmann and his colleagues are now trying to a use a modified version of this saturation labeling technique, called redox DIGE, "to investigate processes which might lead to chemotherapy resistance [in] cells," he says. "Our hypothesis is that certain proteins might influence the detoxification of reactive oxygen species produced by chemotherapeutics. As cysteine residues are labeled with this technique, we are able to monitor, in a certain setup, the redox state of these residues."