Michael Snyder: watching his own diabetes begin

The situation

In 2012, Michael Snyder was chair of the Department of Genetics at Stanford University School of Medicine. By any ordinary measure he was healthy. He was 54 years old and a nonsmoker. His body mass index was 23.9 and he had no significant family history of type 2 diabetes¹². His laboratory had spent years building tools to measure genomes, transcriptomes, and proteomes at scale. He wanted to know what those tools would say about one person followed over time. He chose himself as the subject.

The design was straightforward in concept. He would combine a single genome sequence, which does not change, with repeated measurements of the molecules that do. His group called the result an integrative personal omics profile, or iPOP¹. The question was whether the combination could reveal something about health and disease that neither a genome alone nor a routine blood panel could.

What he measured

Over more than two years he gave about 20 blood samples. That was roughly one every two months while he was well and more often when he was sick². His genome was sequenced on two separate platforms. Complete Genomics read it at about 150-fold coverage and Illumina at about 120-fold. His exome, the protein-coding fraction of the genome, was sequenced again at 80 to 100-fold on three technologies¹. Coverage is the number of times each position in the genome is read. High coverage on independent platforms made it possible to check the two sequences against each other.

At each time point his team measured several molecular layers. One was the transcriptome, meaning the RNA molecules that show which genes are active. The others were the proteome (proteins), the metabolome (small molecules such as sugars and lipids), a panel of 51 cytokines, and autoantibodies. The RNA data came to 2.67 billion mapped sequencing reads covering 19,714 transcript isoforms from 12,659 genes. The proteomics tracked 6,280 proteins, of which 3,731 could be followed consistently across time points¹. The autoantibody array screened reactivity against 9,483 human proteins¹. Standard clinical measurements ran alongside, including blood glucose, glycated hemoglobin, and lipids.

The sampling schedule followed what happened to him. Day 0 of the study was the start of a human rhinovirus infection, a common cold. On Day 289 he caught respiratory syncytial virus, RSV¹. By sampling densely through both infections, the team could watch thousands of RNA, protein, and metabolite levels move during illness and settle afterward.

What the data showed

The genome came first, and it did not match the man. His sequence carried an elevated risk for type 2 diabetes, computed from 28 independent variants. Those included variants in the genes GCKR and KCNJ11¹. The sequence also indicated increased risk for high cholesterol, coronary artery disease, and basal cell carcinoma. It indicated lower risk for hypertension, obesity, and prostate cancer². The diabetes result did not fit the rest of his profile. He was lean, exercised, and had no family history. Stanford’s announcement described the predisposition as unexpected².

The blood told a different story as the study went on. His glucose was normal through the first part of the record. Shortly after the RSV infection, at Day 301, it rose and stayed elevated for months¹. Glycated hemoglobin, or HbA1c, measures the fraction of hemoglobin with sugar attached. It reflects average glucose over roughly three months. His HbA1c came back at 6.4% on Day 329 and 6.7% on Day 369. At that point his physician diagnosed type 2 diabetes¹.

The molecular series added a layer that a clinic visit would not have. The same measurements had been made before, during, and after each infection. That let the team see which pathways changed during the RSV infection and compare them with the changes that accompanied the rise in glucose¹. The paper’s authors were careful about what this proves. One person and one infection cannot establish that the virus caused the diabetes. But the timing is documented in a way that a single blood test never could be.

His lipid panel produced a third finding along the way. Triglycerides were high, at 321 mg/dL. They fell to 81 to 116 mg/dL after he started simvastatin¹².

What he did with it

He changed what he ate and how much he moved. In the paper’s words, the response was a dramatic change in diet and exercise. That meant substantially reduced sugar intake and increased exercise. He also took 81 mg of aspirin daily and ibuprofen for the first six weeks¹. His body mass index went from 23.9 at Day 0 to 21.7 at Day 511. Glucose declined gradually to about 93 mg/dL by Day 602, and HbA1c fell to 4.7%¹. Stanford’s announcement reported that these changes allowed him to avoid diabetes medication at that time². He said at publication that this was the first time anyone had used such detailed information to manage their own health proactively².

The self-study became the seed of a larger one. His laboratory’s iPOP cohort now follows roughly one hundred people over several years. It uses the same combination of genomics, transcriptomics, proteomics, metabolomics, microbiome sampling, and wearables, with a deliberate focus on people who are prediabetic³.

He also kept measuring himself, and moved part of the measurement onto his wrist. For the 2017 PLOS Biology study he wore up to seven commercial biosensors continuously for two years. They recorded heart rate, blood oxygen saturation, skin temperature, sleep, and activity, at a rate of more than 250,000 measurements a day⁴⁵. In 2015 he took a flight to Norway for a family vacation. He noticed that his blood oxygen did not recover the way it always had after previous flights and that his heart rate was higher than usual. Two weeks earlier he had helped his brother build a fence in a tick-heavy part of rural Massachusetts. He suspected Lyme disease, developed a low-grade fever, and obtained a prescription for doxycycline from a Norwegian doctor. Testing after he returned home confirmed the infection. He never developed the characteristic rash⁵⁶. In the paper, the episode appears as Days 470 to 474 of his record. Between 14% and 55% of daily heart-rate readings were flagged as outliers against his own baseline. The abnormal readings resolved the day after treatment began⁴.

His blood sugar did not stay solved. In a 2021 interview he described his glucose control as an ongoing problem. He said that his form of type 2 diabetes appears to involve impaired insulin release rather than insulin resistance. He said he now manages it with a low-carbohydrate diet, daily exercise, and medication⁷. The 2012 result was a reversal, not a cure, and he has been direct about that.

What others can take from it

Snyder had a genetics department behind him. The full set of assays he ran is still not something a person can order at a pharmacy. The treatments and diagnoses are his and his physician’s. The lessons about data are transferable.

A genome is a risk map, not a forecast. His diabetes risk variants were real. They said nothing about when, or whether, the disease would arrive. What made the risk actionable was the repeated blood work that showed it materializing¹.

Baselines have to be measured before the event. His glucose rise was visible within about twelve days of an infection because normal values already existed to compare against². The same logic drove the Lyme detection. The devices flagged deviations from his own baseline heart rate and oxygen saturation, not from a population average⁴.

Sample densely around disturbances. The richest parts of his dataset are the infections, when the team increased sampling frequency and could watch the system perturb and recover². Illness is when molecular data has the most to say. It is also when most people stop collecting it.

Tie the layers together. The genome supplied the prior. The metabolic and clinical data supplied the event. The RNA and protein series supplied the mechanism. Each layer on its own would have been either unremarkable or uninterpretable.

Be plain about what a series of one shows. The 2012 paper reports an association between a viral infection and the onset of high glucose in one person. It describes the study as a demonstration that longitudinal profiling can connect genomic information with dynamic molecular activity¹. It does not claim more, and that restraint is part of why the work has held up.

Footnotes

1. Rui Chen, George I. Mias, Jennifer Li-Pook-Than, Lihua Jiang, et al. Personal Omics Profiling Reveals Dynamic Molecular and Medical Phenotypes. Cell, 2012. https://pmc.ncbi.nlm.nih.gov/articles/PMC3341616/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵

2. Stanford Medicine News Center. Revolution in personalized medicine: First-ever integrative ‘omics’ profile lets scientist discover, track his diabetes onset. Stanford Medicine, 2012. https://med.stanford.edu/news/all-news/2012/03/revolution-in-personalized-medicine-first-ever-integrative-omics-profile-lets-scientist-discover-track-his-diabetes-onset.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

3. Snyder Lab. iPOP. Stanford Medicine, 2026. https://med.stanford.edu/snyderlab/ipop.html ↩

4. Xiao Li, Jessilyn Dunn, Denis Salins, et al. Digital Health: Tracking Physiomes and Activity Using Wearable Biosensors Reveals Useful Health-Related Information. PLOS Biology, 2017. https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.2001402 ↩ ↩² ↩³

5. Stanford Medicine News Center. Wearable sensors can tell when you are getting sick, study shows. Stanford Medicine, 2017. https://med.stanford.edu/news/all-news/2017/01/wearable-sensors-can-tell-when-you-are-getting-sick.html ↩ ↩²

6. KQED. Wearable Sensors Help Diagnose Lyme Disease in Stanford Study. KQED Future of You, 2017. https://www.kqed.org/futureofyou/319192/wearable-sensors-help-diagnose-lyme-disease-in-stanford-study ↩

7. Rhonda Patrick. Dr. Michael Snyder on Continuous Glucose Monitoring and Deep Profiling for Personalized Medicine. FoundMyFitness, 2021. https://www.foundmyfitness.com/episodes/michael-snyder ↩