Adventures in Recreational Genomics

Part 2: The Prediction Controversy

There used to be so much more than ancestry to delve into on 23andMe. Until November 2013, when the US Food and Drug Administration stepped in, 23andMe could analyze your data to make medical predictions based on published research, and pinpoint particular mutations in genes that are associated with certain conditions.

23andMe reported your risk of getting 120 diseases versus the average risk for those with European ancestry. They told you what medical conditions you’d inherited, and what physical traits you possess. They predicted your response to important drugs, and told you the statistical confidence of each prediction all while showing you the science behind it their findings.

I was fortunate enough to sign up with 23andMe just before the FDA restricted access. (Note: Even though they don’t report it, all of your raw data is still there, and you can upload it to other sites to get the analysis…we’ll get to that later.)

The most controversial report had to be Health Risks, and mine was a doozy. They said I had nearly triple the risk of the average European-type-person of getting a rare eye disease called Exfoliation Glaucoma, double the average risk of getting Psoriasis, 45 percent increased risk of late-onset Alzheimer’s, and anywhere from 20-to-40 percent increased risk of Parkinson’s, Scleroderma, Restless Leg Syndrome, Multiple Sclerosis, and Esophageal and Colon Cancer. And it was all backed up by high statistical confidence…YIPES!

My Future: Wait, where's the tall, dark, and handsome stranger?

But what does this all mean, and how does 23andMe make these risk calculations? How soon before I am itching, kicking off the covers, blind and demented? Here is where a little science lesson is in order, and then of course a look inside the tiff between 23andMe and the FDA.

The Hard Science

Let’s look first at my heartbreak-of-psoriasis forecast, a condition, which, by the way, I don’t have. 23andMe cites data from three studies where scientists found mutations in regions of the gene associated with skin disorders. A mutation is simply a mistake, or variant, in the normal order of the chemicals that make up a region of DNA. The chemicals, or nucleotides, are called Adenine, Cytosine, Guanine and Thymine, or A, C, G, and T, and in a normal gene, the chemicals follow a known, common order.

23andMe measures three regions with known mutations. These sections of the chromosome provide instructions to make proteins used in the skin and also used by the immune system to recognize invaders. The regions are known to have specific variants, where one chemical letter in one position can differ from the norm. A, G, C, and T switch-outs are not uncommon, swaps there correlate with psoriasis, where the immune system attacks the skin. I have these mutations.

I studied the science this prediction is based on in my Technical Report. It shows the HLA-C gene, where I have the version where the normal letter has been replaced with a T at one spot. This one mistaken substitution causes the immune system to identify skin as foreign. If you really are a data-geek, you can link directly to the original research that identified the mutations, exploring the Citations at the bottom of the report.

Part of my Technical Report

23andMe used to tell how much the genes could contribute to your disorder. For example, my triple-the-risk Exfoliation Glaucoma is only 13 percent heritable, meaning that environmental factors, such as having diabetes or high blood pressure, are more likely to cause the condition than having the gene. On the other hand, it says Scandinavians are more likely to get the condition. So I may still be at risk for this condition in my senior years. (Oh wait…I am a senior.)

In the Drug Response report, I learned about my body’s inherited responses to specific drugs. I metabolize certain antacids rapidly, meaning I should take a larger than normal dose when I get indigestion. I am sensitive to Coumadin, so should I need to take a blood thinner, I would need a lower dose than normal. (Note: I had no idea what most of the other drugs on the list were.)

Some of my predicted responses to drugs.

Epigenetics to the Rescue  

Now, the big question, why don’t I have any of the diseases on the list? This is where our adventure into genomics becomes less recreational. For the answer, I go back to Dr. Sapolsky’s lectures. Simply put—and this is the newest and most remarkable discovery of genetics research— your 20,500 or so genes are really two, two-mints-in-one. They are the fixed blueprint of instructional code you received from your parents, mutations and all. But your genes are also malleable software your body uses to survive and fight disease, and they can be turned on, off, or modified at any time, which changes the instructions.

We’ve known for decades that any time a cell divides in the body mistakes in the copying of chromosomes are made through random error. Exposure to radiation or toxic chemicals can also alter the DNA. These changes are called “point mutations,” and can cause disease or changes in physical traits. Some point mutations don’t have any harmful effect. And of course, when random copy errors occur in sperm and egg, those mutations are passed down the line and accumulate in the human genome. This is how evolution occurs, when random mutations produce an effect that makes an individual’s survival more likely.

But there’s more, there are “epigenetic” modifications, those that happen “outside the gene.” Just because you have a certain sequence of chemical letters in one gene, whether common or mutated, does not mean that gene is turned on and cranking out the right (or wrong) protein. (Note: There are some important exceptions, like Huntington’s disease and hemophilia.)  Other factors cause genes to be turned on or off, or even dimmed like with a dimmer switch. Or genes can be altered to do something other than what they are supposed to do, and can even move around on a chromosome. Our genetic software is on the go, constantly being upgraded, downgraded or hacked. And then, as these newly modified cells divide, their updated genetic code is passed on to the new cells in that part of the body.

Epigenetic modifications come in many forms and at different times. We’ve known about one kind of epigenetic control over our DNA’s code: it appears during early cellular differentiation. Every cell in our body has the same code, the complete instruction manual to build an entire body, but some cells become muscle, some become blood, some kidneys, and stay that way. Entire stretches of DNA have to be shut down so that your finger remains a finger throughout your lifetime, and doesn’t turn into a nose when the cells divide. Clearly something is telling each cell which chapter of the full instruction manual to follow, and which to ignore.

But our original genetic blueprint can change in other ways. Scientists now know that more than just protein coding is going on: only one percent of DNA actually codes for proteins, the building blocks of the body. These coding sections are called “Exons.”  The remaining 99 percent used to be labeled “Junk DNA,” but we now know that these vast areas include important regulatory genes and punctuation. These are things like “Enhancers,” “Promoters,” “Amplifiers,” “Silencers,” and “Start Sequence” or “Stop Sequence” instructions, all of which control gene expression. They make up an estimated 5% of our DNA, but profoundly affect what the other genes on the chromosome do. The rest of the DNA are non-coding genes called “Introns.

Regulatory DNA precedes the Exons, the actual coding sections, and controls what they do.

What’s important is that mutations in the regulatory DNA can affect the protein-coding Exons. For example, if there’s a mutation that has inadvertently created a new “Start Sequence” or “Stop Sequence” command in the cell’s protein-making machinery, a different protein could be manufactured.

But there’s more. There are regulatory proteins floating around the gene called “transcription factors.”  These bind to the gene to tell it how to transcribe the DNA into RNA. And other chemicals affect the gene. One important type is called a methyl group (an organic compound consisting of one carbon and three hydrogen atoms), which can attach to the chromosome structure, the chromatin, causing it to loosen or tighten its grip on the DNA. The tightening and loosening changes, activates or silences genes.

Our original genetic instructions look like they have less and less significance! Scientists worldwide are collaborating to investigate the structure and function of these non-coding and regulatory elements. The National Institutes of Health is funding the Encyclopedia of DNA Elements (ENCODE), which is a follow-up to the Human Genome Project involving 400 international scientists. Last year, my friend Kate completed her work on the browser for ENCODE.

What fascinates me is the newest, broadest understanding of epigenetics—how factors outside the body alter our genetic blueprint. Researchers have discovered that diet, stress, meditation, exercise, drugs, chemical exposure, social position, and even maternal attention and love, can initiate chemical changes in the body that cascade down to the chromosomal level, altering how your body functions, for good or for bad.

So, because of something not written in the genes, in what we used to call “nurture” or “the environment,” the body changes at the chromosomal level. For example, genes that direct the creation of neurotransmitters in the brain can be hacked by increased levels of stress to make us even more anxious or depressed. Genes that normally suppress tumors might get dialed down by an attached methyl group, ultimately causing cancer. Genes that code for pigmentation in animal fur can be altered by diet, changing fur from light to dark. We are definitely much, much more than the traits we inherited from our parents.

And there is now tantalizing scientific evidence that epigenetic chromosomal alterations in our body’s cells can be passed on to future generations, evidence that these changes make their way to the DNA in our egg and sperm cells. This is a radical notion. Last year researchers learned that stress felt by Dad (a daddy mouse)—whether as a preadolescent or adult—leaves a lasting impression on his sperm that is passed on, giving sons and daughters increased anxiety and depression. Genetic scientists are re-examining the discredited theory that physiological characteristics modified during one’s lifetime may be passed on to future generations.

What Me Worry?

All of this leads to why I’m not worried that my eyes will exfoliate any time soon, and partially explains why the FDA told 23andMe to stop providing Health Risk and Technical Reports. Stick with me, as our journey becomes even more intellectually demanding!

First, we know that on average only 20-to-30 percent of most traits, diseases or characteristics are caused by the genetic code. Take for example the two BRCA gene mutations, which are associated with breast and ovarian cancers. Only 5-to-10 percent of all breast cancers and 15 percent of ovarian cancers are caused by mutations in this gene. Even if a woman’s breast cancer is known to be hereditary, only 20 to 25% of those cancers can be causally linked to the BRCA mutations. So, unfortunately, just because you dont have a specific gene or mutation doesn’t mean you won’t get the disease.

Furthermore, many diseases or conditions are caused by the combined effects of many genes, not just one mutation in one location. Much work is being done to identify these groupings of mutations.

Conversely, if you have a genetic mutation, we know that epigenetic factors affect whether the problematic instruction will be implemented. Recently, for example, a study found that meditation down-regulates genes that normally lead to inflammation, and results in faster recovery from the negative effects of the hormone cortisol in the body. And some epigenetically caused mutations appear to be reversible.

Science is even discovering good mutations that trump bad ones. For example, a rare mutation was recently identified that protects against Alzheimer’s. It’s called A673T, where a T replaces the A in one spot. For those who have the APOE4 mutation that increases the risk of Alzheimer’s, which I have from one parent, possessing the A673T would provide strong protection. 23andMe tells you if you have this mutation.

Another issue involves what these consumer-sequencing companies are actually testing. These services analyze SNPs, or Single Nucleotide Polymorphisms, where a single chemical letter of a gene sequence is known to differ from one person to the next. They are the inherited sections of genes where mutations, also called variants, are common throughout the population. There are an estimated 10 million SNPs in total; 23andMe has selected 610,000 SNPs for analysis. Reading them is inexpensive; 23andMe uses basic chips designed by the company Illumina.

SNPs are fragments of genes that are easy to read and give us a top-line understanding of mutations we possess.

But all the SNPs in the world still represent only a tiny fraction of our full genetic code. The human genome has three billion base pairs, the string of nucleotides A, G, C, and T three billion long, with another three billion matched to them on the other side of the double helix. Understanding the entire code is the Holy Grail of genomic science. Merely reading the full code requires very advanced technology, has, until recently, cost up to 100 times as much as reading selected SNPs, requires vast amounts of server space to store the code, and needs the services of skilled technicians and geneticists to interpret.

More instructive than obsessing over your individual SNP variants would be to overlay them with those of your family. If you, your parents, siblings, and cousins share mutations and some of them have diseases or conditions known to be heritable, then your own variants are more predictive. In reality, simply giving your doctor your family health history is more likely to be of use in disease prediction than studying your SNPs.

Furthermore, it would be more useful to examine the Exome, rather than SNPs. Remember that only one percent of DNA is thought to code for proteins, in the segments called Exons. There are 180,000 Exons in the Exome. Variations here more directly correlate with disease and conditions. Only in the last few years has identifying and sequencing Exons become possible with far more sophisticated equipment than what consumer-genomic companies use.

So what else is wrong with using SNPs to predict disease? ALERT: more genetics lessons ahead!

The vast majority of SNPs used by 23andMe, or any genetic sequencing platform, turn out to be simply random mutations spread through the population through genetic drift. These are variants passed down and accumulating through the generations, but with little or no direct health impact. And for most variants, we don’t know what they affect. In fact, only 30,000 of the 610,000 SNPs measured by 23andMe are associated with diseases and conditions, although research is rapidly accumulating on new correlations. And 23andMe recently customized its Illumina sequencing chips to detect fewer but more diagnostic SNPs.

Another problem is that different consumer gene testing services look at different SNPs. In a recent article in the New York Times, the author submitted her saliva samples to three companies: 23andMe, Genetic Testing Laboratories, and Pathway Genomics. Her results varied widely: 23andMe said she had an increased risk of psoriasis, like me, and the other two companies said she had a reduced risk. Same for Type 2 Diabetes, coronary heart disease, and rheumatoid arthritis—her predictions were all over the place. Why? Different sets of SNPs to assess each disease.

Another shortcoming has to do with how risk is defined. Is a “20 percent increased risk” of a disease low or high? Certainly a 20 percent increased risk of something that occurs only one percent of the time is a lot less scary than a 20 percent increased risk of something half of all people get. And then, a 20 percent risk compared to what group? Only Northern Europeans? Men and women combined? And how are these risk percentages calculated? How big must a database be for assigning probabilities? Is the 15,000-people database of asthmatic 23andMe customers really diverse enough and large enough to draw conclusions about genetic causation? Some scientists would say no. There are no standards used by the industry to define these metrics.

According to some professionals, knowing our genetic code rarely helps us. “A small percentage of people who get tested will get useful information,” said Dr. Robert Klitzman of Columbia University’s Bioethics Program. “For most people, the results are not clinically useful, and they may be misleading or confusing.”  Why? Because there are only 23 diseases for which a) a highly predictive genetic test exists, and b) the disease starts in adulthood, and c) the disease can be treated. Even the American College of Genetics and Genomics admonishes that any test result other than for the 23 diseases is useless.

The FDA Took Our Toys Away

But I only learned these shortcomings after I submitted my spit sample. And I just got in under the wire. Two weeks after I received my results, the FDA told 23andMe to stop analyzing data other than for ancestry. In the FDA’s mind, the company’s service needs regulatory approval. They claimed it was intended for use in the “diagnosis of diseases,” for “cure, treatment or prevention,” providing “steps to take towards mitigating serious diseases.”   Products like that require examination and approval before marketing to the public.

And there had been complaints. Female users learned that they had a SNP that “doubled their risk” of breast cancer and were terrified that they would need to have their breasts removed. Consumers didn’t understand why their doctor’s genetic test said they were not at serious risk of a disease while 23andMe predicted the opposite. People with APOE4 Alzheimer’s mutations were freaked out. A class action lawsuit was filed in California claiming that the company’s ads were misleading, and that the test results “not supported by any scientific evidence.”

Huh? Didn’t the FDA see that the site is plastered with disclaimers? Didn’t the California consumers read the reports that show the actual research findings? Isn’t linking disease to SNPs and genes what the entire field of genomics is about? The regulatory reaction seemed extreme.

23andMe makes it very clear that our ability to interpret genetic data is limited. They explain what SNPs are, and how they selected them. Customers must sign a lengthy disclaimer explaining that the disease-risk estimates are only rough estimates and are made using still-developing science. The company explains that your risks are based on comparisons within a European-origin-only database. And they make it clear that they are not offering any medical diagnosis. To be fair, the science is not easy to understand, but based on some of the comments I read in the 23andMe website discussions, many customers had made no effort to understand what their results mean.

It was said that the execs at 23andMe initially blew off the FDA’s many requests for product information. The company’s in-house attorney tasked with dealing with the FDA apparently had no experience with medical device regulation.

Clearly, heavy advertising and the $99 price—both of which motivated me to buy—were the last straws for the FDA. But some argue that big pharma is in on it. After all, Myriad Genetics charges $4,000 for the BRCA tests, and Ambry rakes in $3,000 for just two mutations. Physicians might feel their turf is threatened. One of my docs pooh-pooh’d my 23andMe findings, but then suggested I follow up with a test to explore one prediction.

You can make an end-run around the FDA’s restrictions. 23andMe still provides your raw data, the strings of A, G, C, and T’s for your half-million-plus SNPs. There are several sites, among themInterpretome, Promethiase, and GeneticGenie, where you upload your raw information for analysis. A warning however, it’s complicated to use. Or you can submit your sample from Canada or the U.K.

Here’s how I used my results in another genetic engine, One of my doctors asked if I had a common mutation that affects how the body metabolizes B vitamins and ultimately makes neurotransmitters like serotonin and dopamine. By uploading my raw data to this engine, I learned that I have variants that reduce serotonin production during a metabolic process called methylation. I’ve got two mutations, in fact, each with the wrong nucleotides, from both my mom and dad, that code for an important enzyme. How to fix? Go to Whole Foods and buy methylfolate supplements, which my body does not produce effectively from the B vitamins in my food. Result: more serotonin, be happy!

My Report
I have non-standard nucleotides in two genes (in red) that code for enzymes that help make serotonin.

Despite the FDA’s restraints, there is no stopping advances in our search for the meaning of our genetic code. This is the frontier, and regulators and the medical establishment may just not be prepared. In Part Three, we’ll look at the latest developments in genomic analysis, and consider if it is wise to peer into one’s chromosomes—remember what happened when Pandora opened her box?

(Start the journey from the beginning with Adventures in Recreational Genomics Part 1)