In my last column (4/10/14 issue) I reviewed recent advances in non-cancer genomics, particularly the fun stuff involving Neanderthals and human migration. Genomic studies are rapidly filling in important blanks in human macrohistory, blanks for which there are no written records and only the scattered bony remains of long-dead ancestors.
Now I'd like to discuss some microhistory, specifically the history of human cancers. We are writing this history as we speak, with almost every issue of Nature and Science and Cell and their lesser relatives. This re-write of microhistory, like that of macrohistory, is the glorious outcome of a decade of falling prices for genomic analysis. While we don't know everything yet-when do we ever?-we are accumulating so much new data at such a rapid pace that every current cancer textbook is seriously out of date the moment it is published.
The last year has given us both 20,000-foot views of the genome as well as up-close-and-personal disease-specific stuff.
Two excellent papers published last year looked at the "20,000-foot view" of cancer. Writing in Nature, Cyriac Kandoth and colleagues in The Cancer Genome Atlas (TCGA) project looked at point mutations and small insertions/deletions across 12 tumor types (2013;502:333-339), overall finding 127 significantly (i.e., driver) mutated genes; most cancers have two to six drivers on average. Some of these (particularly transcriptional factors/regulators) tend to be tissue-specific, and some (such as histone modifiers) are relatively more ubiquitous.
Some mutations (for instance, NRAS and KRAS) seem mutually exclusive: overall there are 14 mutually exclusive pairs. Mutational pairings are common (148 co-occurring pairs), but some pairings are disease-specific (IDH1 and ATRX in glioblastoma, TBX3 and MLL4 in lung adenocarcinoma). There's probably some interesting biology going on related to these pairings and non-pairings-matches made in hell, perhaps. Some mutations are tumor-specific (NPM1 and FLT3 in AML), but most are not.
Lawrence and colleagues at the Broad Institute performed a similar analysis across 27 tumor types in an attempt to examine genomic heterogeneity. Their analysis, published in Nature last year (2013;499,214-218), focused not on driver mutations per se, but on overall somatic mutation frequencies. There are huge differences in mutational frequency, both within and across cancer types: more than three orders of magnitude from the least to the most mutated.
There are also significant differences in the types of mutations seen, differences that appear to reflect (in part) tumor etiology. Cervical, bladder, and head and neck cancers, for instance, share Tp*C mutations that may reflect their viral etiology. GI tumors tend to share frequent transitions at CpG dinucleotides. Etiology, in turn, also affects frequency, with highly mutagenized tumors occurring as a consequence of tobacco, hamburgers, and tanning salons.
Analyses such as those of Kandoth and Lawrence have focused (quite naturally) on genes coding for proteins (the exome). But most DNA is non-coding, what we used to call "Junk DNA." Khurana and colleagues at the Sanger Institute identified some 100 potential non-coding cancer driving genetic variants in breast, prostate, and brain tumors.
Khurana's paper suggests we've barely scratched the surface of genomic analysis. I am reminded of Isaac Newton's fetching phrase (I've quoted it in a previous column, but it appears appropriate to the non-coding DNA story): "I was like a boy playing on the sea-shore, and diverting myself now and then finding a smoother pebble or a prettier shell than ordinary, whilst the great ocean of truth lay all undiscovered before me." But the important thing is that we have, in recent years, dipped our toes into that great ocean of truth.
A whole series of individual cancer genomes are now being rolled out for public display, and the results continue to disentangle cancer-specific biology and (perhaps) suggest new tumor attacks. A recent publication by Ojesina and colleagues in Nature (2014; 6,371-375) identified (as one example among several) HER2 somatic mutations (not the standard breast cancer-type amplifications) in six percent of cervical cancers.
Will these offer new treatment options for these patients? We will see soon, I suspect. We are seeing a profusion of low-frequency mutations in relatively rare diseases. Running the numbers, it's hard to imagine performing Phase III trials to establish benefit for all of these.
What's a clinical trialist to do? How should the regulators deal with the tiny Phase II trials that will result? How are patients (and payers) to understand cost and benefit equations arising from those trials?
Some of the more fascinating studies reported in the past year explored the clonal evolution of human cancers. Sohrab Shah and colleagues in Vancouver analyzed 104 early triple-negative breast cancers. To quote their Nature article (2012:486:395-399): "TNBCs vary widely and continuously in their clonal frequencies at the time of diagnosis[horizontal ellipsis] At the time of diagnosis these cancers already display a widely varying clonal evolution that mirrors the variation in mutational evolution."
There is a point in Jurassic Park where a seasoned big game hunter, busy pursuing a nasty-looking velociraptor, suddenly finds himself outmaneuvered by the predatory dinosaur's concealed partner. His last words before being chewed up are "Clever girl." That's the way I felt after reading Shah's paper. Clever girl: TNBC will be a long, hard slog of a disease if we remain stuck in a morass of kinases.
Similar to the TNBC paper, one by Landau and colleagues investigated the evolution and impact of subclonal mutations in chronic lymphocytic leukemia. This work, published last year in Cell (2013;152:714-726), was both fascinating and deeply disturbing. Measuring mutations at multiple time-points, they showed that chemotherapy-treated tumors (though not untreated CLL) underwent clonal evolution, picking up new driver mutations (and increased measurable genetic diversity) that expanded over time, and that these drivers were powerful predictors of outcome. "Never bet against Charles Darwin" is a good rule in biology, just like "Never bet against Albert Einstein" is a good physics rule.
First Steps Toward Clinical Utility
While most of the excitement in cancer genomics has revolved around the large scale projects such as The Cancer Genome Atlas Project and its kindred studies, projects that in essence serve to describe the mutational landscape of human cancer(s), we are now beginning the first steps towards clinical utility in cancer genomics.
My institution, like many around the country, has wrestled with how to "bring genomics to the masses." The actual performance of tumor deep sequencing has gotten relatively inexpensive (relatively, that is, to those versed in the cost of an hour in an emergency room), and almost technically trivial. But running an assay is only the first step on the journey.
How does one analyze the data from an analysis of three billion base pairs? How does one do it in a timely fashion (patients appropriately want treatment today, not eight weeks from now)? Fortunately these technical challenges are becoming easier to handle: a recent report from the University of Chicago (published in Bioinformatics) demonstrated the use of supercomputers to process 240 genomes in two days.
But analytic problems are more than just technical challenges. They involve real judgment calls. How does one decide what is an important driver mutation, as opposed to a multitude of "passenger" mutations? When is a mutation actionable?
Indeed, what does "actionable" even mean? In current genomics-speak, "actionable" appears to mean something along the lines of "We found a mutation. That mutation is associated with badness in some disease. There is a drug that does something to interfere with that badness in some way in a preclinical model or some other cancer somewhere in the scientific literature."
"Actionable" is easily misunderstood by patients and referring physicians as "we have a drug that works for your tumor." What it really means is "maybe using this drug is not a total shot in the dark."
Commercial enterprises are now leaping blindly into this evidence chasm, and asking us to jump in along with them, paying their development costs. Like all lovers' leaps, I guess you just have to take some things on faith. But some of us would prefer, before ordering a test, to experience the cool soothing balm of carefully curated data demonstrating that obtaining the test actually affects outcome. Old-fashioned, I know.
How to obtain that evidence is the subject of a great deal of noodling by clinical/translational researchers. The upcoming NCI Match trial, a multi-thousand patient attempt to match mutations with drugs in an integrated, cross-disease clinical trials platform, will launch later this year, to much genuine interest and excitement by both patients and physicians.
Most cancer genomics to date has been performed on well-curated, carefully prepared, highly cellular tumors. Can we turn genomic analysis into a blood test? We are now beginning to see publications in the "liquid biopsy" field, such as last year's New England Journal of Medicine publication by Sarah-Jane Dawson and her Cambridge colleagues (2013;368:1199-1209).
To cut to the chase, we can now measure specific genomic alterations (Dawson measured PIK3CA and TP53) in human plasma, and those alterations can be used to track the course of metastatic disease with sensitivity higher than that of circulating tumor cells or protein tumor markers such as CA 15-3. This is a good beginning, though only a beginning: measuring badness is necessary but insufficient to ameliorating badness, as the recent S0500 breast cancer circulating tumor cell trial demonstrated. This technology will really take off when we can use it to identify actionable (there I go again, but you know what I mean) mutations.
All in all, we're constantly being reminded that we live in the Golden Age of cancer biology, a breathtaking rollercoaster ride filled with the excitement of discovery. Have you ever read a book you loved so much that you hated for it to end? One where you wanted to keep on reading after the last page to find out what happened to all those fascinating characters? We're reading that book, the book of Nature, right now. But don't worry: we're still only somewhere in the middle. Maybe even the first few chapters. There's plenty left to learn.
More from George Sledge on His OT Blog!
Read and comment at http://bit.ly/OT-Sledge