This article has been corrected to reflect the proper spelling of Laura MacConaill. In the first published version of this article, her name was was incorrectly spelled MacConnaill. Genome Technology regrets the error.
Cancer is ever growing and evading the body's defenses, but cancer research isn't so different. It, too, is always moving and changing tack. Over the past 12 months, researchers have found new subtypes of glioblastoma, provided a link between epidemiological evidence and genetic risk, and more. To do all that, they have used a variety of tools, from the old standbys of gene expression analysis and sequencing to the relative newcomers to cancer research, such as bioinformatics and metabolomic imaging.
It's a golden age of cancer research, says the National Cancer Institute's Stephen Chanock. There's been a lot of investment into the field and finally all these disparate technologies are being married together. "These are important discovery tools where the coming together of population-based science and epidemiology, together with the genomic technologies of GWAS and SNP scans, and now starting with some sequencing and exome sequencing," he says, "has enabled us to identify many new regions that biologically are quite enigmatic or require a tremendous amount of work to understand the how and why that soft place in the genome may be important for developing prostate cancer or breast cancer or lung cancer or pediatric neuroblastoma cancer."
Gene Expression: TCGA Identifies Subtypes of GMB
The prognosis for glioblastoma multiforme, the most common malignant brain cancer in adults, is rarely promising. Treatment is usually comprised of palliative therapies; in most cases, by the time patients are diagnosed, they have only a few months to live. But a research team from the Cancer Genome Atlas may have made a dent in this deadly cancer with recent findings that indicate GBM is not actually a single disease, but rather four distinct molecular subtypes. Each subtype responds differently to aggressive chemotherapy and radiation, as demonstrated by the fact that patients with one subtype succumb at a rate roughly 50 percent slower than patients treated with less aggressive therapy.
"This study starts by following up on observations [made by] a number of investigators, that gene expression arrays suggest that there are subtypes of glioblastoma," says David Neil Hayes, an assistant professor at the University of North Carolina at Chapel Hill. "What we've done for the first time is shown convincingly not just that there are subtypes, but that the subtypes relate to underlying genomic alterations." Hayes and his colleagues were able to conduct this study because of previously unavailable access to a large and high-quality data set that included not just gene expression arrays, but also copy number data and mutational status on a relevant set of genes, as well as clinical data.
"I think we're going to continue to see exciting and high impact papers coming from the data that's already been generated, but even more exciting going forward is the next tumor, ovarian cancer, followed thereafter by a barrage of tumors over the next year with the ARRA grants," he says.
Genotyping: High-Throughput Oncogene Mutation Profiling With OncoMap
Laura MacConaill sees systematic tumor mutation profiling as a "great first step" in personalized cancer medicine. MacConaill, associate director of the Center for Cancer Genome Discovery at the Dana-Farber Cancer Institute, and her colleagues have recently applied their genotyping platform, OncoMap, for the first time on a large scale.
OncoMap is a mass spectrometry-based genotyping platform for detecting somatic mutations in cancer. MacConnaill and colleagues developed the technique, which is based on alterations to a Sequenom technology designed for germline genotyping, in early 2006.
[ pagebreak ]
MacConaill says that germline mutations in a given heterozygous sample are present at an approximate 50:50 frequency — there can be 100 percent of a given allele, A, or 100 percent of another allele, B, or half A and half B. This is almost never the case in cancer tumor samples; admixture with stroma and healthy tissues makes it "very difficult to get pure tumor," she says. Additionally, other confounding variables — such as amplifications, deletions, or aneuploidy — may come into play. "It's difficult, then, to be able to infer whether a mutation is present in a sample if you're not expecting to see it at a 50:50 [mutation to wild-type] or a 100:0 percent ratio," MacConaill says.
Because of this, the team adjusted the Sequenom algorithm to seek wild-type clusters. In this way, they are able to search against deviations from that cluster "instead of pre-defining what we would expect to see in terms of AA, AB, BB," MacConaill says. In essence, she adds, the team had to train the algorithm on existing data sets from thousands of cancer samples.
It's a complicated process, to be sure, but MacConaill says that the optimization really pays off. In analyzing 91 fresh-frozen and 93 formalin-fixed paraffin--embedded tumor tissue samples, and interrogating them in a high-throughput manner with OncoMap for nearly 400 mutations in 33 known oncogenes and tumor supressor genes, the team was able to achieve greater sensitivity and specificity than with Sanger sequencing.
Their analyses, MacConail says, have identified several actionable mutations — those that can be targeted by specific inhibitors. Not only can OncoMap predict the sensitivity of a target to a particular drug, but it can also determine the likelihood of resistance to it (as they've shown with KRAS mutations). This information could be essential in the clinic because it shows "not necessarily just which drugs should be used, but possibly [which] drugs should not be used for that patient," she says.
OncoMap is not without its limitations, however. To keep the approach up-to-date requires continual assay development; to keep it viable in the clinical setting requires substantial expertise. Beyond performing the genomic analyses and bioinformatic statistical calculations, clinicians and oncologists are left with perhaps the most daunting task — interpreting these data so that they are useful at the bedside. "That will require, going forward, consortia to be developed, more access to clinical trials, and more point people," with substantial expertise, MacConaill says.
Bioinformatics: Tools to Uncover Biomarkers for Renal Cancer Subtypes
As a general rule, being able to take smaller and smaller tumor biopsies has been a boon as it is less invasive for the patient. With those smaller samples, though, there is less tissue left over. "As we move toward less invasive diagnostics, clinical laboratories have less to work with," says Andrew Young, a pathologist at Emory University School of Medicine. But, he adds, "with microarrays, we can sample any number of potential biomarkers."
Young teamed up with Georgia Tech's May Wang to apply her bioinformatics tools to his kidney cancer studies. Together, they used three tools developed by Wang's team to find biomarkers to distinguish the three subtypes of renal cell carcinoma: clear cell, papillary, and chromophobe.
[ pagebreak ]
To begin their search, Wang and Young rounded up microarray data from three previously published studies that examined the various subtypes of renal cancer. They then used the first of Wang's three tools, ArrayWiki, to combine those data sets — giving them a larger sample size to draw from — and to update the annotation of the probes. Like other wikis, ArrayWiki relies on the expertise of community researchers to annotate the microarray data.
That data then flowed through caCORRECT, Wang's Web-based quality control check system that searches through array data to find and remove artifacts — such as scratches or bubbles that can occur on microarrays — that can skew the data.
Next, it was into the wringer of omniBioMarker, which analyzed the data set to identify potential biomarkers. Using support vector machine classifiers — which in this case were previously verified biomarkers — omniBioMarker ranked the genes and then made small changes to the parameters of those classifiers to find the best algorithm. From that, Wang and Young found six potential biomarkers that they then followed up on using qRT-PCR and immunohistochemistry. All of the biomarkers were verified.
"What May's tools have really enabled us to do is, number one, look in great detail and great sophistication at whether there are any mechanical problems with our microarrays that might have skewed our data," Young says. "And then from there, to use the power of computers to really systematically look at any number of algorithms to assess the microarray data to see which one seems to fit best with our preexisting knowledge … and we found much higher rate of verifying our candidates from microarray data using May's tools."
Proteomics: Breast Cancer Biomarkers Found With Modified MALDI-MS
While the proteomics research community has inundated the literature with investigations of potential biomarkers for cancer throughout the last decade, no one approach has been universally accepted as the optimal method for the detection and validation of these small molecule indicators.
Recently, mass spectrometry techniques have begun to bolster — and sometimes even replace — traditional immunohistochemical approaches for biomarker detection because of the reduced cost, increased specificity, and the time savings they afford researchers. Richard Caprioli, director of the Mass Spectrometry Research Center at the Vanderbilt-Ingram Cancer Center, is using an optimized MALDI mass spec approach to differentiate proteomic biomarkers for breast cancer.
"Our focus is to do this kind of imaging so that you're actually looking at the molecules [of interest] themselves, and not at surrogate markers" such as antibodies, which can raise issues of specificity and fidelity, Caprioli says. "Using mass spectrometry, we're actually looking at the molecule itself. … We're looking at signatures for disease, not just one or two markers obtained from immunohistochemical stains."
Because the MALDI-mass spec technique does not require a target-specific reagent, and because some of the newest instruments generate as many as 5,000 mass spectra per second, Caprioli says that a researcher could harvest all necessary molecular data from a small biopsy in a matter of minutes — whereas traditional liquid chromatography-mass spec methods could require hours to process a single sample.
[ pagebreak ]
But rapid, high-throughput spectrometry reads are not necessarily immediately attainable for all types of studies. Caprioli says that in-house optimization is key. He and his team were able to significantly increase the sensitivity of their reads by utilizing custom sample preparation protocols — for example, by augmenting compounds or creating derivatives more amenable to desorption and ionization — and instrumental alterations. While each alteration might only improve the protocol a small amount, collectively, they can add up. "There are a whole bunch of things that one can do. Individually, they might only add 10, 15, or 30 percent to the sensitivity," he says. "But if you do four or five of these things — and they all give you some gain — cumulatively, you get a significant gain in sensitivity."
The same idea goes for statistical analyses of MALDI-mass spec results; Caprioli says that his team uses a combination of off-the-shelf packages, such as the Significance Analysis of Microarrays program from Stanford University, and those custom-designed by his collaborators, which focus on their particular tissue analysis needs.
Caprioli acknowledges that, as with all technologies, the MALDI-mass spec approach has its limitations. "At the moment we see medium- to high-abundance proteins, but we'd like to go down to really low copy numbers," he says. "To integrate the data better and to understand the data to get this molecular information is always a challenge."
Sequencing: UCLA Team Completes Brain Cancer Cell Line Sequence
In an effort to lead the way forward for personalized cancer treatment, researchers at the University of California, Los Angeles' Jonsson Comprehensive Cancer Center recently performed the first complete sequence of a brain cancer cell line. The team, led by Stan Nelson, a professor of human genetics and director of the center's Gene Expression Shared Resource, focused on the U87 glioblastoma cell line, one of the most thoroughly studied brain cancer cell lines. Nelson's group identified many chromosomal translocations, deletions, and mutations that could result in the development of cancer. Taking advantage of ABI SOLID technology, they were able to simultaneously read billions of different DNA fragments from this cancer and analyze it more than a billion times to make sure that the results would be sensitive and accurate.
"It's definitely a technological milestone in the sense of the costs of sequencing, and choosing what you want to sequence is going to shift over time. This is a very commonly used cell line — it probably makes sense to sequence perhaps even the 100 most commonly used cancer cell lines because we base a lot of our cell biology knowledge on the cell lines, yet we're doing it in the absence of knowing how they're all mutated per se relative to the normal genome," Nelson says. "What inhibits people from doing these things is largely cost and effort of generating the sequence." They hope that their study might help pave the way for new methods, such as a sensitive molecular assay, to facilitate earlier diagnosis and treatment for brain cancer by keeping tabs on recurrence.
In addition, these findings could be used to develop a tool to help physicians determine whether a cancer had been effectively eliminated and curb what can sometimes be long-lasting health issues from extended, unnecessary treatment. But for now, the group has developed a website where other investigators can access the sequencing data to use in their own experiments.
Nelson says that in order for this research approach to translate into new treatments, it will be critical to sequence a larger number of brain cancer cell lines, brain tumors from patients, and probably even from selected subsets of cells from individual cancers. In the not-so-distant future, he hopes to have a process by which they can sequence a patient's individual cancer and process that data quickly enough so that an oncologist could use the information to make critical treatment decisions.
[ pagebreak ]
The researchers in Nelson's lab are currently developing a sequence analysis pipeline to enable them to cheaply and effectively sequence other cancer cell lines. "We're sequencing a panel of glioblastomas for confirming if the mutations that we're observing are occurring commonly," he says.
Metabolomics: An Image of Metabolites Helps Pinpoint Prostate Tumor
With the advent of the prostate-specific antigen test, the incidence of prostate cancer diagnoses began to rise–and so did the number of biopsies performed. The problem there, says Leo Cheng at Massachusetts General Hospital, is that some of the men had tumors that were never going to grow and cause a problem, but they still underwent treatment. Furthermore, he adds, the PSA test is prostate-specific, not cancer-specific; other diseases, including enlarged prostates, can raise antigen levels.
Biopsies, then, are the only way to be sure it's not a false-positive. However, only about 10 percent of needle passes are successful in finding the tumor — though with ultrasound-guided biopsies, surgeons are more likely to hit the prostate rather than the bladder.
To tackle the problem, Cheng has developed a mass spectrometry-based approach to determine the chemical profiles of different regions in the prostate to pinpoint where a tumor is and where the biopsy needle should go.
"In this one [we asked] by using the chemical profile in different regions of the prostate, is it possible for us to at least point to the clinician where to take [the] biopsy?" Cheng says. "My clinician collaborator, the chief of urology, Dr. MacDougal, basically told me very clearly: 'Don't tell me how you found it, just tell me: Where is the red zone?'"
By imaging the metabolites — seeing those areas of intensity of the individual chemicals — on top of the anatomical structure, Cheng can determine where a tumor might be. In his images, he includes 36 molecules from metabolic pathways, chosen because they are correlated with disease. He adds that it's similar to what experienced doctors do in their heads: take disparate pieces of information and synthesize them to make a decision.
In a pilot study, Cheng looked at five cases and studied seven cross-sections. For each cross-section, his group looked at 256 spectra to determine if the sample contained prostate cancer. They found that metabolomic imaging was about 93 to 97 percent accurate, though he cautions that the pilot study was small.
MicroRNAs: Connections Among Non-Coding RNA Dysregulations in Several Cancer Types
Because it's known that approximately 50 percent of the genes that code for microRNAs are localized in cancer-associated genomic regions, it's reasonable to imagine that similarities exist in the dysregulation of these small, non-coding RNAs among cancer types. And they certainly do, according to new research out of the Indian Statistical Institute.
Sanghamitra Bandyopadhyay, a professor in the Machine Intelligence Unit at ISI, and her colleagues developed a cancer-miRNA network that visualizes the connections among miRNA modifications in eight cancer types. Bandyopadhyay says that the team created the network by mining the literature of experimentally verified cancer-miRNA relationships, with special attention paid to "upregulation and downregulation, [the] techniques used to measure the expression levels, miRNA chromosomal locations with start and end points, fold change, P-values, and samples and cell lines used in the experiments," among others. "The purpose of the cancer-miRNA network is, primarily, to visualize the involvement of the different miRNAs in a number of cancer types on a global scale," she says. "Such a global view did not [previously] exist."
[ pagebreak ]
The team found nine miRNAs that displayed significantly similar dysregulation in lung, colon, and pancreatic cancer cell lines. They were surprised to find that, of the nine, seven also showed comparable dysregulation in prostate tumor tissue. By maintaining a global perspective, Bandyopadhyay says the team realized that clustered miRNAs show differential coexpression patterns across cancer tissues, and that cancer-specific dysregulation patterns may exist.
"An important observation that became evident from the network was that neighboring miRNAs or clustered miRNAs show similar dysregulation patterns in cancer tissues, suggesting common regulatory pathway," she says.
Indeed, Bandyopadhyay says that there is still much work to be done. In the future, the team hopes to weave other biomolecular interactions into the cancer-miRNA web in order to gain an improved understanding of the complex regulatory mechanisms of the disease.
Mass Spectrometry: High-Throughput Gene Mapping Method for Tumors
A multi-site research effort comprised of investigators from the Broad Institute, the Dana-Farber Cancer Institute, and Harvard Medical School has recently developed a new, high-throughput approach to analyzing genes in cancer tumors that may expedite the development of targeted cancer drug treatments. The researchers tweaked a common mass spec method used to identify single nucleotide variations to help elucidate defects in tumor DNA that alter the function of some oncogenes.
Current methods for exploring the cancer genome rely on large-scale DNA sequencing; however, this becomes cost-prohibitive if the goal is to analyze multiple tumors. Instead, the new method allows them to measure a subset of the most relevant mutated genes among various types of tumors simultaneously. They used this new technique to analyze 1,000 human tumor samples and were able to develop in-depth pictures about oncogene mutations in a slew of both common and rare cancer types. Some of the gene mutations the researchers uncovered had not been previously identified as being linked to certain cancers.
"This study is much larger in scope than anything that's been done to date, and the result is that we're able to discover things that have not been disclosed before and find out new things about the cancer genome because we had such a large data set," says Rameen Beroukhim, a postdoc at Dana-Farber. "The largest sample sets were on the order of 400 or 500 samples on a high-resolution platform, and there have been larger studies on low-resolution platforms, but those could only really give a sense of the very broadest things going on in the genome. We were using these SNP arrays that interrogate about 240,000 sites across the genome, so we had some 240,000 data points for each of the 3,000-some cancers that we were analyzing."
Beroukhim, who is also a physician, says that this type of study will first and foremost help advance cancer treatment from a classification perspective. Currently, the diagnosis and treatment of cancer is based on the tissue type in which it originated. "There's been a recognition that the current categorization scheme for cancer is not sufficient and I think there's an emerging recognition that what we should be targeting is not the tissue type but what's gone wrong in that cancer. Then we can figure out how to set it right," he says. "We found that when you look across cancers you can turn that old classification scheme on its side and find that most of the events that are in some small proportion of lung cancer are also in other cancers too. We have a classification scheme that relies on the molecular events in cancer, which show a lot of shared features in different cancer types–maybe more so than was previously understood."