There are many different forms of cancer, and each continues to change all of the time. To try to keep up, some researchers are using a variety of approaches to study how a benign cell becomes cancerous — and how to detect it as soon as it does. Others focus on how to stratify patients based on cancer types in order to find better, more tailored treatments for a given patient's particular disease.
Over the past year, much work been done in those and other areas. In these next few pages, we present how researchers are using and refining tools and technologies to better understand, and eventually conquer, cancer in its many forms.
Sequencing: Whole-Genome Sequencing Illuminates Rearrangements
Mark Rubin at Weill Cornell Medical College has been intrigued by the puzzling complexity of prostate cancer for several years. Until now, though, the nature of its complexity remained largely elusive, despite researchers' tireless efforts to unmask it. As with most cancers, the mutations that underlie primary human prostate tumors are convoluted indeed. But as Rubin and his colleagues at the Broad Institute and elsewhere show in a February Nature paper, prostate cancers are complex in a rather unique way.
Using whole-genome sequencing on seven primary human prostate tumors and their paired normal counterparts, Rubin and his colleagues identified 3,866 putative mutations per tumor on average, though a median of only 20 non-synonymous variants within protein-coding genes per sample.
For prostate cancer, "over the years, we've seen there are very few mutations in the genes that produce proteins, unlike other tumors," Rubin says. In its recent whole-genome sequencing study, "our group identified some molecular rearrangements, called gene fusions, suggesting that perhaps prostate cancer is more complex than other cancers in a different way; instead of having spelling errors, they have large rearrangements at the genomic level."
The team found several previously unidentified large-scale genomic rearrangements in the prostate cancers they sequenced. Rubin says that whole-genome sequencing enabled the team's discoveries in a way that no other technique could have. "Had we used exome capture, or had we used standard techniques, we wouldn't have discovered these rearrangements," he says. "This was the first time that we were able to shine a light on the genome and see that, in fact … these rearrangements are not entirely random." More specifically, Rubin and his colleagues hypothesize that early gene fusion events predispose a prostate cancer to a particular tumorigenesis pathway.
"This will be important" in a clinical sense, Rubin says, "because right now we think of the 230,000 prostate cancers that are diagnosed each year as one cancer type." Much like breast cancers are now characterized by several subclasses, Rubin predicts that prostate tumors might one day also be recognized by their appropriate molecular subtypes.
It's possible that similar whole-genome sequencing on tumor-normal matched pairs might illuminate large-scale rearrangements in other cancer types in the future, Rubin says. As cost is no longer such a formidable barrier, cancer sequencing efforts such as his group's, "as well as other TCGA [The Cancer Genome Atlas]-type projects are going to provide the really detailed information that is going to allow researchers in the next years to make important discoveries, similar to how the first [human] genome has allowed researchers of the world to make enormous progress," Rubin says. "I think it's going to be an exciting time to see how this information will help accelerate the identification of new treatments for prostate cancer."
Genetic architecture: Pediatric Cancers Have Fewer Driver Mutations than Adult Cancers
Pediatric cancers aren't quite the same as adult cancers, researchers have found. After studying the genetic alterations in four adult cancers, and learning a lot about those diseases, Johns Hopkins University's Victor Velculescu and his colleagues turned their attention to a pediatric cancer. "Something had to be done in order to get at the intricacies of pediatric cancer," Velculescu says. "We decided to tackle medulloblastoma, a very important pediatric brain cancer." Medulloblastoma, a cancer of the cerebellum, affects one in 200,000 children under the age of 15.
Velculescu and his team sequenced the exomes and miRNA-encoding regions of 22 pediatric medulloblastomas — 17 from primary tumors, four from xenograft mice, and one from a cell line — and generated 735 megabases of sequence data. In addition, the researchers used high-density microarrays to identify copy number alterations present in the samples. As they reported in Science in January, researchers compared the mutations seen in this pediatric cancer to the ones present in the adult cancers they had previously studied. Each medulloblastoma tumor only had, on average, 11 genetic alterations — 5- to 10-fold fewer than the adult cancer patients had. "That was one of the striking things," Velculescu says. In adult cancers, most mutations are thought to be passenger mutations that accumulate as the cell divides. The researchers do note that in medulloblastoma, as the patient's age increased, so did the number of mutations. "This is pointing to a lot fewer pathway mechanisms that are altered in pediatric cancer research," he adds.
Velculescu says that this finding suggests that pediatric cancers have fewer broken genetic parts that drive the formation of cancer, and not just in medulloblastoma. "My expectation is this will be generalizable feature of pediatric cancers," he says. Because there are fewer parts to fix, it might easier for clinicians to intervene.
While analyzing the data, the researchers noticed that there were pathways involved in medulloblastoma that had not been associated with the disease before. Of particular interest were MLL2 and MLL3, genes involved in chromatin remodeling and transcriptional regulation, suggesting an epigenetic role in medulloblastoma formation. Genes controlling histone methylation have been implicated in other cancers, but how MLL genes affect tumor formation is as yet unknown. "This theme of mutation in cancer in genes involved in epigenetics looks to be promising," Velculescu adds.
Research efforts over the next five years or so will likely focus on determining which pathways — particularly developmental and epigenetic pathways — are involved in tumorigenesis. "It's like a speeding train that has a number of things wrong with it," he says. Knowing which pathways are involved will then make it easier to find therapeutics to stop the train.
Small RNAs: Small, Conditional RNAs Attack Tumors
Conventional chemotherapies systemically disrupt normal cell division, causing side effects that limit dosage and abate treatment efficacy. In an effort to work around this difficulty, a group of researchers are exploring the possibility of using small, conditional RNAs to selectively attack cancer cells.
"This has been a problem for 50 years, so we decided to start from scratch and engineer chemotherapies that turn themselves on only in cancer cells. The purpose of that was to try to eliminate side effects to make the killing selective," says Niles Pierce, a professor at the California Institute of Technology. "We've shown with human brain, prostate, and bone cancers in cultured cells that we can selectively kill each of those cancers with appropriately designed small, conditional RNAs. With a single dose, we can obtain 20- to 100-fold killing of cancer cells containing the target mutation, with no measurable effect on the other cancers."
These 36 nucleotide-long RNAs can be "programmed" to selectively target a cancer mutation of interest. The small, conditional RNAs can then change shape to mimic viral RNA, thereby effectively tricking the cancer cell into self-destructing through an anti-viral immune response.
The next step for Pierce is to show that he can achieve similar efficacy and selectivity using mouse models of cancer. "Our hope in both the mouse studies and potential future human trials is to leverage delivery technologies for small RNAs that have already been developed to support studies with therapeutic RNA interference," he says. "The reason that therapeutic RNAi is conceptually exciting, and has so much potential, is that it shares the property with our approach of being programmable, but the crucial difference is that therapeutic RNAi is not conditional. It shares the same selectivity problems with conventional chemotherapy because in designing a small RNA for the [RNAi] approach, you're effectively designing both a diagnosis event and a treatment event."
To this end, Pierce is looking to collaborate with delivery companies that have already had some success with RNAi. "Our molecules are chemically very similar, although they're smaller than RNAis, so that's considered to be beneficial. We're hoping that ... formulas that have been validated for RNAi use will be equally suitable for our purposes for mouse studies," he says. "The crucial distinction is that, with our approach, diagnosis occurs when you detect a cancer mutation and that detection event has nothing to do with the drug, it merely turns it on. The shape change leads to this long, double-stranded polymer which mimics viral RNA, which is solely responsible for the efficacy of the drug. You're separating out the diagnosis event from the treatment event in a way that a conventional drug cannot."
ArrayCGH: Algorithms Define Subtypes of Invasive Breast Tumors
Cancer genomes go through all sorts of rearrangements and distortions until they sometimes barely resemble a proper, functioning genome. Oslo University Hospital's Anne-Lise Børresen-Dale and her colleagues had previously found that invasive breast tumors fell into three categories based on their arrayCGH profiles: simplex, complex I sawtooth, and complex I firestorm. This past year, Børresen-Dale and her team developed platform-independent algorithms to separate breast cancers further, into groups based on their genomic architecture. "We needed an algorithm that could more objectively score and what we wanted to do was to score in a more refined way. The first part is to see how often does it break and what is the amplitude of the amplification in the breakpoints. We put into the same algorithm and then make this look more like a phenotype and sub-classify them more," Børresen-Dale says.
There are two main ways in which the genomic architecture is rearranged in a breast tumor: regionally and locally. Regional changes include swaps, gains, or losses of whole arms from a chromosome, while local changes more break points. Drawing on work done by Cold Spring Harbor Laboratory's Michael Wigler, the Børresen-Dale team fine-tuned algorithms to better detect these kinds of changes.
With these two algorithms, the researchers uncovered eight subtypes of breast tumor. The first stratification is based on whole-arm changes seen in the samples. Tumors in group A have a gain of 1q and/or loss of 16q, group B have a loss on 5q and/or a gain on 10p while group AB has both, and group C has neither. Those four groups are then broken down further by whether the tumor has more complex rearrangements on top of those. "Every time that we had this complex arrangement, on top of the translocation, it gave a worse prognosis in all of these four groups," Børresen-Dale says. "And this also holds up in multivariate analysis."
Indeed, she adds, in the multivariate analysis the outcomes of these algorithms "came out as prognostic markers, independent of node stages and independent of grade."
Of interest, Børresen-Dale says, is that triple-negative breast cancers can be found in each of the arrayCGH subgroups. "We think this is very important and they have clinical implications," she says. A triple-negative tumor that falls into one of these classes could be more susceptible to certain therapies. Her group is considering a trial to determine whether that is true. "This is a phenotype that if you have this type of pattern, there is probably something wrong with repair and they may be prone to being more sensitive to being PARP inhibitors," she adds.
Molecular typing of some cancers has been around for a few years now, but Børresen-Dale says that getting typing into the clinic has been difficult. Part of the problem is that there are no good treatments for non-responders. "We can identify the worse prognosis, but we really don't have good treatment for those — yet. I think that we need more targets," she says. "For example, for the triple negative, we really don't have good targets yet. We are able to find the non-responders, but what do you do with the non-responders?"
Proteomics: Researchers Identify Cancer Biomarkers Through Mass Spectrometry
When it comes to cancer research, working to discover treatments goes hand in hand with finding new biomarkers for diagnosis and prognosis. At York University in Toronto, Leroi DeSouza and his colleagues are using a proteomics approach to find protein biomarkers for various cancer types. "If you look at samples from any kind of cancer and compare that with normal tissue from the same organ, then any protein that is consistently differentially expressed in the cancer samples relative to the normal could be indicative of being a cancer marker," DeSouza says.
In 2007, DeSouza and his team published paper in Molecular Cellular Proteomics detailing a proteomics- and mass spectrometry-based approach to find biomarkers for endometrial cancer. The team analyzed 40 endometrial samples — 20 cancerous and 20 normal — and found several proteins that were differentially expressed. More recently, the group published a follow-up paper in PLoS One, in which it refined its technique to work with much smaller samples than those from their original study — 100 micrograms of protein per sample, rather than 200 — and found several additional novel protein markers, DeSouza says.
Their method involves homogenizing a sample of cancer tissue in a buffer, digesting it with trypsin, and labeling both the cancer- and normal-tissue samples from the same organ with mass tags — DeSouza and his team use Applied Biosystems' iTRAQ labels. Once the samples are run through the mass spec, any difference in the peptide abundance between the the sample types is considered "significant," DeSouza say, and is then verified through western blot analysis.
This particular study involved endometrial cancer protein markers, but the team has also been doing research on head and neck cancer — specifically oral squamous cell carcinoma — renal cancer, and glioblastoma multiforme, and this approach can be applied to all these types of cancer, or any other for which researchers can obtain high-quality tissue samples.
"We hesitate to say we've found targets for treatments as of yet," DeSouza says. "At this point it's got more to do with trying to diagnose the cancer. The diagnostic markers could evolve into drug targets eventually, but at the first stage we just want to be able to diagnose it earlier." The team has also studied the proteins to see if they interact with any of the known "usual suspect" cancer genes, like p53, he adds. Some do, which might indicate a pathway involved in the formation of the cancer, and might also lead to therapeutics down the line. And determining when patients' tumors express certain proteins could lead to better prognostic tools for clinicians, he adds, as the proteins could be used as a biomarker for survival.
The next step is to take this approach and apply it to the development of a simple blood test that could check for the presence of differentially expressedproteins. This information would make it easier for clinicians to determine a diagnosis and prognosis for various kinds of cancer, DeSouza says.
Sample prep: Comprehensive Gene Expression Profiling of Partially Degraded RNA
Aware that most clinical oncology samples are fixed in formalin and embedded in paraffin, researchers at the Netherlands Cancer Institute in Amsterdam set out to determine the extent to which gene expression profiles obtained from FFPE-derived RNA are comparable to those from fresh-frozen matched tissue-derived RNA.
"For selecting an untreated control, researchers depend indeed on trials performed in the past when biopsies were mostly only ... FFPE," says Lorenza Mittempergher, a postdoc in Laura van 't Veer's lab. "Our study demonstrated that gene expression analysis on FFPE material is possible, and that it opens up the possibility of using archived FFPE tissue in future cancer research."
By comparing gene expression signatures derived from 20 fresh-frozen breast cancer tissues and those from 20 matched FFPE samples on Illumina's Whole Genome DASL platform, members of the van 't Veer lab show in a Feburuary PLoS One paper that once properly normalized, the data generated from FFPE material is largely comparable to that obtained from their fresh-frozen counterparts. "The DASL assay allows partially degraded RNA to be used, as is the case for RNA extracted from paraffin," Mittempergher says, adding that the team validated its DASL results against those derived using Agilent arrays and standard immunohistochemical measurements. Her colleague and fellow postdoc Jorma de Ronde adds that the team's normalization procedure was a key step.
To make up for the systematic biases of each sample type, the team first subtracted the median expression of each gene they interrogated from a 70-gene signature in the fresh-frozen and FFPE groups separately. "In this way, gene expression values are always centered around zero," de Ronde says. "So, if a gene would always yield a higher signal in the FF samples compared with the FFPE samples, this procedure would account for that."
Post-normalization, the team found that FF and FFPE sample pairs from the same tumor clustered together. "The 70-gene prognostic signature … could, with a reduced gene set of 60, be derived from the FFPE material," de Ronde says. In the future, he adds, the team's study could enable researchers to further investigate the biological biases of each sample type separately. Though "differences between the two sample methods will remain," de Ronde says, they are "largely offset by the fact that FFPE enables transcriptional analysis of far more tissue than would be possible with only FF material."
As such, Mittempergher and de Ronde are hopeful that their results will help cancer researchers mine archived FFPE samples for all that they're worth. "Enabling transcriptional profiling of this [FFPE] material should make large-scale studies possible that were previously unfeasible," de Ronde says, though he adds, "whether it will yield the possibility to be used in a robust diagnostic setting remains to be seen."
Bioinformatics: A Novel Multi-Cancer Computational Approach
After noticing a particularly strong signature in a subset of high-stage ovarian cancer samples containing more than a hundred highly overexpressed genes that are not often overexpressed in other cancers, Columbia University's Dimitris Anastassiou began working on a multi-cancer phenomenon hypothesis. Last year, Anastassiou decided to test his hypothesis that genes that are occasionally highly overexpressed in the presence of a particular phenotype, but otherwise remain unaffected. Using a novel computational technique, he analyzed data from multiple cancers to identify sets of genes whose coordinated overexpression indicate the presence of a high-stage cancer.
"Remarkably, the hypothesis was fully validated. All cancer types we considered, except leukemia and glioblastoma, had this signature in a subset of samples that had exceeded a particular staging threshold, but not otherwise," Anastassiou says. "We also carefully analyzed the signature doing multi-cancer analysis and we found that the most prominent genes of the core signature are COL11A1, INHBA and THBS2. It turns that other computational techniques would also be sufficient to find the signature, except that they would not identify the genes from the beginning as precisely as this custom-based bioinformatics tool."
The signature is largely myofibroblastic, indicative of reactive stroma, and it also prominently contains some epithelial-mesenchymal transition orchestrating transcription factors, such as Slug and Twist. But it also differs in several ways from what is assumed to be typical — the EMT transcription factor Snail is not part of the signature; it appears to be methylated.
Anastassiou says it should be relatively easy to test and validate the findings of his study, which was published in BMC Medical Genomics in November, using a very simple test. "Take any rich gene expression data set from any type of solid cancer and just identify the genes whose expression is most highly correlated with the expression of the collagen COL11A1," he says. "You will find that THBS2 and INHBA are always close to the top of the list, together with some other collagens. This happens because of the presence of the samples that have the signature, but you will not find this property in a data set of noncancerous samples because the ranked list of genes correlated with COL11A1 will be random looking."
What is exciting for Anastassiou and his colleagues is the possibility of developing therapeutics to inhibit cancer invasion and subsequent metastasis by targeting the underlying biological mechanism. "I am currently collaborating with medical researchers trying to decipher the mechanism using both in vivo and in vitro experiments. I believe that it may be a resurrection of pathways from early embryonic development, which is now used to implement the invasive stage of cancer," he says. "As more and more biological data, including methylation and miRNAs, become publicly available from sources such as The Cancer Genome Atlas, we will have the opportunity of using computational techniques to shed additional light on the underlying mechanism."
Pharmacogenomics: Cancer Treatments, One Patient at a Time
When it comes to tailoring cancer treatments to individual patients, or discovering new therapies for various cancer types, the field of pharmacogenomics plays a leading role. "The idea is to tailor an individual's genetic profile and base their therapies to those genomic factors to maximize benefits and minimize harms of treatment," says the National Cancer Institute's Andrew Freedman.
Freedman, who works in the NCI's extramural division that funds and manages grants, has been leading a trans-NCI effort to advance the use of pharmacogenomics in cancer research. He and several colleagues published a commentary in the Journal of the National Cancer Institute in October detailing their recommendations on setting a research agenda to accelerate the translation of cancer treatments and diagnostics. "The hope is that pharmacogenomics and personalized medicine can go hand in hand to personalize treatments for patients, and that we have more predictive medicine, to pick either the right dosage, treatment combination, or treatment intensity to maximize the benefit ... from any treatment regimen," Freedman says.
Indeed, there are many success stories where pharmacogenomics was used to benefit patients. When the Food and Drug Administration approved a drug called cituximab — sold commercially as Erbitux — for the treatment of metastatic colorectal cancer, it was noticed that some patients responded to the treatment while others did not. Using pharmacogenomics, the researchers were able to determine that patients with a particular KRAF mutation didn't respond to the drug. The KRAF mutation had been discovered years before, but the interaction of the drug with that mutation pathway wasn't seen until researchers went looking for answers, Freedman says. Such cases save both clinicians and patients time and effort in determining a proper course of treatment.
Although pharmacogenomics is useful for many diseases, it's particularly effective in the study of cancer, he adds. While germline DNA can be used to study any disease, taking samples from diseased organs can be problematic. In cancer, both germline DNA and tumor tissue are available for researchers to use in tailoring therapies.
In their paper, Freedman and his colleagues make several recommendations. Chief among them is that groups like NCI and other institutions need to develop and support opportunities for researchers to use pharmacogenomics in the lab and in clinical trials, support the use of pharmacogenomics in a clinical setting, support the development of transdisciplinary training programs in cancer pharmacogenomics, and study the ethical, social, and data-sharing implications of collecting specimens for pharmacogenomics research.
When it comes to cancer research, pharmacogenomics is a tool every scientist should know how to use, Freedman says. If they're looking to translate their work from the lab to the clinic, it is particularly useful, he adds.