“Personalized medicine.” You’ve probably heard the term. It’s a bit of a buzzword these days and refers to a vision of future medicine in which therapies are much more tightly tailored to individual patients than they currently are. That’s not to say that as physicians we haven’t practiced personalized medicine before; certainly we have. However it has only been in the last decade or so that our understanding of genomics, systems biology, and cell signaling have evolved to the point where the vision of personalized medicine based on each patient’s genome and biology might be achievable within my lifetime.
I was thinking about personalized medicine recently because of the confluence of several events. First, I remembered a post I wrote late last year about integrating patient values and experience into the decision process regarding treatment plans. Second, a couple of months ago, Skeptical Inquirer published an execrably nihilistic article by Dr. Reynold Spector in Skeptical Inquirer in which he declared personalized medicine to be one of his “seven deadly medical hypotheses,” even though he never actually demonstrated why it is deadly or that it’s even really a hypothesis. Come to think of it, with maybe–and I’m being very generous here–one exception, that pretty much describes all of Dr. Spector’s “seven deadly medical hypotheses”: Each is either not a hypothesis, not deadly, or is neither of the two. Third, this time last week I was attending the American Association for Cancer Research (AACR) meeting in Orlando. I don’t really like Orlando much (if you’re not into Disney and tourist traps, it’s not the greatest town to hang out in for four days), but I do love me some good cancer science. One thing that was immediately apparent to me from the first sessions on Sunday and perusing the educational sessions on Saturday was that currently the primary wave in cancer research is all about harnessing the advances in genomics, proteomics, metabolomics, and systems and computational biology, as well as the technologies such as next generation sequencing (NGS) techniques to understand the biology of each cancer and thereby target therapies more closely to what biological abnormalities drive each cancer. You can get an idea of this from the promotional video the AACR played between its plenary sessions:
Which is actually a fairly good short, optimistic version of my post Why haven’t we cured cancer yet? As I mentioned before, with this year being the 40th anniversary of the National Cancer Act, as December approaches expect a lot of articles and press stories asking that very question, and I’m sure this won’t be the last time I write about this this year.
“Personalized medicine” in CAM
In the meantime, before I discuss a couple of examples of how science-based medicine is moving ever more closely to personalized medicine, I can’t help but note that part of what inspired this bit of my typical blather on this topic was sitting in the audience at AACR hearing about all these tour de force genomic analyses that begin to reveal the individuality and complexity of tumors, and, more importantly, to suggest strategies to target the specific abnormalities that drive the growth and metastasis of each cancer and contrasting in my mind the claims of “personalized” or “individualized” medicine that practitioners of “complementary and alternative medicine” (CAM) and “integrative medicine” (IM) like to make. As I pointed out about a year ago, “individualized” treatment in CAM-world basically means “making it up as you go along.” Consider, for example, homeopathy, which postulates prescientific ideas for the cause of disease, claiming that “like cures like,” and then using unscientific “provings” to determine which remedies can be used to “treat” each condition. Never mind that homeopathy is water (does it really need to be repeated?), consisting of remedies serially diluted and succussed so many times that in many of them it is highly unlikely that there is a single molecule of remedy left in the concoction. Not only that, but one of the most popular homeopathic remedies for flu consists of basically the ground-up liver and heart of a Muscovy duck.
In CAM-world, “personalization” or “individualization” means “making it up as you go along.” A good example of this is a post on that wretched hive of scum and quackery, The Huffington Post, by Dr. Mark Hyman, he of “functional medicine” fame, where, under a section entitled “Treating individuals, not diseases” he writes:
There is no effective known treatment for dementia. But we do know a lot about what affects brain function and brain aging: our nutrition, inflammation, environmental toxins, stress, exercise, and deficiencies of hormones, vitamins, and omega-3 fats.
It is not just one gene, but the interaction between many genes and the environment that puts someone at risk for a chronic disease such as dementia. And we know that many things affect how our genes function — our diet, vitamins and minerals, toxins, allergens, stress, lack of sleep and exercise, and more.
Hyman then goes on to describe an anecdote of a man with developing dementia. Typical of many CAM doctors, Dr. Hyman chased down all sorts of “abnormalities,” prescribed all sorts of supplements to “fix” those abnormalities, and subjected the man to various “detox” regimens, including unspecified “medications that helped him overcome his genetic difficulties by getting rid of toxins.” As Steve Novella pointed out at the time, in reality what Dr. Hyman was doing was a “bait and switch,” in which he extrapolates from preliminary results in real science and uses them to come up with proposed treatments that have not been validated for the purposes that he uses them for. His evidence for success? Not science, not clinical trials, not even preclinical data, that’s for sure. Instead Dr. Hyman presents a couple of anecdotes.
Indeed, much of this sort of “making it up as you go along” is on full display in the anti-vaccine movement. Indeed, the anti-vaccine crank blog Age of Autism has numerous examples of just this sort of “personalization,” including the hijacking of mitochondrial disorders as a predisposing factor for “vaccine injury” causing autism, the serial use of all manner of “biomedical quackery” to “recover” their children, up to and including stem cells, and de facto unethical experimentation on autistic children. Be it chelation therapy, supplements, hyperbaric oxygen, dubious stem cell therapies, and other quackery, the anti-vaccine movement, “biomedical” quacks “individualize” their treatments to each autistic children, using dubious labs like Doctor’s Data to come up with lab abnormalities to “treat.” Of course, there is a fundamental disconnect between the claims of DAN! doctors to “individualize” therapy to each patient with their unwavering belief that vaccines cause autism. No matter how much they try to hide it, vaccines remain The One True Cause of the conditions known as autism and autism spectrum disorders.
There are many, many more examples of this kind of “personalization” of medicine in CAM based on either no science or unjustified extrapolation from existing science, dubious lab tests, practitioner biases, and a veritable panoply of One True Causes of disease. So let’s contrast with evolving personalized medicine in science-based medicine.
Personalized medicine in SBM
Because I’m a cancer researcher and surgeon, I find that currently the most promising examples of how genomics can contribute to personalized medicine come from cancer. This should not be surprising, because cancer is not one disease. It’s hundreds, perhaps thousands, of diseases, and even cancers arising from the same cells in the same organ can have very different biology. Basically, as I put it before, stealing liberally from The Hitchhiker’s Guide to the Galaxy, Cancer is complicated. You just won’t believe how vastly, hugely, mind-bogglingly complicated it is. I mean, you may think it’s complicated to understand basic cell biology, but that’s just peanuts to cancer. This point was driven home in the AACR video above, and I see it driven home with each new study of cancer genomics or heterogeneity that comes out in high impact journals every month.
For example, couple of months ago, I described a tour de force study of changes that occur in the genome of prostate cancer cells compared to normal prostate. The study demonstrated a number of alterations in the genome affecting molecular pathways that drive growth and metastasis and that could potentially be targeted for therapy. Then, not long before going to the AACR meeting, I came across this article in my news feed, Lung Cancer Evolves With Treatment, Study Finds, which refers to this study from Harvard by Sequist et al, Genotypic and Histological Evolution of Lung Cancers Acquiring Resistance to EGFR Inhibitors. The study itself demonstrates something we’ve known for quite some time, which is that tumor cells evolve under selection by various modalities used to treat them. In this case, the authors studied how non-small cell lung cancer (NSCLC) evolves resistance under treatment with drug therapy targeted to the specific mutation driving their growth. In this case, the gene in which mutations result in its being turned on is the epidermal growth factor receptor (EGFR), and the targeted therapy consists of inhibitors of EGFR such as gefitinib (brand name: Iressa) or erlotinib (brand name: Tarceva). Basically, the investigators subjected tumor tissue from patients who had developed resistance to EGFR tyrosine kinase inhibitors to systematic genetic and histological analysis. What they found represented a confirmation of some known genetic changes that occur to result in resistance but also some unexpected changes, including:
All drug-resistant tumors retained their original activating EGFR mutations, and some acquired known mechanisms of resistance including the EGFR T790M mutation or MET gene amplification. Some resistant cancers showed unexpected genetic changes including EGFR amplification and mutations in the PIK3CA gene, whereas others underwent a pronounced epithelial-to-mesenchymal transition. Surprisingly, five resistant tumors (14%) transformed from NSCLC into small cell lung cancer (SCLC) and were sensitive to standard SCLC treatments. In three patients, serial biopsies revealed that genetic mechanisms of resistance were lost in the absence of the continued selective pressure of EGFR inhibitor treatment, and such cancers were sensitive to a second round of treatment with EGFR inhibitors. Collectively, these results deepen our understanding of resistance to EGFR inhibitors and underscore the importance of repeatedly assessing cancers throughout the course of the disease.
Or, as Dr. Lecia Sequist, lead author of the study, put it:
“It is really remarkable how much we oncologists assume about a tumor based on a single biopsy taken at one time, usually the time of diagnosis,” lead author Dr. Lecia Sequist said in an MGH news release. “Many cancers can evolve in response to exposure to different therapies over time, and we may be blind to the implications of these changes simply because we haven’t been looking for them.”
“Our findings suggest that, when feasible, oncogene-driven cancers should be interrogated with repeat biopsies throughout the course of the disease,” Sequist said. “Doing so could both contribute to greater understanding of acquired resistance and give caregivers better information about whether resumption of targeted therapy or initiation of a standard therapy would be most appropriate for an individual patient.”
Now, that would be personalized medicine, based on science, in marked contrast to what passes for “personalized” medicine in CAM.
Another cancer for which new findings in genomics and systems biology hold great promise is the cancer I spend most of my time treating and researching, breast cancer. Current methods to predict prognosis and guide treatment are crude and include stage as measured by volume of the primary tumor; presence and number of lymph node metastases; presence or absence of distant metastases; tumor grade as measured histologically; expression or lack of expression of important hormone receptors such as estrogen receptor (ER) and progesterone receptor (PR); and amplification of ErbB2 (HER2). Used together, these factors allow, albeit roughly, a degree of prediction of prognosis, as well as of personalization of therapies such as hormonal treatments (tamoxifen or aromatase inhibitors) or agents targeted at HER2 (trastuzumab). Aromatase inhibitors had not yetbecome widely available, and Herceptin (trastuzumab) had been FDA-approved for women with HER2-positive metastatic cancer but not yet for the adjuvant therapy of women with earlier-stage HER2-positive breast cancer. (That did not come until 2006, and then only with chemotherapy.) Then, in 2000, Perou et al published a classic paper in Nature that used then state-of-the-art cDNA microarrays to divide breast cancers into subtypes based on gene expression patterns, which, based on this work and work done since then, currently include normal-like, basal-like, luminal (A and B), and HER2(+)/ER(-), and there is a growing body of literature (for example, this study) that suggests that these different subtypes respond differently to different chemotherapy and targeted agents.
We learned many things from this work, which has accelerated over the last decade. For example, we now know that there are several intrinsic groups of breast cancer based on the patterns of gene expression they exhibit, and these groups subdivide many of the “classic” divisions we have been using for at least two decades, such as ER(+) or ER(-). More importantly, there is one form of breast cancer that expresses none of these markers. Dubbed “triple negative breast cancer” (TNBC), this form of breast cancer is defined as expressing neither ER, PR, nor HER2. TNBC is a close relative of a type of breast cancer categorized a decade ago through gene expression profiling and dubbed “basal-like” (synonymous terms include “basal-type,” “basal-epithelial phenotype,” “basal breast cancer,” and “basaloid breast cancer”). For the most part, currently the same treatments are used for TNBC and basal-like breast cancer because the sine qua non of TNBC is that there are no known molecular targets in this breast cancer subtype, while non-TNBC basal-like breast cancer tends to express HER2 and thus be susceptible to Herceptin. Because TNBCs do not respond to drugs targeting ER or HER2, cytotoxic chemotherapy is currently the only option for adjuvant or neoadjuvant therapy in women with operable TNBC or for systemic treatment for metastatic disease. Paradoxically, TNBCs are more sensitive than ER(+) luminal tumors to standard chemotherapy regimens, but unfortunately this increased chemosensitivity does not translate into prolonged overall or disease-free survival. Consequently, the identification of new molecular targets or oncogene signatures that can be targeted for therapy, either with new agents and/or of new synergistic combinations of old agents, is a critical problem to be overcome for TNBC. Tantalizing hints of how this might be done arose at AACR, for example, this study in which the genomes of 50 breast cancers were sequenced.
Another area in which genomics might assist us as clinicians in breast cancer is through answering a rather vexing question regarding racial disparities in cancer outcome. For example, although the incidence of breast cancer among premenopausal African-American women is lower than among Caucasian women, African-American women are more likely to die from their disease, with a breast cancer-specific mortality of 33 per 100,000 and five year survival of 78% compared to 23.9 per 100,000 and 90%, respectively, for Caucasians. Breast cancer among African-American women tends to be characterized by higher grade, later stage at diagnosis, and worse survival, even after controlling for age and stage. Although it is true that the causes of these observed differences are likely to be multifactorial and include socioeconomic factors, such as differences in access to screening and treatment, there also appear to be biological differences in breast cancer in AA women. Indeed, evidence supporting biological differences as a major part of the explanation for these observed racial disparities was reported from the Carolina Breast Cancer Study. This study reported that that the TNBC/basal-like breast cancer subtype is nearly three times more common among premenopausal African-American women than among Caucasian women In marked contrast, the HER2(+)/ER(-) subtype did not vary appreciably with race or menopausal status, and the less aggressive ER(+) luminal A subtype was less prevalent in premenopausal African-American women. These results suggest that two questions remain open: (1) whether there is a difference in breast cancer biology that drives the tendency of young AA women to develop TNBC at a much higher rate than Caucasian women and (2) whether these biological differences, if they exist, can be exploited therapeutically to develop personalized regimens targeted at patients’ individual tumors. These are the sorts of questions that genomics can potentially answer and personalized medicine be based upon.
Don’t get me wrong. We are not yet near true personalized medicine for breast cancer. Indeed, if you want to get an idea of the challenges that remain, several of the talks I attended are available for free at the AACR website. Talks that are worth watching include:
- Harold Varmus, New Directions in Cancer Research
- Lynda Chin, Translating cancer genomics: From discovery to medicine (unfortunately, nearly none of the slides from this talk have been made available)
- Waun Ki Hong, The landscape of cancer prevention: Personalized approach in lung cancer
- Andrea Califano, A systems biology approach to integrative cancer genomics (Another one without slides.)
There are many others, but unfortunately most of the relevant talks are either not posted yet, require payment, or both. It’s annoying to me, but on the other hand I understand that it costs money to produce these and put them up on the web. Still, the sheer number of talks on The Cancer Genome Atlas, which goes by the annoyingly cutesy acronym TCGA, is telling. Every day, or so it seemed, there were multiple talks on TCGA, reporting findings, telling investigators how to access the data on its website, and discussing the progress. Basically, TCGA is becoming a repository of genome sequences of many cancers, a resource that can be mined. It’s not too hard to envision that one day, when it has many thousands of cancer genomes stored away from before and after treatment, its database might serve as the basis for computer algorithms that compares a patient’s tumor genome to the database and comes up with a list of recommended treatments.
Unfortunately, the move towards personalized medicine not without its share of opportunists and companies selling kits based on genetic tests that either haven’t been validated in clinical trials sufficiently to support their clinical use, for example Anne Wojcicki of 23andMe, whose pitch is has a lot more in common with “health freedom” arguments than it does with actual scientifically validated uses of genomic data, complete with heavy promotion in various social media. It’s a trait shared with enthusiasts of direct-to-consumer genetic testing, whose language really does harken to that of the “health freedom” movement. For example, compare this post with this post by Mike Adams, and the main difference you’ll find in the arguments will be in degree, not kind. It’s “health freedom” all around. Even while promoters, in a fit of cognitive dissonance, simultaneously accuse physicians of paternalism on this issue and admit that the burgeoning personal genomics industry needs to be “purged of scammers and bottom feeders,” it’s an effort designed to create an army of people who “will go nuts” at any attempt by legislators to legislate direct access to personal genomic data or regulatory agencies to control more tightly access to direct-to-consumer genetic testing. As Harriet Hall put it correctly, when it comes to routine genomic testing, we’re not there yet, not the least of which because, as Scott Gavura has pointed out, there are lots of problems with the testing. As I put it, much of the promotion of personal genomic testing is disturbingly similar to the promotion of various autism “biomedical” therapies by DAN! doctors or Dr. Hyman’s panoply of woo to which he subjects his patients in that it is an extrapolation from data that are too preliminary to justify widespread use. That may well change one day, but today is not yet that day, and saying so, to me at least, is akin to pointing out that the “do it yourself” use of unproven cancer therapies like dichloroacetate is usually not a good idea.
We are, however, making progress, and we’re making that progress not based on speculative extrapolation of preliminary science or accepting dubious science as true. I’ll close with an example that is now routinely used in the treatment of ER(+) breast cancer. As I mentioned above, ER(+) breast cancer tends not to be as sensitive to chemotherapy as ER(-) (and in particular, triple negative) breast cancer, even though it has a better prognosis. Over the last decade, a 21-gene assay has been developed for women with ER(+)/HER2(-) cancer that has not yet spread to the axillary lymph nodes called the OncotypeDX® assay. Based on the results of this assay, a recurrence score can be calculated. If it’s high, the tumor is likely to be sensitive to chemotherapy, which will improve the woman’s chances of survival. If it’s low, the tumor is likely to be insensitive to standard chemotherapy, and the recommendation is for the woman not to undergo anything other than anti-estrogen therapy. In this way, thousands of women will be spared chemotherapy that will not help them. Current trials are investigating the utility of this and other genetic prognostic tests in node-positive tumor or, in the case of Oncotype, in determining whether patients with intermediate scores can be spared chemotherapy, and, if so, which ones.
The revolution in genomics has been likened to a flow of water. In the late 1990s, it was a trickle. Five years later, it was a firehose. Today, it’s Niagara Falls, with terrabytes of data being produced every month. The cost of sequencing an entire genome has fallen from $100,000,000 ten years ago to under $30,000 per genome by late 2010, with the era of sub-$1,000 genome sequences in sight. There is definite promise in genomics to result in truly personalized medicine. The key will be to combine it with proteomics, metabolomics, and an understanding of environmental influences. Doing that will not be easy, and, despite the Niagara Falls of data currently deluging us, it will not be fast. Unlike the claims of personalized medicine that arise from CAM and IM, it will take science and clinical trials to tease out the true associations from the noise, to differentiate correlation from causation, to separate and quantify effects of environment from effects of genome, and to figure out the interactions between them all. Oddly enough, that’s why I find last year’s AACR promotional video to be a bit more realistic than this year’s, annoying techno background music aside:
There are successes, challenges and even failures, but always hope. It’s our time, but we have to stick to science-based medicine.