5

Therapeutic Development and Biomarkers for Sleep Disorders

Sleep disorders are heterogeneous, ranging from more common disorders, such as insomnia, to rare disorders, such as rapid eye movement (REM) behavior disorder or idiopathic hypersomnia, for which comparatively little is known about the underlying pathophysiology, said Andrew Varga, associate professor in the Integrative Sleep Center at the Icahn School of Medicine at Mt. Sinai. Moreover, he said, there are limited treatment options for sleep disorders and those that are available predominantly address symptomatology rather than pathophysiology. Nathaniel Watson added that while there are effective treatments for insomnia, stigma against them prevents many people from achieving their positive benefits.

To address this conundrum, “Collaborative, cross-disciplinary research can help us forge empirical links in causal chains that go from genes to diagnosis and can lead to mechanistically guided treatments,” said Dara

Manoach. These treatments include pharmacological and non-pharmacological treatments, such as neurostimulation and cognitive behavioral therapy, ideally tailored to provide the most effective treatment based on circuit dysfunction or a marker that identifies the nature of the problem, said Andrew Krystal.

PHARMACOTHERAPY FOR INSOMNIA

Although CBT-I is recommended by clinical guidelines as first line therapy for insomnia, limitations in terms of access, cost, adherence, availability of trained therapists, and effectiveness for patients with objective short sleep means that other therapeutic options including pharmacotherapy are also needed, said Margaret Moline.

In the United States, the antidepressant trazodone is the most commonly used drug prescribed for insomnia, said Moline, despite the fact that it is not approved for this indication. The other major drugs used to treat insomnia are benzodiazepine receptor agonists (BzRAs, or z-drugs), such as zolpidem, eszopiclone, temazepam, and triazolam, a new class of drugs called dual orexin receptor agonists (DORAs), melatonin agonists (e.g., ramelteon), and doxepin (a low dose tricyclic antidepressant). The z-drugs and DORAs work through different neurotransmitter systems: z-drugs through GABA and DORAs through orexin. “So [there are] fundamentally different ways of treating insomnia,” she said, noting that BzRAs increase sleep drive while DORAs decrease wakefulness.

All of these drugs—with the exception of trazodone, doxepin, and ramelteon—are “scheduled” drugs, which means they have been identified as having some abuse potential. Different factors go into the decision making for scheduling, and in some cases, there are very limited data available, said Moline (see Table 5-1). Trazodone was approved years before guidelines for assessing abuse potential were established, said Moline, adding that it has never been tested for insomnia in large-scale, well-controlled studies.

Other safety concerns for z-drugs include tolerance, dependence, rebound insomnia, daytime sedation, fall risk, potential respiratory depression, memory and cognitive difficulties, and aberrant nocturnal behavior. They require gradual discontinuation and are not recommended for elderly patients. The most common safety concerns for DORAs include somnolence, headache, nightmares, and abnormal dreams, she said.

TABLE 5-1 Abuse Potential of Drugs for Insomnia

NOTES: The Food and Drug Administration uses different types of data to determine abuse potential. Both benzodiazepine receptor agonists (BzRAs) and dual orexin receptor agonists (DORAs) have been designated as schedule IV drugs, although BzRAs have a higher misuse potential.

SOURCE: Presented by Margaret Moline on November 3, 2022.

Doug Williamson, who has a long history of working in the pharmaceutical industry, suggested that DORAs are prescribed less often because they do not provide patients with the effects they desire, namely, the heavy sedation they may want, in addition to their cost. “I think that stems from an underappreciation of the importance of sleep quality and the lack of insight people have into their sleep quality,” he said. For example, 5HT2A antagonists have profound effects on sleep quality, increase total sleep time, and reduce waking after onset, but patient rating scales used by the Food and Drug Administration (FDA) fail to pick up these effects, noted Williamson.

PHARMACOTHERAPY FOR NARCOLEPSY

To treat the sleepiness associated with narcolepsy, Scammell said common treatments include amphetamines, modafinil, and oxybates, along with newer agents including pitolisant and solriamfetol. Cataplexy responds well to antidepressants because they suppress REM sleep. However, even at optimal doses of these medications, many patients still fall asleep easily,

have difficulty focusing, and have cataplexy, he said. “We’re helping them, but we’re not helping them enough.” said Scammell, and by understanding the circuits that underlie sleepiness and cataplexy, more targeted therapies should be possible.

Restoring orexin signaling is one promising approach to the treatment of narcolepsy, he added. An orexin antagonist was approved for the treatment of insomnia in 2014, and the first orexin receptor agonist, YNT-185, was synthesized in 2015, with proof of concept for treating narcolepsy type 1 established in 2017, said Hao Wang (Irukayama-Tomobe et al., 2017). Takeda developed an intravenous orexin-2 receptor agonist, TAK-925, and showed in a narcolepsy mouse model that it eliminated frequent sleep/wake transitions, restored wakefulness, and eliminated cataplexy, said Wang (Evans et al., 2022). She added that multiple oral orexin-2 receptor agonists are currently in development by multiple companies.

In clinical studies, three efficacy endpoints are commonly used, said Wang. The maintenance of wakefulness test (MWT) evaluates through self-report a trial participant’s ability to stay awake in a dimly lit room; the Epworth Sleepiness Scale (ESS) is a subjective report on the likelihood of falling asleep under various circumstances; and a Weekly Cataplexy Rate (WCR) quantifies, again through self-report, the number of episodes experienced by a trial participant.

In a double-blind, placebo-controlled crossover study, TAK-925 at two doses was shown to significantly improve the ESS score, increase sleep latency in the MWT test, and reduce weekly cataplexy, all in relation to healthy controls, said Wang (Evans et al., 2022).

Wang noted that to accurately capture the impact of narcolepsy on patients’ lives, other tests are needed that are robust and not overly burdensome to patients. For example, patients often report “brain fog,” she said. Patient diaries and cognitive performance tests could provide data to assess this aspect of the disease, said Wang. Actigraphy and electroencephalogram (EEG) measures from wearable devices may also provide more accurate patient-relevant measures of sleep and activity, she added. However, these measures need validation and testing in diverse real-world settings and international consensus on their validity, said Wang.

A better understanding of the relationship between nighttime sleep and excessive daytime sleepiness is also needed, said Wang.

NON-PHARMACOLOGICAL STRATEGIES

In addition to pharmacological approaches to treat sleep disturbances and disorders, several workshop participants highlighted the need to improve sleep hygiene through changes in behavioral and other non-pharmacological strategies.

One population group that could benefit from such strategies is shift workers, who comprise about 20 percent of the workforce, according to Nathaniel Watson. He added that shift work is associated with higher rates of prostate, breast, and colon cancer as well as sleep insufficiency, diabetes, and hypertension. How much of this is due to circadian desynchrony versus sleep insufficiency is unclear, said Louis Ptáček. However, Charles Czeisler at Harvard Medical School and others have shown that appropriate management of the light environment in the shift workspace can improve circadian alignment with hormones such as melatonin, noted Erik Herzog. Food and the timing of food intake may also provide therapeutic benefits, added John Hogenesch.

Dayna Johnson noted that individual-level interventions, such as those employing CBT-I, have shown effectiveness in treating insomnia. Tailoring these programs to specific population groups can increase engagement among those groups. For example, a randomized clinical trial of a culturally tailored, Internet-delivered CBT-I program to Black women was more effective than the non-tailored version in engaging Black participants and achieving improvements in sleep, said Johnson (Zhou et al., 2022).

Tiffany Schmidt added that individuals may also be able to improve their sleep by, for example, exposing themselves to bright light and daylight during the day and avoiding light at night.

BIOMARKERS OF SLEEP AND SLEEP DISTURBANCES

An array of biomarkers, ranging from molecular to digital, will be critical to evaluate circadian behavior in humans, said Hogenesch. Manoach added that because diagnostic categories alone do not indicate mechanisms, biomarkers are increasingly needed for development of new treatments for the sleep disturbances associated with specific disorders such as schizophrenia and autism.

Molecular Biomarkers

The first circadian blood-based biomarker, which measures the onset of melatonin production in dim light (DLMO), was identified in 1989 (Lewy and Sack, 1989), said Hogenesch, adding that DLMO can also be measured in saliva. Transcription profiling emerged in the early 2000s through the pioneering work of Hiroki Ueda (Ueda et al., 2004). Several years later, Ueda and colleagues applied this technique to human blood samples, demonstrating the ability to detect internal body time (Kasukawa et al., 2012).

Subsequently, scientists discovered that skin provides even more robust circadian biomarkers, said Hogenesch (Brown et al., 2005). “Skin is far more rhythmic than blood, and probably more rhythmic than almost all of your internal organs,” he said. Using a hybrid design that combined

longitudinal and population studies, he and his colleagues identified a set of skin-derived biomarkers that reflect human circadian rhythms.

Transcriptomes from isolated blood monocytes have also been shown to provide robust biomarkers of circadian rhythms, said Hogenesch (Wittenbrink et al., 2018), noting that algorithms built using transcriptomic data have shown good prediction values from measuring about 12 genes. Herzog added that metabolomics measures may provide more accurate and proximate measures of circadian output.

Digital Biomarkers of Circuitry Differences

Krystal has been studying clinically relevant phenotypes that are associated with different treatment responses, which may reflect circuitry differences. For example, while many short sleepers do not respond well to CBT-I, short sleepers with diminished homeostatic sleep drive when they go to sleep—meaning they feel less pressure to go to sleep—and elevated arousal when they awaken do particularly well, he said. He has identified slow wave activity and increased high-frequency activity in the non-REM EEG as markers for that phenotype. Identifying such mechanisms that underlie specific sleep problems may enable targeted therapy, said Krystal.

Computational approaches applied to sleep metrics can provide digital biomarkers for disorders, according to Aarti Sathyanarayana, assistant professor of computer science and health science at Northeastern University. Digital biomarkers include patterns and pathological signatures gleaned from digital measurements of physiology, such as an EEG or actigraphy signal, she said.

Sathyanarayana has applied this approach to study epilepsy and its relation to sleep. “Sleep is a modifiable target that has a profound effect on neurological function and CNS disorders,” said Sathyanarayana. “Epilepsy, in particular, has a very close relationship to sleep.” For many types of epilepsy, seizures occur primarily during sleep, so understanding how sleep affects the brain overall and the brain’s propensity to have a seizure is important, she added.

Sathyanarayana models brain electrodynamics using non-linear measures of complexity, including a metric called sample entropy, which estimates mathematical chaos in the signal. During sleep, sample entropy drops more significantly in an epileptic brain compared to control, which indicates that patients with epilepsy have lower chaos and lower complexity in their brain electrodynamics, said Sathyanarayana. This occurs both when the patient is awake as well as asleep, but the difference becomes more pronounced when asleep. “While sleep itself is not the biomarker for epileptogenicity, it has facilitated the identification of sample entropy as a candidate biomarker for the brain’s propensity to seize,” she said.

Sathyanarayana and colleagues developed a method using a 30 second EEG to determine whether a medication will work for a patient. They have also expanded this work to explore the complex dynamics of the brain by using topological data analysis, which enables holistic modeling of the brain as a network. Because epilepsy is so complicated, they first tried applying this technique to sleep apnea and were able to show that they could capture the diagnosis with a 30 second EEG while the patient was awake, she said.

Digital Biomarkers from Sensors, Wearables, and Other Devices

These digital EEG biomarkers, combined with behavior modification, can also change outcomes for sleep, said Sathyanarayana. Actigraphy data indicate not only when sleep is occurring, but also sleep quality. Sathyanarayana and her team have further shown, using actigraphy data from wearable sensors, that physical exertion during the day affects sleep quality at night. “We’ve built a model that uses machine learning with a novel neural network architecture to handle this low-dimensional longitudinal data, and we use the physical activity of an individual to predict whether or not they will have good sleep quality,” she said. Next, they built a system that, based on predicted sleep efficiency, gave activity recommendations to individuals to improve their sleep quality. They found that 81 percent of people who followed these recommendations had great sleep quality, while 79 percent who did not follow the recommendations continued to have poor sleep quality, said Sathyanarayana. Her team is now working on developing algorithms that train on only an individual’s personal data and are device agnostic and generalizable.

The results of this study suggest that behavioral modification can mitigate sleep problems, “and so may have the potential to mitigate central nervous system disorder exacerbation,” said Sathyanarayana. To realize the potential of this approach, she said cross-disciplinary collaboration between sleep experts and individuals with computational expertise in signal processing and machine learning is essential. Several workshop participants also noted the importance of engaging individuals with social and behavioral expertise.

Data collected remotely by sensors and wearable devices may also provide novel and more informative biomarkers of real-life sleep and sleep disorders, said Vadim Zipunnikov, associate professor of biostatistics at Johns Hopkins Bloomberg School of Public Health. Scaled up, these data could be used to define population-level references by age, gender, or in specific clinical cohort as well as patient-level normative values and longitudinal trajectories. These norms could account for day-to-day variability and dynamics, and integrate sleep data with daytime motor activity and circadian rhythms,

he said. Further, integrating these data with ecological monitoring assessments and light sensors could also provide insight into the impact of social schedules, encounters, and experiences as well as the environmental factors impacting sleep, such as day and nighttime light exposure, said Zipunnikov.

However, to realize these opportunities, Zipunnikov noted several challenges that will need to be addressed. First, he said, sleep biomarkers from wearable devices are contingent on the specific device and algorithm used. Many devices, particularly consumer-level devices, use proprietary hardware and algorithms that derive sleep biomarkers in ways that often need additional independent validation, which may prevent generalization of the results and harmonization of data across studies. Zipunnikov noted that there may be variability in how different devices define key metrics such as resting heart rate. Given this, uniform standards for validation agreed upon by all stakeholders are critically needed, he said.

For harmonization and collaboration, Zipunnikov advocated for open-source open-platform solutions. For example, he cited the challenges of integrating data from studies using polysomnography (PSG), which may have only one or a couple of nights of data, with studies using actigraphy, where data often is collected over a much longer period.

He added that the Johns Hopkins University medical system is working on developing and integrating a “bring your own device” concept, where, as an example, patient-donated historical Fitbit data are analyzed and reported to clinicians. These analyses can be used, for example, prior to elective surgery to provide clinicians with information about the functional status of the patient, enable better risk stratification, and track post-surgery recovery, said Zipunnikov. He added that his team is trying to understand how best to train clinicians to use these data and the outcomes of predictive models. As wearable devices and sensors become more ubiquitous and mainstream, he suggested there could be a course for clinicians on how to interpret the data they provide. First, he said, it will be necessary to demonstrate that incorporation of these data in clinical routines can improve long-term outcomes for patients.

Existing data repositories such as the National Sleep Research Resource discussed in Chapter 4 and large population-level studies such as the UK Biobank¹ and the All of Us study² are already collecting data that could be used to build and validate these predictive algorithms, said Zipunnikov.

___________________

¹ To learn more about the UK Biobank, go to https://www.ukbiobank.ac.uk/ (accessed January 9, 2023).

² To learn more about the All of Us Research Program, go to https://allofus.nih.gov/ (accessed January 9, 2023).

This page intentionally left blank.