Twelve months after the debut of the first personal genomics services, Kevin Davies expectorates and explores the pros, cons and controversies of “23 et al.”
By Kevin Davies
Nov. 12, 2008 | One year ago, Iceland’s deCODE Genetics kicked off a new era in personalized genetics with the launch of a direct-to-consumer service dubbed deCODEme. A few days later, California’s 23andMe debuted its own service with the slogan: “Genetics just got personal. Don’t worry. We’re here to help.”
For less than $1000, deCODE and 23andMe offered customers the exciting opportunity to gander their unique genetic quirks and glitches, as revealed by the identities of more than 500,000 DNA markers (single nucleotide polymorphisms, or SNPs). Such services would have been unthinkable just two years ago, before the explosion of genome-wide association studies (GWAS) began pinpointing susceptibility genes for common, complex diseases such as cancer, heart disease, diabetes, and mental illness.
Some of the attraction of these web-based services is undoubtedly recreational—what’s my haplogroup? Do I have fast-twitch muscles? Am I a fast or slow caffeine metabolizer? Some people may be curious about tracing their family’s roots or comparing genomes to friends and family members. Adoptees may be interested in gleaning something of their biological background. But these services, which now include offerings by Navigenics and SeqWright, translate the latest peer-reviewed genomic research into estimates of individualized disease risk, potentially allowing consumers the opportunity to confirm indications from their family history or revealing wholly unsuspected predispositions to a given disorder.
The proselytizers of personal genomics acknowledge that these services are works in progress, but some anecdotal results have been remarkable.
Last summer, deCODE Genetics co-founder Jeffrey Gulcher, 48, reviewed his deCODEme data, which indicated he had a doubled lifetime risk for prostate cancer. Even though his PSA levels were fairly normal, Gulcher ended up getting a biopsy that revealed a grade 6 (Gleason scale) prostate carcinoma, which was successfully resected. “I’m not only the CSO, but I’m also a client,” Gulcher joked at CHI’s Beyond Genome conference last June. “This test may have saved his life,” says deCODE CEO Kari Stefansson in all seriousness.
The co-founder of Navigenics, Los Angeles physician David Agus, received similarly alarming results when he took his firm’s Health Compass analysis—an 80% risk of a heart attack. He told guests at the Navigenics launch party last Spring that he is taking appropriate lifestyle, dietary and medical action.
Sergey Brin, the co-founder of Google and husband of 23andMe co-founder Anne Wojcicki, recently disclosed that he carries a mutation in a Parkinson’s disease (PD) susceptibility gene called LRRK2. By searching all of his genotyped LRRK2 SNPs through his 23andMe account, Brin learned that he carried the same point G2019S mutation as his mother, who has PD. This puts Brin, he says, in a unique and fortunate position: “I know early in my life something I am substantially predisposed to. I now have the opportunity to adjust my life to reduce those odds… and support research into this disease long before it may affect me… I have a better guess than almost anyone else for what ills may be mine—and I have decades to prepare for it.”
A handful of anecdotal stories is no substitute for hard data of course, and the consumer genomics industry has sparked deep concern among the medical establishment and certain state regulatory agencies about the wisdom, validity, and legality of delivering medical information to consumers—and the public’s limited ability to interpret their odds ratios and relative risks without expert assistance. Rather than fork out $1000 for a genome scan, the New England Journal of Medicine argued earlier this year that it made more sense to buy “a gym membership or a personal trainer” and implement dietary changes to reduce the risk of heart disease and diabetes.
Last summer, the State of California issued “cease and desist” letters to 13 consumer genomics providers, but while some firms retreated, the major providers satisfied the State that their tests are performed in CLIA-certified labs with appropriate physician oversight.
Regulatory rumblings notwithstanding, it is easy to overlook just what has been achieved in the past 12 months thanks to the collective efforts of what BioTeam’s Michael Cariaso dubs “23 et al.” These genome scans are not, the providers stress, diagnostic tests, but a means of providing information that mirrors the stunning progress researchers are making in dissecting the genes underlying complex traits. Indeed for most genes, the individual impact on a person’s relative disease risk is minor. One exception is Apolipoprotein E (ApoE), which was strongly associated with late-onset Alzheimer’s disease (AD) by Duke University professor Allen Roses and colleagues (see, “A Virtual Pharma Organization,” Bio-IT World, June 2008). This is the gene that James Watson had redacted from his genome sequence last year rather than learn his potential AD risk.
“The odds ratio of ApoE4 is 3.0, which is pretty substantive,” explains Boston University’s Robert Green, who is conducting a trial to study reactions to Alzheimer’s genetic testing. By contrast, most gene associations in complex diseases produce odds ratios of 1.1 or 1.2 and thus have limited significance by themselves. “But people who are not statistically sophisticated will not necessarily understand that. All they’ll see is that the risk is higher or lower,” Green warns.
Green talks of a “nightmare scenario” in which a woman is sophisticated enough to order a personal genome scan, but unable to distinguish the BRCA1 hereditary breast cancer gene from the breast cancer markers offered by 23 et al. “Oh my God, I got the gene for breast cancer,” she thinks. Her surgeon studies the printout and concurs: “Yeah, you’re positive for the gene, I guess we better [schedule surgery].”
The author holding a mock Navigenics Health Compass scorecard that allows users to immediately assess their health risk profile. Conditions are grouped in columns according to predicted lifetime risk (<1%, 1-10%, etc.). Orange boxes denote conditions of higher than average risk to the individual.
Running the Bases
During 2008, I signed up for personal genomics services from four providers: deCODE, Navigenics, 23andMe and Houston-based SeqWright. The firms’ philosophies may vary but the procedures are similar. The consumer’s DNA is surveyed on a commercial or customized Affymetrix or Illumina DNA chip for 500,000 to 1 million SNPs. The results are delivered via the web and (with one exception) regularly updated as new findings are published. These services are now as cheap as $399, thanks to 23andMe’s recent price cut.
Navigenics: Health Compass
At $2500, the Navigenics service is the most expensive whole-genome SNP service on the market—excluding Knome’s six-figure sequencing service. Navigenics is working closely with the medical community and does not consider itself simply a direct-to-consumer offering. For example, it is collaborating with Scripps, Affymetrix, and Microsoft in a 20-year study to determine if personal genomic testing encourages participants to make beneficial lifestyle changes.
Last Spring, I visited Navigenics headquarters in Mountain View, Calif., to discuss my results with Elissa Levin, the director of the company’s in-house, on-call team of genetic counselors. Navigenics is the only company that offers this direct resource. “I think people would benefit from having some expert consultation,” of their genome scans, says Green, noting that the family physician isn’t necessarily qualified for that role.
There’s nothing recreational about the Navigenics Health Compass: it won’t tell me whether I can smell asparagus in my urine or whether my maternal ancestors hail from Doggerland. The focus is strictly on medical conditions—in particular, actionable conditions for which Navigenics believes there is a beneficial medical or lifestyle intervention. “If we can detect Alzheimer’s at age 21, we open up 50 years—I call it the therapeutic window,” explains co-founder and CSO Dietrich Stephan. “As opposed to presenting when half your brain has been melted by disease, you present when you’re totally cognitively normal.”
The Health Compass currently presents information on 21 diseases. The results are presented clearly and intuitively: conditions are arrayed in columns according to estimated lifetime risk, from less than 1 percent to above 50 percent. This reminds clients that their lifetime risk of a heart attack, although perhaps less than the general population, is still greater than say restless leg syndrome, even if their individual risk for that disease is increased. Disease boxes shaded orange indicate a significantly elevated individual risk. I quickly survey only five orange boxes—a good omen perhaps.
The single most predictive result is for AD. For the 3 percent of the population that carries two copies of the ApoE4 allele, the risk of AD is increased ~ 15-fold, and my curiosity is heightened because the disease stole my great aunt Gwyneth. The silver panel puts my lifetime risk at 4.5 percent—half the national average.
There is similarly reassuring news for lupus, age-related macular degeneration, multiple sclerosis and type 2 diabetes. For each disease, I can learn my precise genotype and a wealth of medical information. I can also print out a thick report to take to my physician. But some conditions do bear further inspection: my risk of heart disease is no better than the national average, about 40 percent, and my risk of prostate cancer is slightly elevated. In most cases, these risks are calculated on just a few known genes, with the likelihood that many more genes remain to be detected, whether by GWAS or resequencing for rarer variants. My calculated risk will be revised as each new variant impacts the overall likelihood of that disorder.
One final entry catches my eye: obesity is in an orange box, my lifetime risk pegged at the national average of 34 percent, because I carry a risk allele for the “fatso” gene, FTO. At 6’2” and barely 165 pounds, with a limited exercise regimen consisting of refereeing a few high-school soccer matches, that isn’t something I will worry about.
While deCODE Genetics is best known as a drug development and diagnostics company, Kari Stefansson says “The original business proposal was always to use genetics for preventative health care. The idea of marketing direct to consumer is something we’ve been toying with for several years.”
Unlike its competitors, deCODE is driving much of the original research in complex traits (see sidebar: “Assessing SNP Associations”). Stefansson says Navigenics and 23andMe are basically “dot com companies… They are mostly marketing discoveries that we have made.” In October, deCODE scientists published two SNPs associated with skin cancer (basal cell carcinoma), and introduced those SNPs into the deCODEme service. But any advantage was short-lived: within 24 hours, 23andMe reminded its users that they could search for the same SNPs in their raw data. Although competitors can use deCODE’s diagnostic markers “with impunity,” Stefansson warns “that is not going to last long. Pretty soon, these patents are going to issue.”
Assessing SNP Associations
Before including a disease SNP in their consumer genomics offerings, 23 et al. employ teams of very smart geneticists and bioinformaticians to assess the merits of every peer-reviewed association. Many of these papers—and the SNPs genotyped as a result—stem from deCODE’s ground-breaking research on complex diseases. “We have isolated 50-70% of the new discoveries,” in the past two years, says Stefansson, who ticks off precedents in type 2 diabetes, heart attack, and prostate cancer, to name a few. “We aren’t marketing tests unless they have a greater effect on the risk of disease than elevated cholesterol.”
At Navigenics, Michele Cargill, who has the distinction of having worked with both Eric Lander and Craig Venter, leads a group of more than half-a-dozen geneticists who weigh every reported gene association in the top-tier journals. Not only must the association pass muster statistically, the disease must be actionable. If a robust gene association for Lou Gehrig’s disease (ALS) should materialize next month, Navigenics won’t include it unless it believes there is a genuine medical or lifestyle change that can benefit the individual.
- Stephan says his colleagues perform a “decision heuristic,” a laundry list of criteria that includes:
- The genome-wide association study must include at least 500,000 evenly spaced SNPs.
- The study must be replicated in at least two independent populations.
- The report must be controlled for ethnicity, gender, age, etc.
- Ideally there is some functional evidence surrounding the association.
At 23andMe, a similarly qualified team of geneticists, many of them Ph.D.s formerly with Stanford, review papers as they appear in press. The Gene Journal features many more traits and conditions, but like newspaper movie reviews, each condition is rated one to four stars. Only a four-star rating is deemed “established research.” In one case, a result published in Science earned a single star.
Most of my own results across the three major platforms that offer aggregate likelihood risks were in close agreement, but there were some intriguing disagreements. These could arise for a number of reasons, including different platforms (and SNPs), statistical programs and criteria. For example:
- 23andMe reports a 50% increased risk for age-related macular degeneration based on three SNPs, but Navigenics and deCODEme indicate a lower than average risk.
- Navigenics reports a reduced risk of rheumatoid arthritis, contrary to deCODE and 23andMe.
- There is a wide discrepancy between the reported lifetime risks of abdominal aneurysm between deCODEme (17%) and Navigenics (3.1%).
The deCODEme service offers information on more than 30 conditions. It includes information on ancestry and some personal traits, including hair and eye color and—unfortunately in my case—baldness. More surprisingly, I’m an asymptomatic carrier for hemochromatosis (ironic, as I’d published the paper on the cloning of that gene a decade ago). There’s also a tool to compare my genome to dozens of people of different ethnic ancestry. This tells me I am genetically more similar to Craig Venter than, say, an Mbuti pygmy—a feature some consumers might deem “recreational,” others not so much. A more useful tool is deCODE’s gorgeous java-fueled genome browser: using this, I can pinpoint all of my SNPs matched against the reference genome, zooming into the raw sequence if desired.
23andMe: Gene Journal
The publicity surrounding character traits earned 23andMe a “frivolous” tag in some quarters that Avey and Wojcicki reject. “We really applaud the fact that people are interested in their genomes for reasons other than just their health,” says Avey. “The ancestry work we do—I wouldn’t call that frivolous at all.”
Not surprisingly, 23andMe, co-founded by Linda Avey and Anne Wojcicki, has enjoyed lavish media attention for the past 12 months. That’s attributable to the Google connection, the celebrity spit parties, and non-stop appearances from the Today Show and New York Fashion Week to the World Economic Forum and even the Bio-IT World Expo (see, “23andWe,” Bio-IT World, June 2008). The company provides the deepest dive into your genome data, presenting information on ancestry, traits (earwax, heroin addiction), carrier status for Mendelian disorders, and risks for common diseases.
In September 2008, 23andMe slashed the price of its service from $999 to a mere $399, using a second-generation Illumina platform. The Gene Journal assigns one to four stars for more than 90 conditions, based on the firm’s internal assessment of the validity of the original research. I learn, for example, that I’m not a carrier for the most common cystic fibrosis mutation. But there are differences with reports from other services—for example, 23andMe calculates I have an increased risk of macular degeneration.
||23andMe, co-founded by Linda Avey and Anne Wojcicki.
The medical reach of the Gene Journal is illustrated not only by Brin but also by his wife. Wojcicki has revealed that she is a carrier for Bloom’s syndrome, a recessive disease that, like Tay Sachs and a dozen or so other conditions, is more frequent among people of Ashkenazi Jewish descent. For Wojcicki and her husband, such information is anything but recreational: they are expecting their first child.
The ability to search the full SNP dataset is also valuable. For instance, a recent New England Journal report identified a SNP that predisposes towards statin-associated myopathy. It was trivial to key in the identity of that SNP and learn I did not carry that allele.
The ancestry painting involves a proprietary algorithm, explains senior director of research Joanna Mountain, to compare groups of SNPs against those catalogued in the HapMap project. Other services had said my genome contained traces of African or Asian genetic material, but the 23andMe report is unequivocal and consistent with my own self identity: my chromosomes are 100% blue, or European. Avey, who has the same pattern, called her chromosome painting “very boring.”
23andMe has pioneered the social networking aspect of its service, encouraging customers to compare genes to friends and family and join a community effort called “Research 2.0.”
Based in Houston, SeqWright’s Genome Profiling Service (GPS) has received scant public attention compared to its competitors since its launch in early 2008. The company, co-founded by Baylor genome center director Richard Gibbs, is a genome service contract research organization, but it already possessed a CLIA-certified microarray facility.
SeqWright’s GPS lacks some of the bells and whistles of its competitors. My report was delivered in May, but it is not dynamically updated. The report is presented as a single HTML page—mine presented 16 medical conditions. It conveys odds ratios for each tested SNP, but the GPS doesn’t offer an aggregate likelihood ratio or lifetime risk for each disease. When presented with a list of 17 SNPs for Crohn’s disease or 26 for type 2 diabetes, it’s unclear for a geneticist what to make of the information, let alone a lay person. Nor does it list the genes that contain each SNP, citations to the original literature, or the ability to dynamically search raw SNP data.
If you are not content to wait for the personal genomics companies to decide which conditions are officially presented, then Michael Cariaso’s Promethease program may be just the ticket. “The pure computer programmer in me has always wanted to write Promethease,” says Cariaso. “I want to have my own DNA as data and figure out what the hell it means.”
After downloading my raw SNP data, I ran Promethease, which annotates my genotypes based on information in SNPedia, a wiki for SNP data built by Cariaso and Greg Lennon. This reveals hints of associations with many other conditions, although this is very preliminary data. Cariaso recently demonstrated the potential of his program when he won a 23andMe-sponsored competition to predict the traits and characteristics of 23andMe’s anonymous demo account holder, based solely on a handful of SNPs.
All of the consumer genomics firms have expressed interest in shifting to whole-genome sequencing once the price is right. Stephan says Navigenics will introduce some form of exome sequencing in 2009. Knome CEO Jorge Conde says he expects to have sequenced 20 individuals by the end of 2008. The sequencing is performed by the Beijing Genomics Institute in Shenzen; clients fly to Boston for a daylong debriefing before receiving their data on a USB thumb drive.
Knowing one’s full genome sequence won’t be a godsend. There are still glaring deficiencies in our ability to interpret genome data or predict the significance of the thousands of novel SNPs and copy number variants unearthed in each new human genome. After originally being told last year he carried a mutation in the BRCA1 breast cancer gene, James Watson later learned it was nothing more than a “benign Irish polymorphism.” That may not concern an 80-year-old man, but such a misinterpretation error could have catastrophic results for a young woman with a family history of the disease.
Yet despite legitimate concerns about the validity and significance of every reported gene association, Stefansson has no doubt where this is going: “Within the next 3-5 years, pretty much every college-educated person in the U.S. is going to have a profile similar to the one found in deCODEme or 23andMe or Navigenics.”
Navigenics’ Stephan goes further. “Ultimately,” he says, “[the goal] will be: sequence the genome, put it in a big computer, push a button, and get a rank order list of things you are at risk for. It will supplant newborn screening and all molecular diagnostics.”
This will surely hasten some sort of federal regulation, which would be widely welcomed. The consumer genomics trend is “absolutely unstoppable,” says Green, who adds: “The question isn’t really, should it be happening or can we stop it? The question is to foster an incremental path along which it’s done better and better.”
And let’s not forget the environment. Two years ago, my soccer days came to an abrupt end when my face latched onto a thunderous clearance at close range, resulting in a detached retina that no genome scan could have seen coming.
Editor’s Note: The author thanks deCODE Genetics, Navigenics and SeqWright for providing complementary personal genome services.
This article appeared in Bio-IT World Magazine.
Subscriptions are free for qualifying individuals. Apply Today.