Genetics typically modify the abundance or functionality of proteins, and it is proteins not genes that mediate physiological effects. Further, proteins are modified by both genetic and environmental effects, making them closer to physiology. Linking genetics to proteins to the phenotype of interest establishes the importance and usefulness of a pathway. Our lab is interested in establishing multivariate protein panels that are predictive of physiological function. For example, in the circadian field, the inability to easily measure the phase of the central biological clock is a significant limitation. The current gold standard, a test called "DLMO", measures the time at which melatonin rises under dim light. Though effective, it requires the collection of blood, urine or saliva at multiple points over a 7-hour period. Using a large panel of 5,500 proteins and mathematical modeling, we developed a test we call “plasma-time” that can predict DLMO using a single blood sample taken anytime during the day. We are currently establishing protein panels that predict sleep deprivation (“sleep debt”) or presence of sleep disorders (narcolepsy/hypersomnia, sleep apnea, insomnia, restless leg syndrome, REM behavior disorder). These assays are being validated in clinical trials and help identify why particular patients have insomnia or sleepiness. We also plan to use these panels in large scale genetic studies.
Seventy million Americans complain of sleep problems, and thousands of sleep clinics treat patients with sleep disorders. The main sleep disorders are sleep apnea (20%), Insomnia (10%) , Restless Legs Syndrome/Periodic Leg Movements (PLM) (10%), Hypersomnia/Narcolepsy (~4%). We work on all these disorders.
To understand sleep problems and disorders, we must understand the mechanisms underlying why we are tired or have difficulty falling asleep. These mechanisms are regulated by two processes: sleep “homeostasis”, which represents a need to catch up on lost sleep (called sleep debt), and the circadian clock, a consequence of the time of day that is independent of prior sleep (which explains why we experience jet lag). One of our major goals is understanding why lack of sleep or sleep disorders make us feel tired.
To discover novel molecular mechanisms regulating the need for sleep or sleep disorders, we are currently conducting large scale genetic studies in patients with EEG-characterized sleep or sleep disorders. This is complementing efforts by the UK biobank and 23&me that are collecting genetic association data with subjectively defined sleep symptoms. To achieve this goal, tens of thousands of polysomnography (PSG) studies have been collected or curated from epidemiological samples and from our own clinical repository.
During sleep, we are immobile and undisturbed by sensory input: data can be easily captured without disturbance or loss of productivity. At the technological level, the gold standard to measure sleep is nocturnal sleep polysomnography (PSG). The PSG is a complex set of physiological signals including electroencephalogram (EEG), electroocculogram (EOG, eye movements for REM sleep), nasal and oral air-flow (to measure breathing and sleep apnea), chin electromyogram (EMG, muscle tone loss for REM sleep), electrocardiogram (ECG), right and/or left leg EMG (for RLS or PLM), snore sensor or microphone, inductance plethysmography (belts to measure respiratory efforts in sleep apnea), blood oxygen saturation (SpO2, sleep apnea), and body position. At the end of the day, or more accurately the beginning of the night, this amounts in a ton of wires.
This technology is 30 years old. With advanced sensors, it is now possible to create comprehensive, non-intrusive, minimalist home sleep systems. Although there is currently an explosion of devices measuring “sleep” through various proxies (tri-axial accelerometers, ECG, PPT, cardioballistograms, oxymetry, selected EEG leads), these devices are only predictive of normal sleep or of specific sleep disorders. The right sensors still need to be integrated into a non-invasive, comprehensive sleep system. In our lab, we are running numerous clinical trials studying and validating new devices, including less invasive EEGs, sound recordings to predict sleep apnea, and leg actigraphs measuring PLMs.
Sleep devices and sleep studies, including Polysomnography studies (PSGs), return complex signals that are often scored or analyzed by hand. One of the areas where the Mignot Lab has recently made significant progress, is in applying deep learning to sleep study signals.
PSGs are currently scored by humans who page through a full night of recorded data, in 30-second segments, to extract simple features such as sleep/REM sleep latency, sleep stage proportions, number of sleep apnea breathing events (Apnea-Hypopnea Index, AHI), and number of PLMs (PLM Index). This is time consuming and varies based on individual scorer. In some centers, semi-automatic programs now assist scoring, but these programs are limited and error prone. This type of human annotated data is ideal for deep machine learning and indeed, we have been able to create algorithms that can automatically score sleep stages, detect EEG and autonomic arousals, sleep apnea, PLM, and other currently annotated features2. These automated methods are more reliable and consistent than human annotations. More importantly, they are also creating and revealing new features that can be used for more than just PSG scoring. For example, the output our deep learning algorithm generated for sleep stage classification proved a useful diagnostic for the sleep disorder narcolepsy. Similarly, we found that subtypes of sleep apnea with differential effects on sleepiness and the cardiovascular system could be objectively identified. Deep learning of sleep PSGs can also be applied to totally new problems. For example, we have designed a method that uses our PSG model to predict age. This was subsequently used to show in epidemiological studies that older “EEG” age predicts mortality.
These models are being developed by a team of 3-5 electrical engineers and computer scientists who are part of a long-term exchange program with the Danish Technical University (Helge Sorensen: https://orbit.dtu.dk/en/persons/helge-bjarup-dissing-s%C3%B8rensen) and University of Copenhagen (Poul Jennum: https://mhd.ku.dk/behind/staff/jennum/) that has led to the publication of over 35 papers in the last 5 years (see: https://www.mignotlab.com/collaborations/). Our end goal is linking sleep study and device data with genetics (genome wide association), proteomics (blood biomarkers), and phenotypes (various disease and sleep disorder).
Mignot lab is primarily known for the study of narcolepsy, a neurological disorder affecting 1 in 2000 individuals. Narcolepsy is characterized by sleepiness and a rapid transition into Rapid Eye Movement Sleep (REM sleep). We discovered that the cause of human narcolepsy is an autoimmune loss of ~70,000 hypothalamic orexin/hypocretin neurons, leading to deficient neurotransmission. This is currently leading to the clinical development of orexin antagonists as sleeping pills, and orexin agonists as new therapies for narcolepsy: clinical trials are being run by our lab.
In our lengthly study of narcolepsy, we were the first to positionally clone a disease gene in dogs (HCRTR2 mutations causing canine narcolepsy) and the first to identify HLA-DQ0602, T-cell receptor gene variants, and other immune polymorphisms as major genetic factors in human narcolepsy. We also found that narcolepsy is triggered by molecular mimicry between specific influenza strains and orexin sequences; this initiates an autoimmune process that destroys orexin neurons. This research is increasing our understanding of how the immune system can be dysregulated through a combination of genetic and environmental factors and attack neurons. Our next goal is to link every identified genetic factor to a specific step in the pathophysiological process.
An extension of this work is in the field of hypersomnia. Many patients who are sleepy do not have narcolepsy. Instead, as mentioned above, they have dysregulated circadian or homeostatic regulation of sleep. Interestingly, genetic studies of subjects who report needing a lot of sleep show an overlap between hypersomnia and mood disorders, notably bipolar. Another specialty of the center is the study of Kleine-Levin Syndrome (KLS), a rare disorder characterized by severe episodic hypersomnia with cognitive impairment accompanied by apathy or disinhibition. Using genetic studies in over 600 cases, we found that KLS is associated with birth difficulties and genetic variants in the TRANK1 gene locus (a gene associated with bipolar disorders). We are continuing to study these disorders through a combination of genetic and proteomic biomarker studies.
Work in narcolepsy has convinced us of the value of establishing causality through genetic approaches. As we continued our research, we also realized that numerous brain autoimmune diseases are left unstudied. Our lab started to collaborate with Professor Honnorat in Lyon, a neurologist focused on neuroimmune disorders that are sometimes associated with cancer (paraneoplastic syndromes). A big advantage of studying paraneoplastic syndromes is that the primary causal autoantigen of these conditions is often known (because it is also tumoral). With Dr. Honnorat, we started to survey these conditions using HLA typing and GWAS, discovering, for example, an effect of Killer Immunoglobulin-like Receptor (KIR) and Immunoglobulin (IG) genetics in anti-NMDA encephalitis (one of the most common paraneoplastic disorders). This discovery is leading us to study Natural Killer (NK) cell immunity, a nascent area of research. Like narcolepsy with DQ0602, a number of paraneoplastic conditions have monoallelic HLA associations, for example DQ0501 in anti Iglon5, or DR0701 in LGl1; this suggests that these conditions also have HLA-T cell interaction effects, facilitating the application of our narcolepsy research to these disorders. Other neuroimmune disorders have links to neurodegeneration, which is leading us to study the protective mechanisms of the adaptive immune system. The study of these conditions will link pathologies to specific immune cell subsets, further guiding pathophysiological work and therapies in this area, and enhancing our understanding of CNS and tumor immunity. Work here involves genetic studies, tetramer studies, and single cell transcriptomics.
Although our laboratory is primarily conducting basic pathophysiological studies, we have always had a strong interest in clinical translational studies. As mentioned earlier, our work in narcolepsy has led to new diagnostic procedures for narcolepsy (machine learning algorithm to diagnose narcolepsy using PSGs3, measuring hypocretin levels in the CSF for diagnosing narcolepsy) and new therapies for sleep disorders (hypocretin drugs for insomnia, circadian disorder or narcolepsy). We are currently conducting over 30 clinical trial studies involving the study of drugs, aiming at establishing novel biomarkers, and/or studying various devices such as smart watches, EEG bands. These studies are conducted by a team of experienced patient coordinators at the sleep clinic.
The website was designed by Samuel Mignot. Illustrations are by Dani Darling.