Photo by redwood1

Somewhere deep in my grandmother’s veins, a blood clot breaks free. Her blood carries the clot past her heart, to her lungs, where it becomes stuck in a pulmonary artery. This is when my grandmother feels a sudden sting in her chest and loses her breath. She is suffering a pulmonary embolism.

My grandmother is tired. She wants to go home, but understands that she has to stay in the hospital for another while. Outside, the autumn sun colours the treetops a golden red. “At least the view is wonderful.” My grandmother manages a faint smile. We eat some of the strawberries that I brought, careful not to stain the sheets. Visiting hours are almost over.

As I drive back home, an old plan wrests itself free from the back of my mind. I know it is a selfish plan, driven by my own curiosity and of no immediate benefit to my grandmother. I want to have her DNA analyzed. Before time runs out.

I’ve never discussed genetic testing with my grandmother. I’m not even sure my grandmother would know what a gene is. After all, she was just an 18 year old girl in a war-torn country when Avery, MacLeod, and McCarty concluded for the first time that strands of DNA, and not proteins, carry our hereditary information. Later, when scientists cracked the genetic code and learned how to sequence genes, she was still a young woman, raising two young boys. The boys grew up and became fathers. DNA sequencing became faster, cheaper and more robust by the year. Now my grandmother has grown old and grey. Now she has a grandson with a plan.

Stomach duel

Curious to peek behind the curtains of me, I ordered a testing kit from 23andMe two years ago. I filled a tube filled with some of my finest spit, sent it back to Los Angeles and before long, I received an e-mail informing me that my results were ready. I still remember the duel held by excitement and anxiety in my stomach that evening. Here I was, just a few clicks away from my genes. What if one betrayed me, revealing itself as a harbinger of some untreatable or life-threatening disease?

I surfed to 23andMe’s website and there they were. All the SNPs that are me (SNPs – say ‘snips’ – are the one letter variations in our DNA that companies like 23andMe genotype and interpret). There were SNPs for my brown eyes, curly hair and wet earwax. And for dozens of different diseases and conditions, all numbered in red, green or grey.

Rheumatoid arthritis was red. 3.5% of the European men with my SNPs get this joint disease, compared to 2.9% on average. A significantly higher risk, yes, but not something to lose sleep over. After unlocking the report, Alzheimer’s disease turned out green. The average risk for a man of European ancestry is 7.2%, but my SNPs lower my risk to 4.9%.

There were no betrayals. Instead, their were small victories and defeats all across my genome. For every SNP that protected me from disease X, there was SNP that increased my risk for disease Y. In the disease lottery, I pulled more migraines and tremors, but less back pain and restless legs. It was hard to make sense of it all. Was I healthy, sick or something in between?

This is when I thought of my grandmother. There is no way for me to know which predictions will come true, except to sit around and wait until old age and disease come knock on my door. But for my grandmother, a predicted embolism would be old news. Her life could be the benchmark for a genetic test.

The conductor

A few weeks after my grandmother is discharged from the hospital, I look her up. She’s still short of breath. As we sip our tea, I try my best to explain her how a DNA test works. “You have to spit into a small tube. I will send that tube to a lab in the US for your, where they will isolate your DNA from the cells in your saliva that normally line the inside of your mouth.”

“And then they will know everything about my spit?” Okay. One step back. “Much more than that”, I begin. “You inherited your DNA from your parents, and they got it from their parents. I got mine partly from you, through papa. Your DNA says something about who you are on the most fundamental level. Whether you have brown or blue eyes, where you come from, but also whether you are predisposed to get cancer.” My grandmother nods thoughtfully.

A relief. She’s willing to have her DNA tested, as long as I walk her through the results once they’re in.

It is Sunday, a few months after the embolism. The frilled, purple blouse my grandmother is wearing seems too chic for the task at hand. Carefully, she spits into the tube that I brought. And again. And again. The saliva level barely rises. Between spitting, my grandmother tells stories of relatives and acquaintances I barely know. “I’m only halfway,” my grandmother sighs after half an hour.

Grandma gets there, eventually, and three weeks later the email arrives: ‘Your results are ready.’ I log in and start scrolling through the list of diseases and conditions. Colorectal cancer, 5 percent. Diabetes, 15.6 percent. Prediction after prediction, as if my grandmother has her whole life ahead of her. She hasn’t.

Click here to see what a typical 23andMe health report looks like

Armchair genetics

Armed with a printout of her test results, I go visit my grandmother. I move the wooden stool besides her comfortable armchair, so that we can run through the results together.

We start with the diseases for which my grandmother is at increased risk. Age-related macular degeneration (AMD), an eye condition in which the light-sensitive patch in the center of the retina slowly degrades, is at the top. AMD is the most common cause of blindness amongst the elderly: over 15 percent of white women over 80 in the US have it.

“It says here that with the genetic variants you carry, your risk for AMD is 9.3 percent”, I explain. “The average risk is seven percent. But your eyes are still OK, right?” “Yes, yes. But I know my father did have AMD. He even went to Nijmegen for treatment. And some of my sisters have it too. A very nasty disease.”

Next up is rheumatoid arthritis. Like me, my grandmother is unlucky when it comes to this joint disease: thanks to her SNPs, she is almost twice as likely to suffer from RA. Four out of hundred European woman get rheumatoid arthritis on average, compared to eight out of hundred women with my grandmother’s genotype. I don’t need to ask whether my grandmother has RA. I know she has it.

Onto the good news: diseases my grandmother is less likely to have. Diabetes is on the top of this list. The average risk of getting diabetes is 20.7 percent, whereas my grandmother has a chance of 15.7 percent. “But I do have diabetes”, my grandmother objects. “It’s not severe though, I don’t ever need to inject insulin.” I don’t say it out loud, but I suspect my grandmother’s lifestyle overshadows her genes here. The heritability of diabetes is only 26%, as 23andMe mentions. Environment does the rest. And truth be told, my grandmother has been overweight for as long as I remember.

We reach the end of the list. My grandmother is impressed by the predictions 23andMe has distilled from her DNA, but I’m not so sure. With the benefit of 85 years of hindsight, the predictions seem haphazard and irrelevant, a mixed bag of near hits and misses. Only when I ignore half of the health reports, and stare at the other half through eyes half closed do the contours of genetic destiny become visible.

Against all odds

Disease prediction is tough. Correlations between certain SNPs and diseases are often weak and difficult to interpret, despite what roaring headlines (‘Scientists discover genes for X’, with X ranging from diabetes to coffee addiction) would make us believe. This uncertainty, inherent to science, stands at odds with the desire of companies like 23andMe to present disease risk as clear, single digits.

Genome-wide association studies (GWAS) make up the scientific fuel on which 23andMe runs. Look past the intimidating name, and you’ll find GWAS are relatively straightforward as far as scientific studies go. Most follow a simple, three-step recipe: geneticists divide test subjects into a patient and control group, genotype their DNA and fish out the SNPs that are overrepresented in the patient group. These overrepresented SNPs contribute to the onset or progression of disease, is the assumption.

Geneticists quantify this contribution in the odds ratio: the proportion of people in the patient group with a certain SNP, divided by the proportion of people in the control group with that same SNP. Suppose 12 percent of your patients carry SNP A, compared to only 8 percent of people in the control group, then the odds ratio for SNP A is 1.5.

This is where things get interesting, and tricky. To calculate disease risk, 23andMe multiplies the odds ratios of someone’s SNPs with the average risk of disease. Ideally, the odds ratios in such a calculation should be well established and reproducible. The only problem: they’re not. On the contrary. Most are preliminary and prone to change. In a recent review of AMD genetics, the reviewer didn’t bother to list the odds ratio’s of all the SNPs that have been associated with AMD, because “these values are constantly shifting and vary based on the population that is studied as well as the (phenotypic) features of the AMD”.

As science progresses, the number of SNPs known to be associated with a certain disease is bound to increase. 23andMe’s currently calculates AMD risk using five different SNPs, yet many more have already been discovered. In a recent screening of common SNPs associated with AMD, geneticists identified almost twenty different variants.

Now I do think it is right for 23andMe to not include every SNP and odds ratio it can lay its hands on right away, because these numbers can be so fickle. But the continuous discovery of new SNPs does make clear that no prediction is final. If 23andMe decides to include additional SNPs for its calculation of prostate cancer risk tomorrow, your risk will change, for better or worse.

Problems arise elsewhere in the equation too. As a white woman over 80, my grandmother’s baseline risk for AMD should be up to at least ten, and maybe even twenty percent. 23andMe only lists the risk for women between the age of 43 and 79, which stands at 7 percent. Unfortunately, it’s not clear why data for other age ranges is lacking, or even where these figures come from. Whereas 23andMe’s geneticists meticulously cite the GWAS they have plucked odds ratios from, they remain silent when it comes to the population studies that is the source for this baseline risk. This is more than a sloppy omission. There is as much potential for error here as there is with the odds ratios. Only this time, 23andMe keeps its customers in the dark.

The contribution of environmental factors to disease risk, on the other hand, is something that 23andMe covers well. For every disease, the company lists how much of the risk can be attributed to genetics, and how much to the environment. Note: for a geneticist, the environment encompasses almost everything we do and experience that’s not genetically determined, from smoking, to physical activity and diet.

A hundred flips

But let’s just suppose for a minute that environmental factors contribute nothing to disease risk at all. And suppose that 23andMe’s prediction are infallible, and that my risk for rheumatoid arthritis really is 3.5%. In the end, whether I get the disease or not is still a chance event, no different than the flip of an unfair coin. In our lives, we have to make hundreds such flips. Most will come up in our favour. A few of them won’t. Chances are that the long list of predictions includes the disease that will be the end of me, in red, grey or green. But which one?

Two months ago my grandmother was hospitalized again. It was a stroke this time, in the middle of the night. The view isn’t as great as it was one year ago. My grandmother is asleep when I enter room 1D. She looks even more brittle than she did before. When I get home, I search through her test results until I find the health report I was looking for: ‘Typical risk of having a stroke.’

Disclaimer: this blog post is an adaptation and partial translation from an article that appeared in the Dutch newspapers (08.16.2012) and NRC Handelsblad (08.23.2012). Major parts have been deleted and rewritten. The events described took place from November 2010 to August 2012.

How does 23andMe genotype its customers?
23andMe uses a microarray chip to profile around 1 million SNPs (say ‘snips’), one letter variations in our DNA, of its customers. To put this number into perspective: geneticists of the HapMap project estimate there are around 10 million common SNPs in the entire human population.
Other types of genetic variation are not covered by 23andMe’s analysis, like differences in gene copy number. Some persons have more copies of certain genes than others, but 23andMe cannot detect the number of gene copies with its assay.
Whereas there are many companies that genotype SNPs, commercial sequencing still has to take off. Last year, 23andMe announced it would start sequencing complete exomes, the part of the genome that is translated into proteins. Back to text.