What is Sleep?/VanSavage
Notes by user Van Savage (UCLA/SFI) for What is Sleep?
1+ paragraphs on any combination of the following:
- Presentation highlights
- Open questions that came up
- How your perspective changed
- Impact on your own work
- e.g. the discussion on [A] that we are having reminds me of [B] conference/[C] initiative/[D] funding call-for-proposal/[E] research group
SFI WG Big Unanswered Sleep Questions
1. What is sleep? Behavioral definition? Phenomenological/experiential definition? EEG/physiological definition? Which metrics—fitbit, online, over what time span, Etc.? How closely matched in terms of time of onset? How closely matched in terms of functional or health effects? How do we separate sleep ability from sleep need? Do we need a species-specific definition and can we come up with a very general definition that abstractly covers everything? How strict should we be with definition of REM and SWS by voltage or time-scale and frequency characteristics across ages versus qualitatively similar behaviors? What are the building blocks of sleep? When do definitions fail? What breaks these definitions? Which questions correspond to which definitions? Can we make that mapping? Is there a universal rescaling of EEG or something else where sleep does look the same across species, life stages, etc?
2. What are the functions of sleep? Can we construct a phylogenetic tree of life and developmental stage tree of life for the function and features of sleep? Are all about evolutionary/energy optimization of processes that don’t happen as well during waking? Restoring functions vs learning functions vs other
a. Feel rested/cure sleepiness
b. Performance of tasks
c. Emotional regulation
d. Metabolic function
e. Maintenance of brain and/or body
f. Immobility during dangerous times for predation etc.
g. Memory and information processing
h. Energy conservation, adaptive inactivity
j. Inflammation (molecular) and pain (behavior/perception)
k. Endocrine function
l. Growth (SWS)
m. Protein homeostasis
n. Endogenous sensory input with higher signal/noise than otherwise for neonates
o. Maintain mitochondrial function
p. CSF/glymphatics clearance
q. Synaptic homeostasis
r. Breathing and cardiovascular regulation
s. Sleep cycling for vigilance to monitor fires, danger, etc.
t. Appetite control
u. Increased/decreased predation
w. Kidney function
x. To dream
Consequences of sleep. These would be costs/tradeoffs in an optimization function
a. Reduced alertness
b. Reduced foraging
c. Reduced social contacts
d. Reduced activity
f. Reduced information gathering
3. Does evolutionary origin for function of sleep correspond to current functions of sleep? Is one sleep function dominant and the rest subservient or piggybacking? Are all on equal footing now? How big is the selection pressure in terms of evolutionary theory and measurement? Is there a group-level selection for sleep? What is first common ancestor of sleep? Circadian rhythm came before “sleep”.
4. Do different functions compete for time? Which function has longest time constant, which is distinct from which is most important? How does this change with life stage?
5. Are there different amounts of sleep time required for different functions of sleep? IF sleep is limited, are these different functions competing for the available time? If you restrict sleep, what is the ordering and magnitude of loss of different functions? Much known for learning and memory but what about other functions?
6. What are the characteristic time scales of sleep and what sets them? Which ones are invariant, and which ones change across size, species, developmental stage, brain region, task, etc.? What data already exist for this? What were our models predict? If you take differential equations models and parameterize by correct time constants (based on scaling of neuronal scaling and axon conduction velocities or based on physiological/allometric scaling or both), as you tune them what dynamical shifts or phase transitions do you see? Why does a 90 minute nap help more with learning tasks than 5 hours of sleep? Go from fastest (ripples) all the way up to long timescales. Can we list characteristics timescales for each function above?
7. Do we mean mechanism or function? How are these all reconciled? All equivalent?
8. If sleep is for repair, does whole body repair work differently or over different time scale than just for brain? Immune response, wound healing, and bone density connection to sleep and circadian rhythm? How to quantify?
9. Does brain “saturation” cause sleep? Global saturation (no?) or local saturation? How are local limits set and what sets partition of space for each category, say physics versus French literature? Seems like it should be experimentally measurable/testable now. What exactly needs to be measured? Why don’t we do it?
10. Do we need and can we get better developmental sleep data?
11. How do we include effects of temperature in sleep models? Via metabolism? Via loss of thermoregulation in sleep, hibernation, torpor in birds? What about sleep changes in total time, %REM, sleep cycle, etc across body temperatures in ectotherms, even for similarly sized organisms.
12. Is dreaming a way for brain to devise better computational algorithms for exploring large dimensional data or parameter space for theory? Why does this require sleep? Is it because it isn’t accessible to our conscious brain because done by some other part of the brain and takes lots of energy so have to save energy from elsewhere? If it’s so useful, why do we forget dreams? Maybe because their actual content isn’t useful? Maybe if you remember too well, you start to confuse dreams with reality and that becomes big problem? Is there some wake correlate or default state or meditation or exercise in which something similar can occur?
13. What are the individual and Societal health implications of all the above?
14. Does sleep time change with city/community/social network/college size?
15. How can we best leverage large digital datasets? Can we align 23andme with fitbit and other data?
16. Can we develop a quantitative, mathematical, predictive, universal theory of all this? What would it take to do this in terms of data, assumptions, mathematical and computational tools?
17. How can we capture the heterogeneity of the data? Deviations from universal or described on its own terms or correlates and cross-effect for different features. Can we isolate which factors impact sleep? Gender, genome, SES, race/ethnicitiy, age, etc. Higher-order interaction effects.
18. How do recent experimental tools enable the answering of the questions above?
What/Who are we missing?
Ken Wright UC Boulder Metabolism and Microbiome
Eve Van Cauter U of Chicago Metabolism and Immune function
David Paydafar UT Austin Engineering Mathematical Biologist and modeling of SIDS
Rosemary Braun Northwestern Statistical Analysis of Circadian Phase from 1 or 2 samples
Phyllis Zee Northwestern Sleep in elderly (bug picture thinker/ great at networking)
Michael Perlis U Penn Insomnia and suicide
Daniel Buysse Sleep health
Sonia Ancoli-Israel UCSD sleep and chemo
Laura Lewis Harvard Clearance of CSF during sleep
Monica Haack Harvard Immunology/Inflammation/Sex differences
Mark Opp UC Boulder
More traditional Evolutionary Biologist
Book audience more towards industry, Gates foundation, funding agencies, scientists, CDC,
Present more scientific basis and evidence-based approach to importance of sleep and understanding it and its consequences.
Mostly about basic science questions we’ve discussed throughout but with bullets at end of each chapter about implications and a chapter or two at end that is more focused on implications for health and policy.
Make more focused in terms of modeling and timescales of sleep as theme of book.
Data generated at many timescales, timescales can be organizing principle but must reconcile and synthesize them, modeling needed to reconcile and synthesize these timescales, this process actually unique vantage point for answering and asking big questions about sleep.
Help bring model-resistant scientists to understand importance of modeling.
Let’s be provocative and say we can try to model dreaming!
How does a brain calculate at the neuronal level which memories are valuable?
How do we model sleep across timescales?
Connecting machine-learning to modeling approach to make it better
Importance of working and talking across disciplines
How do a mapping from questions to appropriate scales. Hard to do it as just individual group or lab.
Susan Sara: Locus Coerelus, PreFrontal Cortex, Sleep Spindles all increase about 2 hours after leaning task in rats. Higher LC firing rates even during Slow Wave Sleep, suggesting important learning and consolidation happening during sleep. LC is silent during REM sleep. Much more firing together during SWS than during wake. What sets time scales of 2 hours? It is task dependent because some take and hour. Does characteristic time scale change with species or brain size? What sets this time scale? Does it change across development? What is theoretical expectation? Any data? How to disentangle age effect from increasing difficulty of task being learned. Must be careful. Coordination and coupling of spindles and ripples. Which is first? How are coordinated? Which is functionally more important? Temporal relationships. Can replay happen without ripples? Replay in visual cortex is not related to ripples. Not clear ripples and reply are causative for memory. Maybe it’s the subsequent ripple-spindling coupling that’s disrupted instead of the present one.
Gina Poe: Norepinephrine at Beta Adinergic receptor is to increase/encourage long-term potentiation. Slow Wave Sleep goes aing with absence of Acetochioline even in unihemispheric sleep in seals. Define sleep as a behavioral state but maybe that’s hard because of dissociated states. Maybe sleep must be defined as functional question to ask what’s needed or what counts. Is millisecond sleep enough to do something functionally? Lack of norepinephrine is required to erase memory so erasure can only happen in sleep. And targeted erasure (meaning targeting memory) can only happen in REM sleep thru depotentiation. Takes about 5 days or learning mazes for consolidation and reversal of learning. Again what sets this time scale across task, species, development, or amount of brain involved? What tags or systematically decides what information you want to remember and what you want to forget? Firing rate lowest in REM sleep. Hard to see sleep in insects etc. because electric signals cancel out because not layered or striated or organized in way that gives clear signals.
Is familiarity versus episodic memory two separate states or is it a continuum? Familiarity might be the feedback to the hippocampus that this can be eraser. Seems like two separate states. Why do we want to forget? Storage limitation, not useful, scary/traumatic, updating of previous thought like Santa Claus, etc. Danger of seizure from overload or saturation but seems like little evidence for this. Maybe saturation is only at the local circuit level. Flips questions around to ask why do we partition brain memory in way we do and why we do give this chunk of memory to this specific task. Nobody has looked at this question experimentally but could be done now. Maybe not really forgetting anyway but more like downweighting certain memories compared with others. Very modular structure and organization to cholinergic and adinergic systems. Perhaps people with super memories have less redundancy of memory. Would they forget more in old age? Is this observed?
Sara Aton: Foot shock creates fear memory that isn’t consolidated if transcription/translation is disrupted, neuron firing is blocked, or sleep is not allowed. Increase in theta and ripples (so oscillations in general) after learning event across sleep. If you induce similar pace (7 Hz) oscillations you recover consolidation even without sleep. Not clear of side effects of what happens to other functions of sleep if this is all you do. Use of optogenetic tools to do cool experiments. Able to make it so there is fear response but without awareness of what they’re afraid of. Correct time scales of rhythms are main thing needed for appropriate spike timing and that happens during sleeps and that’s why sleep is needed for memory consolidation even if that’s not primary function (although it could be). Is the 7Hz timescale task dependence, species dependent, developmental dependent, etc? Going from rate code to phase code to renormalized rate code and strengthening connections to propagate information through circuit. Thinks 7Hz is circuit dependent and circuit frequency is tag in that way that depends on wiring. Like a resonance frequency. She thinks phase matters but not frequency but not clear. Could you decipher or reverse engineer these codes to read out dreams or other things that are happening. Seems like there are experimental tools where we could start to do that. Circuit itself and post-synaptic partners are themselves making decision of what to remember or forget. Is certain cells can’t keep they are in some sense filtered (probably literally) out? Could at least ask this question theoretical and see what happens. Does lag time correspond to spindles?
Slow oscillations are 0-1 Hz. Delta is 1-4 Hz. Slow oscillations go much deeper that allows for calcium rebound. Slow oscillations are more global and Delta are more local. Even individual cortical column or relay neuron can exhibit Delta by itself. Spindles are 10-15Hz layered on top of these other waves. Ripples are 80-200Hz on top of spindles. Gamma are 40-80 Hz and occur while awake. Only 15% of spindles coupled to oscillations even though learning relates to total number of spindles in sleep. Seems like paradox. 200mV for waves. Most power is in slow waves. How much of the brain’s metabolic rate does this take? A lot or a little? 1/f power happening for this. Measures of fractal dimensions of these time series? Do follow a power law. These are like temporal correlations. What about spatial correlations (using calcium imaging) or overlap in temporal and spatial correlations? What percentage neurons contribute to this? About 5% of hippocampal cells involved in a learning task and memory. 30% have place fields. About 20% decrease in glucose and oxygen consumption during sleep. About 37% decrease for humans in NREM. Would these timescales change across species. Similar in mice, rats, and humans. Calcium imaging could say more.
Kimberley Whitehead: Looks at spindle bursts that’s different from sleep spindles because also happen in wakefulness. People are now claiming to see spindles in adult. Maybe these two things aren’t that different and we just haven’t realized that yet. Help refine sensoricortical maps. If you prevent bursting, barrel cortex doesn’t get organized properly. Sleep-wake state on the scale of minutes for humans. Pre-term baby has loads of delta in all their sleep. Higher power in pre-term. Role of nREM becomes more important as you grow. Maybe phasic and tonic REM sleep have different roles. Sub-sleep state in scales of tens of seconds. Active sleep movement on scale of seconds. Sleep oscillations important for sensory cortical organization in gestation and early development. Log normal for frequency of wake bouts and active sleep bouts. Role of benefits of wakefulness versus benefits of sleep and which are more important for survival and selection. Sensory experience more intense during sleep than during wakefulness so reverse of sleep model for information input proposed in paper (Cao, Herman, Poe, West, Savage). Wakefulness set more by time to birth than time to gestation age. Why not plot versus body weight instead of age? Age explains more of the variance. It’s experience dependence response and processing which seems like the generic definition of learning, not just laying the substrate for it. Surge to breathe, arousal, etc at birth so that’s enough to switch dynamics of wakefulness and sleep.
Elizabeth Klerman: Sleep-wake cycle and circadian cycle Homeostatic cycle and circadian cycle must be coupled. Must desynchronize two clocks to decipher effects of each one. Homeostatic is need for sleep overall and not just circadian. Clocks shift for 25 or 26 hour timing. Scored in 30 second chunks of sleep or wakefulness. Organized around core body temperature minimum that typically occurs a couple of hours before you wake up. No circadian rhythm in SWS but there is in REM sleep. Not understood physiologically why that is? Tononi’s model is that SWS reverse buildup of LTP. SWS percentage is same regardless of sleep debt so total SWS decreases. Related I think to questions about %REM and %NREM across development and across species. Remember to ask this question during my own talk. Drive to wake strongest right before you go to sleep and drive to sleep is strongest right before you wake up. Like hanging on by your fingertips but need some force you’re fighting against then. Needed to help you consolidate sleep is hypothesis. Humans can know time of day even as seasons change by adjusting melatonin levels up and down. People used to be seasonally reproductive. Children and adolescents have much faster buildup of sleep pressure than older people. People who are out in bright sunlight during day aren’t as affected by devices at night. So something to do with change in magnitude of light, but just current amount of light. Different models needed for short exposure to light and effects on phase delay. Additional process coupled to process L or some type of modification of process L. How do go from average- or group-level models to more individual-specific models? How does caffeine or other drugs affect it? How does aging, disease, mental disorders, etc. affect it? What is physiological analogue of the two variables for the phenomenological model? Some amalgam of the physiological factors? Maybe different amounts of sleep are needed for different functions of sleep? Maybe they are competing ofr the time they get?
Cecilia Diniz Behn: Sleep depends nonlinearly on amount of sleep deprivation and timing in circadian clock. Characteristic time scales affect shape of circadian waveform that affects timing of sleep and wake. Can we get a more physiological perspective and include that in modeling framework. Network between SCN, Wake, REM, and NREM. Choose scale at which network is specified, ranging from neurons to whole-brain regions or functions. How do you choose scale and how that affect choice of math methods. Depends on what questions you’re asking as to what scale you should choose. Time constants of homeostatic sleep for humans can be tuned to work. For rats homeostatic sleep build up too quickly so can’t compensate with circadian clock. Shows importance of time constants and that it can really give different effects and conclusions depending on what you think or use here. Could be order-of-magnitude difference between rats and humans. Populations of neurons can shift clock and likely mediated by electrical activity and firing in SCN. Can use Hodgkin-Huxley type models here. Circadian rhythm plays a role in sleep timing but not the ultimate clock! Feedback with other sleep clocks!
Marishka Brown: It is a mandated program but funding is not mandated. Largely instigated with push from circadian researchers. Trans-NIH sleep coordinating committee to build partners across all the different parts of NIH. Starting to build with NIAID. Studies on fatigue and sleep and performance joined together for studies. Connecting also with Human Health and Services. Circadian, disease, development, understanding are big parts of funded NIH sleep proposals. Turning discovery into health is motto. We scientists need to communicate better with those who fund us and general public to explain why basic science is so important to even be able to have something to translate into clinical applications. Need to do a better job to verbalize priorities to get funding behind us. Sleep disturbance is one of the best predictors of suicide. Lots of effects on heart and other functions. Correlation between sleep deficit and every type of risky behavior of adolescents. Public is convinced about sleep, just not academic medicine. Sleep and circadian rhythms are fundamental to health and life. How does circadian timing (NIAID) and sleep restriction affect immune response and vaccinations? Funding opportunity now. Ideas about chronotherapy. Does timing of organ donor for transplant need to match circadian timing of recipient or does it not matter? How does sleep deprivation affect it? What about sleep and health disparities compared with woman, URM, low SES, rural communities, etc. Why do these differences exist? (Not just cardio event in AA adults.) DOD strongly aware of importance of sleep and funding for research for it. Big upsurge in funding over last 3 years. Funding connections with cancer are expected to grow stronger. Silos of sleep research based on citation networks. NIH wants to break that down and create more communication.
Victoria Booth: Shortest time scales are bouts of minutes. (In principle, could this go down to even milliseconds because memories can form in that small of a time.) Wake durations are power law and sleep durations are exponential. Wake durations seem to follow Zipf’s law. Sleep duration follows random Poisson process like radioactive decay. More fragmentation in sleep bouts happens towards the end of the night. How strict should we be with definition of REM and SWS by voltage or time-scale and frequency characteristics across ages versus qualitatively similar behaviors. Randler et al. 2019 in Sleep Medicine for time in bed across ages. 3 to 4 months for baby to get used to day-night cycle. Sleep is more fragmented with aging. Trying to use fine time-scale modeling to predict changes in sleep across development and aging. Both wake and sleep bouts start exponential and random but become power law for wake around P15. Is this same as pre-term to term? Lesioning SCN you don’t see as much of the power-law behavior being established. By synaptogenesis P14 in rats is similar to about 3.5 years of age in humans. Younger subjects have longer tails than older subjects. Difference has a lot to do with NREM to REM transitions. Probability of waking up is much higher in older people so they sample the long-tail of the distribution and have longer individual waking bouts. LC develops early in fetus (Nakamura) and transitions/switches to alpha-2 autoinhibitory mechanism around P10 to P15. LC might be off in REM sleep. True in fetal rats but unclear in humans or other species? Diniz Behn and Booth (J Neurophysiol 2010) have physiologically-grounded model to investigate and explore this with and look at mean firing rate using ODEs. Stochasticity lengthens tail. Huge switch at 3 to 4 weeks to entering sleep in NREM phase. Can mode predict this?
Jerry Siegel: Sleep changes across species. Argues it’s driven by ecological factors. Argues sleep is adaptive in evolutionary sense. Perhaps animals are dying from repeated wakenings instead of loss of sleep for things like disk-over-water technique or experimental deprivations in general. Carnivores show little change in total sleep time with size. Recorded conditions aren’t much like natural conditions so how much does that affect sleep results you get. For example, temperature regulation. Example of platypus believed to have no REM and then they built platypusariums and there was lots of REM. Sleep changes a lot during migration. Amount of sleep in wild typically less than in lab. Elephant sleeps half as much in wild as in lab so about 2 hrs. Do they sleep at best time to thermoregulate. Bats sleep a lot (~20 hrs) but also have a very high metabolic rate. Go into hibernation though SWS that gets deeper and deeper. Could REM sleep be happening while elephants are seemingly awake and walking around? Might it be happening in only certain parts of the brain? Could you see how Mom responds to meat in terms of speed and orientation to determine how lack of sleep affects her. EEG basis for REM sleep but not based on brainstem recording. Call to include more temperature effects in models. Temperature and light are correlated. How to disentangle which drives sleep? Janet Best has paper on how temperature drive sleep cycles.
Geoffrey West: Body cannot have nearly perfect repair because of increase in entropy and second law of thermodynamics. But brain can be treated as not a closed system. It is an open system that draws from and drains off of resources of rest of body to keep nearly perfect repair. Lifespan and aging both about damage. Should be able to have formula in biology textbook for why we sleep 8 hours a day and live roughly 100 years.
Alex Herman: Subcortical REM more important in relation to number of synapses.
Bob Stickgold: More creative after sleep or dreaming. Because new creations are explored then or puts brain in better space or configuration to then do exploring once you wake up. How does this relate to patients with narcolepsy because seems like they have higher levels of creativity or lucid dreaming. Function of dreaming is not to solve problems as much as to explore different solution spaces and find what’s promising. Reminds me of trying to devise better computational algorithms for exploring large dimensional data or parameter space for theory. Maybe our brain has found good way to do this. Question is why does this require sleep to occur. Is it because it isn’t accessible to our conscious brain because done by some other part of the brain and takes lots of energy so have to save energy from elsewhere? If it’s so useful like this, why do we forget dreams? Maybe because their actual content isn’t useful? Or maybe we unconscious access to them but not direct access. Or maybe if you remember too well, you start to confuse dreams with reality and that becomes big problem.
Reference material notes
- Here is [A] database on [B] that I pull data from to do [C] analysis that might be of interest to this group (insert link).
- Here is a free tool for calculating [ABC] (insert link)
- This painting/sculpture/forms of artwork is emblematic to our discussion on [X]!
- Schwartz et al. 2017 offers a review on [ABC] migration as relate to climatic factors (add the reference as well).