← All self-study courses

Sound, sleep and memory: how the research unfolded (10 sessions, 15 minutes each, v1.0)

Goal of this course: Trace how sleep-and-memory research developed, and see what is new in our approach.

This is a self-study course for high-school students and first- and second-year undergraduates. In 10 sessions of 15 minutes each, you will trace how research on "using sounds or smells to call up a specific memory while a person is asleep" has grown over about 20 years. Building on that history, the goal is for you to be able to explain, in your own words, what is new about the approach our lab reported: presenting a sound related to a person's own experience during deep sleep.

A Japanese version of this course is available (講座_TMRの歴史とSES_v1.0.md). This English version has the same content.

Introduction (How to use this course)

What this is

• This is a teaching material that the Department of Neurophysiology, Kobe University Graduate School of Medicine, makes public for students' self-study.
• You may have heard that "memories get organized while you sleep." But can we play sounds to a sleeping person and pick out just one particular memory to strengthen? The line of research that has tried to answer this question is called TMR (targeted memory reactivation).
• This course starts from classic findings that appear in textbooks, reads representative TMR papers in chronological order, and finally reads our lab's paper (a study that examined the feasibility of SES) with an eye to how it differs from TMR.
• Each session lists the papers on which that session is based. All of them are published (or are publicly available preprints or accepted papers), and each is given a PubMed number (PMID) or a DOI (an identifier for each paper).

Who this is for

• Roughly second-year high-school students (about age 16–17) through first- and second-year undergraduates.
• High-school biology (the basics of nerves and memory) is enough background. The math involves only averages, division, and percentages.
• It suits people who are interested in research on sleep and memory, and people who want practice reading the "Methods" and "controls" of a paper.

What you need

• A conversational AI (one that answers your questions in a chat). The one you usually use is fine.
• Something to take notes on (paper or screen). Some sessions include simple hand calculations.
• In Sessions 5 and 10 only, you can run a short piece of code (Python) if you wish. A free environment that runs in a browser is enough. You can complete the course without running it.
• An environment where you can open paper pages. Putting https://doi.org/ in front of a DOI opens the paper's page. Adding a PMID after https://pubmed.ncbi.nlm.nih.gov/ opens the abstract page.

How to proceed

1. Load this file into a conversational AI (attach the file or paste it in).
2. Type "Please start Session 1."
3. Answer the AI's questions in your own words. It is fine to make mistakes. The explanation starts from where you went wrong.
4. Stop after 15 minutes. The next day, load the same file again and type "Please start Session 2." - Each session's content is sized to finish in 10–12 minutes. Use the remaining time to ask again about anything you did not understand.

Cautions

• AI can make mistakes. Check important points in textbooks or in the papers themselves.
• Do not enter your own or other people's personal information, medical information, or other people's data into the AI.
• This course is not intended as medical advice. Sessions 6, 9, and 10 deal with research on a condition called post-traumatic stress disorder (PTSD), but what they introduce is the way of thinking in research and how to read papers, not treatments. If you have concerns about your body, mind, or sleep, please consult a medical institution.
• Do not try this yourself. Playing sounds that remind you or your family of painful events while asleep is not an assignment in this course. The research is carried out with approval from an ethics review board and with the person's consent, with a physician watching over the participant.
• This course does not deal with the content of painful events themselves (what kind of experience it was). If reading becomes distressing, you may skip that session.
• You do not need to paste the full text of papers into the AI. Read the abstract on the paper's page, and show the AI a summary you have written in your own words.

Request for feedback (please be sure to send it)

This course is revised every time based on the voices of the people who use it. If you used it, please be sure to send us feedback.

• Send to: the address listed on the Contact page of the lab website
• Subject: "Course feedback: History of TMR and SES"
• What we would like you to write (4 points): (1) how many sessions you completed (2) parts that were hard to understand (3) parts that were good (4) how long one session took

Feedback from people who stopped partway is also very helpful.

Terms of use

• You are free to use this for personal study or in classes.
• If you modify and distribute it, please contact the lab (via the Contact page).
• The content is as of November 2026. Research moves on, so please check for yourself whether new papers have come out.

List of all 10 sessions

1. Do we forget less after sleeping? — Sleep and memory consolidation
2. The sleeping brain "replays" the day's events — Replay
3. Calling up memories with smell — The first TMR
4. Targeting one memory at a time with sound — Sound cues and skill memory
5. What got stronger, and how was it measured? — Controls, within-person comparison, effect size
6. Emotional and fear memories, and cues during sleep
7. Looking at the whole picture — Meta-analysis, and studies that could not be replicated
8. A different idea — CLAS, which plays sounds in time with the rhythm of brain waves
9. SES — What is new?
10. Reading a small clinical study critically — What can and cannot be said

Instructions for the AI (how to act as the teacher)

This section is for the AI that has loaded this file. You do not need to read it aloud to the learner.

• Role: You are a teacher who teaches high-school students and undergraduates one-on-one. Teach using only the "Session N" section of this file.
• Time: Each session is limited to 15 minutes. The explanation is sized to finish in 10 minutes. If it looks like it will go over 15 minutes, tell the learner the rest can be carried over to the next session.
• First question (only once, at the start of Session 1): "What is the one thing you most want to understand from this course? Please tell me one thing." Match the answer to the "Map" below and tell the learner, in one line, which sessions will be especially useful. Adapt that session's examples to the learner's interests.
• Start of Session 2 onward: Ask just one recall question taken from the "Summary" of the previous session.
• Flow of one session: State the aim in one line → convey the key points of the explanation one at a time → ask one question per key point → wait for the answer → hands-on task → check question → at the end, 3 key points and 2 mini quiz questions → ask about understanding on a 3-level scale (understood well / mostly / still unsure) → end by previewing the title of the next session in one line.
• Ask only one thing at a time: After asking a question, wait for the learner's answer. Do not give the answer first.
• Rephrase: If the learner seems not to understand, change the example, rephrase one level more simply, and ask one more question. If the answer is correct, acknowledge it briefly and move on.
• Be brief: Each turn should be within 100 words. Go longer only when the learner says "tell me more."
• Do not move to the next session on your own: Do not start the next session until the learner says "Please start Session N."
• Terms: Briefly paraphrase specialist terms the first time they appear. When you introduce an abbreviation, give the full term followed by the abbreviation (e.g. targeted memory reactivation (TMR)).
• Do not invent: Answer about the numbers and conclusions of papers only within what is written in this file. If asked about something not written here, answer "Let's check that together in the main text of the paper," and point to where to read (which figure, which section). Qualify general statements with "according to textbooks." Say so when your confidence is low. Do not cite paper authors, years, or journal names that are not in this file.
• Do not mix up TMR, CLAS, and SES (important): The three are different methods. Do not carry over the properties of TMR (linking a sound to a memory during waking learning or therapy) or the properties of CLAS (playing sounds timed to the phase of slow waves to strengthen slow waves) into your explanation of SES. SES is "a standalone method (not combined with waking therapy) that presents sounds related to the person's own traumatic experience during slow-wave sleep (deep NREM sleep, N3)." SES does not aim to change the amount or size of slow waves (that is the aim of CLAS). Whenever you compare the three, always name each one.
• Do not state benefits definitively: The SES paper is a small "feasibility" study without a control. Do not say "SES cures PTSD," "it works," or "its safety has been confirmed." Convey it in the paper's own wording (the symptom outcomes are only descriptive; a comparative trial with a sham condition is needed).
• About the lab: If asked about the lab's research, participation in clinical studies, or specific people, do not answer by guessing; instead guide the learner: "Please check what is publicly available on the Research and Publications pages of the lab website." Do not recruit participants for clinical studies.
• Decline: If the learner writes about a medical consultation (symptoms of themselves or their family, medication, worries about sleep or trauma) or personal information, do not deal with that content; say "This course cannot handle medical consultations or personal information. Please consult a medical institution for medical matters," and return to the course. If the learner seems distressed, tell them it is fine to stop that session. If the learner says "I want to try it myself," tell them not to, and add the reason in one line (it is research carried out under ethics review, consent, and a physician's supervision).
• Full text of papers: If the learner tries to paste the full text of a paper, tell them they do not need to. Have them read the abstract on the paper's page and show you a summary written in their own words.
• End of the final session: Together with the learner, put the 4 points of the "Request for feedback" in order, and give it to them as text they can send by e-mail as is.

Map (what the learner wants to understand → session)

• Why memories stay better after sleep, memory consolidation → Session 1
• Neurons in the sleeping brain replaying daytime activity → Session 2
• The beginning of TMR, smell cues → Session 3
• Sound cues, practicing skills or musical instruments and sleep → Session 4
• Controls, experimental design, effect size → Session 5
• Frightening memories, emotional memories and sleep, application to research on illness → Session 6
• Meta-analysis, reproducibility, limits of research → Session 7
• Brain-wave rhythms, closed-loop sound stimulation (CLAS) → Session 8
• SES, the lab's research → Session 9
• How to read clinical study papers, what can be said → Session 10

Session 1: Do we forget less after sleeping? — Sleep and memory consolidation

Papers this session is based on

• Jenkins JG, Dallenbach KM. Obliviscence during sleep and waking. American Journal of Psychology. 1924;35:605–612. DOI: 10.2307/1414040
• Diekelmann S, Born J. The memory function of sleep. Nature Reviews Neuroscience. 2010;11:114–126. PMID: 20046194 DOI: 10.1038/nrn2762 (review)
• Rasch B, Born J. About sleep's role in memory. Physiological Reviews. 2013;93:681–766. PMID: 23589831 DOI: 10.1152/physrev.00032.2012 (review)

Aims of this session

1. Be able to distinguish the three stages "learn → stabilize → recall."
2. Be able to say what the sleep stages are (N1–N3 of non-REM sleep, and REM sleep) and what slow-wave sleep (N3) is.
3. Be able to sketch the simplest form of an experiment that tests "we forget less after sleeping."

Explanation

Question: The night before an exam, which lasts better until the next morning: staying up all night memorizing, or memorizing and then sleeping? Answer: Since an experiment about 100 years ago, it has been shown repeatedly that memories last better if you sleep after learning. However, the answer to "why do they last?" has changed over those 100 years.

• Key point 1: Three stages of memory: Memory has the stages (1) learning (encoding), (2) stabilizing (consolidation), and (3) recalling (retrieval). A memory right after learning is fragile and becomes stable over time. Sleep is thought to be involved in this "stabilizing" stage.
• Key point 2: A classic from 1924: Jenkins and Dallenbach had people learn meaningless syllables (for example, sequences of sounds with no meaning, like "nukeso") and compared how much they could recall several hours later when they had slept versus stayed awake. There were only 2 participants, but recall was better after sleeping.
• The interpretation at the time was: "While awake, new events keep coming in and interfere with earlier memories. Sleep reduces the interference." This is the idea that sleep merely protects.
• Key point 3: From "protecting" to "nurturing": Later research (summarized in the 2010 and 2013 reviews) strengthened the idea that sleep does not merely reduce interference; rather, the brain actively reorganizes memories during sleep. This idea of "active consolidation" leads to replay in Session 2 and to TMR from Session 3 onward.
• Key point 4: Sleep stages: Sleep is broadly divided into non-REM sleep (non-rapid eye movement sleep) and REM sleep (rapid eye movement sleep, REM). Human non-REM sleep has three stages, N1 (dozing off), N2, and N3; N3 has many large, slow brain waves (slow waves), so it is also called slow-wave sleep (SWS). It appears mostly in the first half of the night. In this course, "slow-wave sleep," "N3," and "deep non-REM sleep" are used with the same meaning.

Figure: The simple form of the experiment (schematic)

Hands-on task (hand calculation, 3 minutes)

These are made-up numbers. People learned 10 syllables and were tested 8 hours later.

• Condition A (slept): recalled 6 on average
• Condition B (stayed awake): recalled 4 on average 1. What is the "proportion retained" for each, in %? 2. How many times more did A retain than B? 3. Can this experiment alone tell us that "sleep actively consolidated the memory"? If not, what other explanation is possible? Give one.

Check questions

1. State the three stages of memory in order. Which one is sleep thought to be involved in?
2. In one line, the difference between the interpretation of the 1924 experiment at the time and the later view.
3. What kind of sleep is slow-wave sleep (N3)?

Summary

• Memory goes learn → stabilize → recall. Sleep is thought to be involved in "stabilizing."
• The 1924 classic showed that "we forget less after sleeping," but the explanation has shifted from "protection" to "active reorganization."
• Slow-wave sleep (N3) = deep non-REM sleep with many large, slow brain waves. This is the main stage in this course.

Session 2: The sleeping brain "replays" the day's events — Replay

Papers this session is based on

• Wilson MA, McNaughton BL. Reactivation of hippocampal ensemble memories during sleep. Science. 1994;265:676–679. PMID: 8036517 DOI: 10.1126/science.8036517

Aims of this session

1. Be able to say what the "place cells" of the hippocampus do.
2. Be able to say what was compared with what in the 1994 experiment to show replay.
3. Be able to say the difference between "replay occurs" and "replay is necessary for memory."

Explanation

Question: How could we confirm that daytime experiences are being "replayed" in the sleeping brain — by some method other than asking about dreams? Answer: We can count whether groups of neurons that worked together during the experience also work together during sleep.

• Key point 1: Place cells: In the hippocampus (a brain region involved in memory), there are neurons that are active only when an animal is at a particular location in a room; they are called place cells. Because different cells respond at location A, location B, and so on, you can read "where the animal is now" from the pattern of cell activity.
• Key point 2: The 1994 experiment: Wilson and McNaughton simultaneously recorded the activity of many place cells from the hippocampus of 3 rats. They compared three periods: (1) slow-wave sleep before the maze task, (2) during the task, and (3) slow-wave sleep after the task.
• "Pairs of cells that worked together at the same location" during the task were more likely to work together in sleep after the task. This was not the case in sleep before the task.
• This increase was not seen for cells that were not active during the task, or for pairs of cells whose locations did not overlap.
• This tendency gradually faded as sleep went on.
• Key point 3: "Sleep before" as the control: The clever part of this experiment is that it used "the same animal's sleep before the task" as the comparison. This rules out the explanation that the pairs simply tended to work together from the start.
• Key point 4: Occurring ≠ necessary: What this experiment showed is that "the pattern of daytime activity reappears during sleep." Whether it is necessary for memory has to be tested by other experiments (blocking it or strengthening it). This is where the idea comes in: "Can we give a cue from outside and deliberately trigger the replay?" = TMR, from Session 3 onward.

Figure: Groups of place cells and replay in sleep (schematic)

Hands-on task (reading a table, 3 minutes)

These are made-up numbers. For pairs of cells, we counted the "number of times they fired together within 1 second," per 10 minutes.

1. Calculate "after ÷ before" for each pair.
2. What role does the A and D pair play in this experiment?
3. If only "sleep after" had been measured, without measuring "sleep before," what could no longer be said?

Check questions

1. What are place cells?
2. What did the 1994 experiment compare with what to show "replay"?
3. Why are "replay occurs during sleep" and "replay is necessary for memory" separate questions?

Summary

• In the sleeping hippocampus, pairs of cells that worked together during the day work together again (replay, reactivation).
• The 1994 experiment showed this phenomenon by using "sleep before the task" and "non-overlapping pairs" as comparisons.
• "Occurring" and "necessary" are separate questions. The idea of deliberately triggering replay from outside leads to TMR.

Session 3: Calling up memories with smell — The first TMR

Papers this session is based on

• Rasch B, Büchel C, Gais S, Born J. Odor cues during slow-wave sleep prompt declarative memory consolidation. Science. 2007;315:1426–1429. PMID: 17347444 DOI: 10.1126/science.1138581

Aims of this session

1. Be able to state the basic idea of TMR: calling up a memory with a "cue."
2. Be able to state the 4 controls of the 2007 experiment.
3. Be able to state the difference between "memories you can put into words (declarative memory)" and "memories learned with the body (procedural memory)."

Explanation

Question: Can the "replay in sleep" we saw in Session 2 be triggered from outside? Answer: If a smell that was present while learning is presented again during sleep, the memory learned at that time is retained better — this was shown in 2007.

• Key point 1: The idea of a cue: When you learn something, the sounds and smells present at that time become linked with the memory. Later, smelling that smell makes it easier to call up the memory (similar to the experience of a certain smell reminding you of the past). The idea is that if that cue is given during sleep, replay in sleep might be triggered deliberately. This method later came to be called TMR (targeted memory reactivation) (Session 4).
• Key point 2: The 2007 experiment: Participants performed a task of learning the locations of objects in a room with a smell. That night, when the same smell was released during slow-wave sleep, the location memory was retained better the next morning.
• Key point 3: The 4 controls (the most important part of this paper):
• (1) Releasing the smell during REM sleep → no effect
• (2) Releasing the smell while awake → no effect
• (3) Releasing the smell during sleep when there was no smell during learning → no effect
• (4) For a task learned with the body (a finger-movement skill = procedural memory) → no effect
• In other words, the effect was seen only for the combination of "a smell present during learning," "released during slow-wave sleep," and "a hippocampus-dependent memory that can be put into words (declarative memory)."
• Key point 4: Inside the brain too: Using a method that images brain activity (functional MRI), they also showed that releasing the smell during slow-wave sleep activates the hippocampus.
• Key point 5: Why start with "smell"?: One reason is that smells were thought to be less likely than sounds to wake a person up. In the next session, we move to "sound," which can target memories more precisely.

Figure: Combinations of conditions in this experiment (schematic)

Hands-on task (prediction table, 3 minutes)

Without looking at the table above, answer the following questions, then check with the table.

1. Without Control (3) (no smell during learning), could you rule out the explanation that "the smell itself simply deepened sleep and improved memory"?
2. What is Control (2) (releasing while awake) a control for?
3. If you were designing the next experiment, which one condition would you add? (e.g. change the type of smell, change the amount learned)

Check questions

1. State the basic idea of TMR in one line.
2. In the 2007 experiment, which combination showed an effect?
3. Choose one of the 4 controls and say which explanation it rules out.

Summary

• TMR = a method of giving again, during sleep, a cue that was linked with a memory during waking learning.
• The 2007 study showed that giving a smell during slow-wave sleep makes hippocampus-dependent memories retained better.
• The value of this paper lies in its 4 controls. Being able to say which explanation each comparison is meant to rule out is what builds the skill of reading papers.

Session 4: Targeting one memory at a time with sound — Sound cues and skill memory

Papers this session is based on

• Rudoy JD, Voss JL, Westerberg CE, Paller KA. Strengthening individual memories by reactivating them during sleep. Science. 2009;326:1079. PMID: 19965421 DOI: 10.1126/science.1179013
• Antony JW, Gobel EW, O'Hare JK, Reber PJ, Paller KA. Cued memory reactivation during sleep influences skill learning. Nature Neuroscience. 2012;15:1114–1116. PMID: 22751035 DOI: 10.1038/nn.3152
• Oudiette D, Paller KA. Upgrading the sleeping brain with targeted memory reactivation. Trends in Cognitive Sciences. 2013;17:142–149. PMID: 23433937 DOI: 10.1016/j.tics.2013.01.006 (review)

Aims of this session

1. Be able to state the difference between smell (a cue for the whole room) and sound (a cue for each item).
2. Be able to state the advantage of a design that "compares cued items with uncued items within the same person."
3. Be able to state that TMR has been tested not only on verbal memories but also on skill memories (such as playing a musical instrument).

Explanation

Question: The smell in Session 3 was linked to "everything" learned that night. Then, out of 50 items learned, can we strengthen only 25 chosen items? Answer: It was shown that if a different sound is linked to each item, playing only some of those sounds during sleep strengthens only the targeted items.

• Key point 1: Sound can target items one by one: A smell spreads through the whole room and cannot be removed quickly. Sounds are short, and a different sound can be prepared for each item (e.g. "meow" for a picture of a cat, a "whistle" for a picture of a kettle).
• Key point 2: The 2009 experiment (Rudoy et al.): Participants learned the locations of pictures placed around a screen, together with a sound for each picture. During slow-wave sleep in a subsequent nap, only some of the sounds were played quietly. When asked for the locations after waking, participants remembered the locations of the pictures whose sounds were played during sleep more accurately than those whose sounds were not.
• Within-person comparison: In this experiment, the same person has both "cued pictures" and "uncued pictures." Differences in memory ability between people (people with naturally good or poor memory) enter both sides of the comparison equally, so the difference becomes easier to see. We will check this with numbers in Session 5.
• Key point 3: The 2012 experiment (Antony et al.): Participants practiced playing two melodies by pressing keys in time with markers moving across a screen (similar to a music game). When only one of the melodies was played during a nap, after waking their performance of the melody that was played had improved relative to the one that was not. This means TMR was tested not only on memories that can be put into words but also on skill memories.
• Compared with Control (4) in Session 3 (no effect on a finger skill), this shows that we cannot generalize that "it does not work for skill memories." The type of cue (a smell, or the sound of the skill itself) and differences in the task can change the result.
• Key point 4: It gets a name: As such studies accumulated, by the time of the 2013 review the term "targeted memory reactivation (TMR)" had come into wide use as the name of this line of research.

Figure: Within-person comparison design (schematic)

Hands-on task (hand calculation, 3 minutes)

These are made-up numbers. The error in the location answers (distance from the correct location; smaller is more accurate) was measured before and after sleep.

1. For each, by how many cm did the error increase from before to after sleep (= how much was forgotten)?
2. Compared with uncued pictures, by what % less was the "amount forgotten" for cued pictures?
3. Can we say "the cued pictures got better after sleep than before sleep"?

Check questions

1. Why is sound better suited than smell for "targeting items one by one"?
2. What is good about comparing "cued" and "uncued" within the same person?
3. How does the 2012 experiment relate to Control (4) of 2007?

Summary

• Using sound cues, only some of the learned items can be targeted and strengthened (2009).
• A design that compares "cued" and "uncued" within the same person cancels out differences between people.
• It was also tested on skill memories (2012). By the time of the 2013 review, TMR had become the name of this line of research.

Session 5: What got stronger, and how was it measured? — Controls, within-person comparison, effect size

Papers this session is based on: We reread, with numbers, the methods of the papers in Sessions 3 and 4 (Rasch et al. 2007, Rudoy et al. 2009).

Aims of this session

1. Be able to say how many kinds of "comparisons (controls)" there are in TMR research.
2. Be able to calculate an effect size (a number that measures the size of a difference using variability as the yardstick) by hand.
3. See, through numbers, that results from experiments with few people fluctuate even when the underlying truth is the same.

Explanation

Question: "Items whose sounds were played during sleep were remembered better." Before believing this sentence, what would you want to check? Answer: Three things: (1) what it was compared with (control), (2) how large the difference is (effect size), and (3) how many people were measured.

• Key point 1: Kinds of control: The following 4 controls are often used in TMR research.
• Uncued items (comparison within the same person, Session 4)
• Sounds that were not linked during learning (rules out an action of the sound itself; corresponds to Control (3) in Session 3)
• Playing during a different sleep stage or while awake (tests whether timing matters; Controls (1) and (2) in Session 3)
• A group with a sham procedure (pretending to play sounds) (rules out the influence of people's expectations)
• Key point 2: Size of the difference = effect size: Being told "the difference in means is 3 points" means different things depending on whether scores naturally vary by 10 points or by only 1 point. So we use the difference divided by "variability (standard deviation (SD), a measure of spread around the mean)." This is called the effect size. When comparing within the same person, a common form is the mean of the "per-person differences" divided by the SD of the "per-person differences."
• Commonly cited rough guides are 0.2 ≈ small, 0.5 ≈ medium, 0.8 ≈ large (though the sense of scale differs by field).
• There are several kinds of effect size that differ in the details of calculation ("Hedges' g," which appears in the meta-analysis in Session 7, is one of them). The idea is the same, "difference ÷ variability," but avoid lining up numbers of different kinds side by side and comparing them directly.
• Key point 3: What you count is "people": Even if one person learns 50 pictures, the unit of comparison is the "person." If you count 50 items × 20 people = 1000 items, the difference looks far more certain than it is. Count how many people were measured first, not the number of items.
• Key point 4: Small experiments fluctuate: Even if the true difference is the same, in an experiment with about 20 people the result may come out large or hardly at all. This is why we do not draw conclusions from the numbers of a single paper, and it leads to the meta-analysis in Session 7.

Hands-on task (Python, 5 minutes, optional)

This is made-up data. Each of 20 people has an accuracy (%) for cued items and for uncued items. The data are built so that cued items are 3 points higher on average.

1. The data were built to be +3 points, but the mean difference was 1.9 points. Why did it shift?
2. Change the 1 in random.seed(1) to 2, 3, or 42, run it, and see how the effect size changes (if you are not running it, look at the numbers in the answer below).
3. For those not running it (hand calculation): when the mean difference is 1.9 points and the SD of the differences is 7.0 points, what is the effect size?

Check questions

1. Name two controls often used in TMR research, and say which explanation each rules out.
2. What is an effect size? Why is the difference in means alone not enough?
3. Why must we not count "50 items per person × 20 people = 1000 cases"?

Summary

• Before believing a result, check (1) what it was compared with, (2) the size of the difference (effect size), and (3) how many people.
• Effect size = difference ÷ variability. Within-person comparison cancels out differences between people and makes the yardstick finer.
• Small experiments give widely fluctuating results even from the same truth. That is why we look at many studies together (Session 7).

Session 6: Emotional and fear memories, and cues during sleep

Papers this session is based on

• Cairney SA, Durrant SJ, Hulleman J, Lewis PA. Targeted memory reactivation during slow wave sleep facilitates emotional memory consolidation. Sleep. 2014;37:701–707. PMID: 24688163 DOI: 10.5665/sleep.3572
• Hauner KK, Howard JD, Zelano C, Gottfried JA. Stimulus-specific enhancement of fear extinction during slow-wave sleep. Nature Neuroscience. 2013;16:1553–1555. PMID: 24056700 DOI: 10.1038/nn.3527
• He J, Sun HQ, Li SX, et al. Effect of conditioned stimulus exposure during slow wave sleep on fear memory extinction in humans. Sleep. 2015;38:423–431. PMID: 25348121 DOI: 10.5665/sleep.4502
• (Lab paper) Purple RJ, Sakurai T, Sakaguchi M. Auditory conditioned stimulus presentation during NREM sleep impairs fear memory in mice. Scientific Reports. 2017;7:46247. PMID: 28401950 DOI: 10.1038/srep46247 (full text freely available)
• Schwartz S, Clerget A, Perogamvros L. Enhancing imagery rehearsal therapy for nightmares with targeted memory reactivation. Current Biology. 2022;32:4808–4816.e4. PMID: 36306786 DOI: 10.1016/j.cub.2022.09.032
• van der Heijden AC, van der Werf YD, van den Heuvel OA, Talamini LM, van Marle HJF. Targeted memory reactivation to augment treatment in post-traumatic stress disorder. Current Biology. 2024;34:3735–3746.e5. PMID: 39116885 DOI: 10.1016/j.cub.2024.07.019

Aims of this session

1. Be able to explain fear conditioning and "extinction" in your own words.
2. Be able to name two studies in which cues during sleep acted to "weaken" fear responses.
3. When TMR is used in research on illness, be able to say where the cue is linked to the memory (conditioning in the laboratory, waking therapy).

Explanation

Question: We have learned that TMR is a method of "strengthening" memories. Then, if a cue is given for a frightening memory, does the fear get stronger or weaker? Answer: Depending on the study, results in the direction of the fear response weakening (extinction progressing) and results in the direction of the memory being disrupted have both been reported. Whether something "strengthens or weakens" depends on which memory it acted on (the frightening memory, or the memory of having learned that something is safe).

• Key point 1: Fear conditioning and extinction: If a certain sound is repeatedly followed immediately by an unpleasant stimulus (in human experiments, a mild electric stimulus, for example), the body comes to brace itself at the sound alone. This is called fear conditioning, and the sound is called the conditioned stimulus (CS). Afterward, if the sound is heard many times without the unpleasant stimulus, the bracing gradually weakens. This is called extinction, and it is thought that rather than "the frightening memory disappearing," a new memory that "this sound is now safe" is formed on top of it.
• Key point 2: Emotional memory and TMR (2014): Cairney et al. had participants learn the locations of unpleasant and neutral pictures together with sounds related to the pictures, and played half of the sounds during slow-wave sleep in a nap. Faster location judgments for the unpleasant pictures whose sounds were played were related to the duration of slow-wave sleep and to sleep spindles (short, fast waves during sleep) (there were 15 participants).
• Key point 3: Studies in which fear responses weakened (2013, 2015):
• Hauner et al. carried out fear conditioning using smells, then released the same smell during slow-wave sleep. The fear response linked to that smell weakened, limited to the targeted stimulus. In brain imaging, hippocampal activity decreased and the pattern of activity in the amygdala (a region involved in emotion) was reorganized. It attracted attention because extinction progressed without showing the frightening stimulus itself again.
• He et al. showed that after fear conditioning with a sound, playing the conditioned-stimulus sound for 10 minutes during slow-wave sleep weakened the fear response the next day (compared with a group played an unrelated sound and a group given nothing). They reported that sleep structure did not change.
• Key point 4: In mice, the memory was disrupted (2017, lab paper): Purple et al. (a lab paper) fear-conditioned mice by pairing a sound with a mild electric stimulus, then played the conditioned-stimulus sound during non-REM sleep. When the sound was played later, the fear memory had weakened. This did not happen when the same sound was played during REM sleep. This is in the same direction as human extinction studies (fear responses decrease), but whether "a new safety memory was formed" or "the original memory was disrupted" is a point where interpretation differs depending on how the experiment is built.
• Key point 5: Toward research on illness — the cue is linked during waking therapy:
• Schwartz et al. had patients with nightmare disorder listen to a certain sound while undergoing a therapy in which they "rewrite the plot of the nightmare in a positive direction and rehearse it in their mind" (imagery rehearsal therapy), and then, for the following 2 weeks, played that sound during REM sleep while they slept at home (using a headband that automatically determines the sleep stage). The group in which the sound was linked had fewer nightmares than the group in which it was not, and the reduction was maintained 3 months later.
• van der Heijden et al. studied patients with PTSD (post-traumatic stress disorder, a condition in which symptoms such as the memory of a life-threatening event coming back again and again persist after that event). In the sleep after one session of waking therapy (EMDR, eye movement desensitization and reprocessing: a psychotherapy in which the person recalls a painful memory while receiving stimuli such as eye movements alternating left and right, or click sounds), they played the click sound used in that therapy as a cue (compared with a sham-stimulation group).
• What these two have in common is that the cue sound is linked to the memory in the setting of waking therapy. Because TMR is a method of "playing, during sleep, a cue that was linked during waking learning (here, therapy)," it takes the form of using sleep to support the effect of the therapy. This point becomes important when we compare it with SES in Session 9.

Figure: "Where" was the cue linked to the memory? (schematic)

Hands-on task (sorting into a table, 3 minutes)

Sort the 5 studies above into the following two boxes.

• Box A: studies in which the cue was linked through "conditioning within the experiment"
• Box B: studies in which the cue was linked through "waking therapy" Then think of one possible "way of creating a cue that is neither Box A nor Box B" (this is a preview of Session 9).

Check questions

1. Is fear "extinction" the disappearance of the frightening memory?
2. Name one point that attracted attention in Hauner et al. 2013.
3. In TMR studies of nightmare disorder and PTSD, where was the cue sound linked to the memory?

Summary

• Fear extinction = a new memory that "it is now safe." There are studies in which cues during sleep weakened fear responses (humans 2013 and 2015, mice 2017).
• Whether something "strengthens or weakens" depends on which memory it acted on.
• In TMR research on illness, the cue is linked in the setting of waking therapy and played during sleep to support the therapy.

Session 7: Looking at the whole picture — Meta-analysis, and studies that could not be replicated

Papers this session is based on

• Hu X, Antony JW, Creery JD, Vargas IM, Bodenhausen GV, Paller KA. Unlearning implicit social biases during sleep. Science. 2015;348:1013–1015. PMID: 26023137 DOI: 10.1126/science.aaa3841
• Humiston GB, Wamsley EJ. Unlearning implicit social biases during sleep: A failure to replicate. PLoS One. 2019;14:e0211416. PMID: 30682167 DOI: 10.1371/journal.pone.0211416 (full text freely available)
• Hu X, Cheng LY, Chiu MH, Paller KA. Promoting memory consolidation during sleep: A meta-analysis of targeted memory reactivation. Psychological Bulletin. 2020;146:218–244. PMID: 32027149 DOI: 10.1037/bul0000223
• Paller KA, Creery JD, Schechtman E. Memory and sleep: How sleep cognition can change the waking mind for the better. Annual Review of Psychology. 2021;72:123–150. PMID: 32946325 DOI: 10.1146/annurev-psych-010419-050815 (review)

Aims of this session

1. Be able to say what "replication studies" and "meta-analyses" do.
2. Be able to state the main conclusion of the 2020 meta-analysis (in which sleep stages an effect was seen).
3. Be able to say how to take a single striking paper.

Explanation

Question: In Session 5 we learned that small experiments give fluctuating results even from the same truth. So does TMR "really" work? Answer: One paper cannot decide that. We look at both studies that redo the same method (replication studies) and studies that combine many studies in a calculation (meta-analyses).

• Key point 1: A striking study (2015): Hu et al. had people practice weakening implicit biases (assumptions about race or gender that are hard to notice in oneself) with linked sounds, and then played one of the sounds during subsequent slow-wave sleep. They reported that the reduction in bias was maintained for the cued side, and that a difference remained 1 week later. It attracted a great deal of attention as a sign that sleep research might be applicable to social problems.
• Key point 2: Redone, and it did not appear (2019): Humiston and Wamsley redid the experiment with the same procedure on 31 university students. The result: playing the sounds did not weaken bias (either immediately or 1 week later), and they had tested enough people to adequately detect an effect of the size in the original study. While raising both possibilities — "the original report was a lucky hit by chance, or ours was an unlucky miss by chance" — the authors concluded that confidence that cues during sleep can weaken bias had decreased.
• This is not a matter of one of them being a "lie." As you experienced in Session 5, results splitting between small studies happens routinely.
• Key point 3: Calculating them together (the 2020 meta-analysis): Hu et al. combined 91 TMR experiments (212 effect sizes, about 2000 people in total) in a calculation.
• Overall, TMR showed an effect of aiding memory (effect size Hedges' g of about 0.29 = small, short of medium. This is an effect size built on the idea of "difference ÷ variability" and corrected for the bias caused by small sample sizes).
• The effect was seen when cues were given during non-REM sleep (N2 and slow-wave sleep), and no effect was seen when they were given during REM sleep or while awake.
• Effects were seen for several kinds of memory, including memories that can be put into words and skill memories.
• They also examined, by several methods, the influence of the bias whereby studies with effects are more likely to be published (publication bias).
• Key point 4: How to take a single paper: (1) Look at the effect size and the number of people. (2) Look for whether there are other similar studies and whether there are replications. (3) If there is a meta-analysis or review, place the paper within it. The 2021 review (Paller et al.) discusses both sides: memory processing during sleep may be involved not only in retaining memories but also in problem solving and processing emotions, whereas in forms such as repeatedly ruminating over worries it may be harmful.

Figure: The idea of a meta-analysis (schematic)

Hands-on task (hand calculation, 4 minutes)

These are 5 made-up studies. Average the effect sizes, weighting by the number of people (a real meta-analysis determines the weights from variability, but here we approximate with the number of people).

1. What is the simple average of the 5 effect sizes?
2. What is the average weighted by number of people (Σ people × effect size ÷ Σ people)?
3. Why are the two different? When smaller studies have larger effect sizes, what does that prompt you to suspect?

Check questions

1. What does a replication study do, and what does a meta-analysis do?
2. In the 2020 meta-analysis, in which sleep stages was an effect seen, and in which was it not?
3. After reading one striking paper, what do you look for next?

Summary

• There is a case in which a striking study (2015) was not replicated by a redo with enough people (2019). Neither is a "lie"; small studies can split.
• The 2020 meta-analysis: the effect of TMR overall is small, short of medium. The effect was seen in non-REM sleep, and not in REM sleep or while awake.
• Read a single paper by placing it among effect size, number of people, replication, and meta-analysis.

Session 8: A different idea — CLAS, which plays sounds in time with the rhythm of brain waves

Papers this session is based on

• Ngo HV, Martinetz T, Born J, Mölle M. Auditory closed-loop stimulation of the sleep slow oscillation enhances memory. Neuron. 2013;78:545–553. PMID: 23583623 DOI: 10.1016/j.neuron.2013.03.006
• (Reference, Session 6) van der Heijden et al. 2024 (a study that combined TMR with a method timed to the rhythm of brain waves)

Aims of this session

1. Be able to say how closed-loop auditory stimulation (CLAS) is a different idea from TMR.
2. Be able to say why it matters to time sounds to the "peak (up-state)" of slow waves (slow brain waves).
3. Be able to distinguish, in one line each, "what TMR and CLAS aim at."

Explanation

Question: Is all research that plays sounds to sleeping people TMR? Answer: No. There is another line of research that tries to strengthen the brain waves themselves through the timing of sounds, rather than calling up memories through the content of sounds. That is CLAS.

• Key point 1: Slow waves have peaks and troughs: The brain waves of slow-wave sleep are slow, large waves at roughly once per second (slow oscillations). At the "peak" of a wave (up-state), neurons are active together, and at the "trough" (down-state) they become quiet together. The rhythm of these peaks, and the sleep spindles riding on them, are thought to be involved in memory consolidation.
• Key point 2: Closed-loop = adjusting while measuring: Closed-loop refers to a system that continuously measures brain waves and delivers stimulation on the spot according to the result (similar to an air conditioner adjusting its airflow while measuring room temperature). In closed-loop auditory stimulation (CLAS), a short burst of noise (a brief crackle, like static) is played in time with the arrival of the slow-wave peak.
• Key point 3: The 2013 experiment (Ngo et al.): Playing sounds in time with slow-wave peaks strengthened the slow-wave rhythm, increased the sleep spindles timed to it, and made memories of word pairs learned before sleep better retained. On the other hand, playing sounds at timings shifted away from the peaks had no effect. It was not the content of the sound but the timing that was decisive.
• The sound here is noise that has nothing to do with the learned words. The idea is not to call up a specific memory but to support memory consolidation as a whole by strengthening the rhythm of sleep itself.
• Key point 4: The difference between TMR and CLAS (one line each):
• TMR: plays, during sleep, a sound that was linked to a memory during waking learning (or therapy), aiming at replay of that memory. The content matters.
• CLAS: plays a short sound unrelated to memory in time with slow-wave peaks, strengthening the slow waves themselves. The timing matters.
• The two can also be combined. van der Heijden et al. 2024 in Session 6 played the click sound used in therapy (a TMR cue) timed to slow waves.
• Key point 5: The map so far: Research using sounds during sleep had two major lines: "calling up a specific memory through the content of the sound (TMR)" and "strengthening brain waves through the timing of the sound (CLAS)." In the next session, we will look at an idea that differs from both.

Figure: Timing to the slow-wave peak (schematic)

Hands-on task (hand calculation, 3 minutes)

These are made-up numbers. Suppose a person's slow waves continue at a steady rhythm of exactly 0.8 Hz (0.8 times per second).

1. How many seconds is one wave cycle?
2. From a peak to the next trough is half of that. How many seconds?
3. Suppose there is a 0.5-second delay between the device detecting a peak and playing a sound. Where in the wave is the sound that was meant to be timed to the peak likely to land? From this, what is required of a CLAS device?

Check questions

1. What is closed-loop? Give an everyday example.
2. In the 2013 experiment, which condition had an effect and which did not?
3. For TMR and for CLAS, what about the sound matters?

Summary

• CLAS = a method that plays short sounds unrelated to memory in time with slow-wave peaks, strengthening the slow waves themselves.
• The 2013 study: timing to the peaks strengthened slow waves, spindles, and memory; shifting the timing had no effect.
• TMR is about "content," CLAS is about "timing." The two are different ideas, and there are studies that combined them.

Session 9: SES — What is new?

Papers this session is based on (lab paper)

• Ino K, et al., Kim Y, Sakaguchi M. Sound Exposure During Sleep (SES) in PTSD Patients: An Open-Label Feasibility Study. medRxiv (preprint: a manuscript made public before peer review). Posted May 2026. DOI: 10.64898/2026.05.02.26352243 (available at https://www.medrxiv.org/content/10.64898/2026.05.02.26352243v1)
• The same study has been accepted by the journal npj Mental Health Research under the title "Sound Exposure During Sleep in individuals with PTSD: An Open-Label Feasibility Study" (October 2026, not yet published). The explanations in this course follow the wording of the accepted version.
• Trial registration number: jRCT1030230706 (anyone can read an outline of the plan in jRCT, Japan's registry of clinical studies)

Aims of this session

1. Be able to state the definition of SES in your own words, following the wording of the paper.
2. Be able to state 3 points in which SES differs from both TMR and CLAS.
3. Be able to state, in order, the procedure of one night of SES (when, what, and how it is stopped).

Explanation

Question: In Session 6, in TMR studies of PTSD, the cue sound was linked to the memory during waking therapy. Then, can we conceive of acting on the trauma memory itself during sleep, without newly linking a cue in a therapy setting? Answer: SES is the idea of using as the cue a sound that is already linked to the memory within the person's own experience.

• Key point 1: Background (the line of argument in the paper's Introduction): One of the treatments for PTSD whose benefit has been established is a psychotherapy in which the person deliberately and repeatedly confronts painful memories in a safe setting (e.g. prolonged exposure, PE). However, deliberately and repeatedly confronting them while awake is very distressing, and quite a few people drop out partway. So there is a need for methods that do not require deliberate confrontation while awake. There is also an accumulation of research showing that sleep is involved in the consolidation of fear-extinction memories (Session 6).
• Key point 2: The definition of SES: The paper describes SES as follows (paraphrased in plain words).
• Sounds related to the person's own painful experience (an autobiographical traumatic experience)
• are presented during slow-wave sleep (N3),
• as a standalone method (stand-alone approach) that is not combined with waking therapy (psychotherapy).
• At the end of the paper's Introduction, put briefly, it states to the effect that this is "the first human intervention (to our knowledge) that presents individualized trauma-related cues during sleep as a standalone method that does not require waking psychotherapy."
• Key point 3: The difference from TMR (the most important part):
• In TMR (Sessions 3–7), the cue was a sound (or smell) newly linked to the memory in the setting of waking learning or therapy. Even in TMR studies of PTSD and nightmares, the cue was linked through conditioning in the lab or through waking therapy (the table in Session 6).
• The sound in SES is not a sound newly linked in an experiment or therapy. It uses a sound related to a memory that already exists within the person's experience. Specifically, based on the experience the person described, a sound that recalls that scene is created together with the study physician (the specific content of the sounds is not disclosed even in the paper, out of consideration for readers). In other words, the link between sound and memory already exists within the experience itself. The way to tell them apart is "whether the study includes a procedure that newly connects sound and memory (conditioning, or playing it during therapy)." TMR has such a procedure and SES does not; in SES, creating the sound is the work of confirming "which sound recalls that experience." The paper states that this is how it differs from previous TMR studies.
• SES is not a method used together with waking therapy in order to use sleep to support the effect of the therapy.
• Key point 4: The difference from CLAS:
• CLAS (Session 8) was a method that plays short sounds unrelated to memory in time with slow-wave peaks, strengthening the slow waves themselves.
• SES does not aim to strengthen slow waves. Nor is it a method that plays sounds timed to the phase (peak or trough) of slow waves. It looks at the sleep stage (whether it is stable slow-wave sleep) and, during that stage, presents sounds related to the memory.
• The difference between "phase" and "stage": phase is "peak or trough" within a single wave lasting about 1 second. CLAS plays a short sound matched to each individual wave. Stage is "is this slow-wave sleep now?" judged in 30-second segments. SES confirms that slow-wave sleep segments have continued for 90 seconds, and then plays a 30-second sound regardless of the peaks and troughs of individual waves. The signal to start the sound is "the technician judging that it is now slow-wave sleep," not the shape of each individual wave.
• Key point 5: The procedure for one night (from the paper's Methods, roughly):
• (1) Creating the sounds in advance: in an interview with a physician, the person, while awake, briefly listens to candidate sounds and is asked for a distress score (SUDS, Subjective Units of Distress, answered by the person on a scale of 0–100). The physician observes the person's reaction and decides whether the sound may be used during sleep. The paper explicitly states that this step is a preparatory procedure for assessment and is distinct from waking therapy (prolonged exposure or EMDR).
• (2) Before sleep (evening): the person, while awake, listens once more to the fixed 30-second sound and is asked for a distress score. The upper volume limit is set according to each person's hearing. The sound is delivered through earphones placed in the ears.
• (3) During sleep: a technician trained in sleep scoring watches the EEG in real time, judges whether it is "stable slow-wave sleep," and enters this into the device. When slow-wave sleep has continued for 90 seconds, playback of the 30-second sound begins. The volume is raised gradually from a very low level. If the person leaves slow-wave sleep, playback is paused; if the person wakes up, it is stopped completely, and the system waits until slow-wave sleep has again continued for 90 seconds. It ends when the total time of sound played reaches 600 seconds (10 minutes).
• (4) The next morning: the person, while awake, listens to the same sound once more and is asked for a distress score.
• Key point 6: What this study examined: The main aim is "whether this procedure can actually be carried out with due care for safety" (feasibility). It is not a study to test whether it works. How to read the results is dealt with in Session 10.

Figure: Comparing the three methods

Hands-on task (hand calculation and table, 4 minutes)

1. If a 30-second sound is played up to a total of 600 seconds (10 minutes), how many plays of the 30-second sound is that?
2. According to the paper, when the night's recordings were later re-scored according to the sleep-scoring rules (the manual of the American Academy of Sleep Medicine, AASM), the proportion of time that was slow-wave sleep, with the 600 seconds of sound playback as the denominator, was on average about 68% across the 6 people (about 31% to about 89% depending on the person). The remaining time was time that looked like slow-wave sleep to the real-time judgment but was classified as N2 or other stages in the later scoring. The official values are those from the later scoring. The paper gives no threshold at which 68% would count as "good" or "bad," so here, read from it that "it differs greatly between people." Out of the 600 seconds, about how many seconds on average were within slow-wave sleep? About how many seconds for the lowest and the highest person?
3. Without looking at the comparison table above, write one line each on the "sound content" and "when played" of TMR, CLAS, and SES. Then check against the table.

Check questions

1. State the definition of SES in 3 parts (what sound / when / combined with what).
2. What is the difference between the sounds in SES and the sounds in TMR studies of PTSD?
3. Is having the person listen to the sound while awake before sleep and the next morning a form of therapy?

Summary

• SES = a standalone method, not combined with waking therapy, that presents sounds related to the person's own painful experience during slow-wave sleep (N3).
• Difference from TMR: the sound is not newly linked through learning or therapy (it is a sound from the person's original experience), and it is not combined with therapy. Difference from CLAS: it does not aim to strengthen slow waves and is not timed to phase (it looks at the stage).
• The procedure for one night: start the sound when stable slow-wave sleep has continued for 90 seconds, stop when the person leaves it, and end at a total of 600 seconds. The main aim is to examine feasibility.

Session 10: Reading a small clinical study critically — What can and cannot be said

Paper this session is based on: the paper from Session 9 (Ino et al., medRxiv 2026 / accepted in npj Mental Health Research).

Aims of this session

1. Be able to state the difference between a "feasibility study" and a "study to test whether something works."
2. See, through numbers, that in a study without a control, before-and-after changes can also arise from other causes (such as regression to the mean).
3. Be able to state, following the paper's wording, what can and cannot be said from the results of the SES paper.

Explanation

Question: "The distress score fell the morning after listening to the sound." Can we say from this that the sound worked? Answer: No. Without a comparison (control), there can be many other reasons why it fell. The paper itself says so.

• Key point 1: Stages of research: A new method is not suddenly tested for "whether it works." First, with a small number of people, researchers check "whether the procedure can actually be carried out, whether it can be continued, and whether problems arise" (feasibility study). To test whether it works, a comparative trial is needed after that: one with a sham-procedure group, in which which group each person is in is decided by lot (randomization), and in which neither the participants nor the assessors know which group is which.
• Key point 2: The design of this study: It is a study with only one group (single-arm) in which both the participants and the researchers know what is being done (open-label). 13 people consented, 8 proceeded to sound creation, sounds were finalized for 6, and all 6 of them completed one night of SES and the follow-up assessments. There were no adverse events (unfavorable events occurring during the study) judged to be caused by the study procedure. However, the paper states that because the number of people is small, its conclusions about safety are provisional.
• Key point 3: The procedure changed partway: For the first 2 people (version A), as a safety precaution, the sounds used during sleep were limited to sounds with "a distress score of 30–40." After confirming that there were no problems with these 2 people, the procedure was revised with the approval of the ethics review board, and for the remaining 4 people (version B), that upper limit was removed and replaced by a method in which the physician judged from multiple perspectives.
• The "distress score" here is the person's own rating of distress (SUDS, 0–100) when listening to that sound while awake. It is not the severity of the symptoms themselves. It is asked with the same sound on the previous evening (before sleep) and the next morning.
• The % here is the value obtained by calculating, for each person, "(next-morning distress score − previous-evening distress score) ÷ previous-evening distress score," and averaging it. As described in the paper, this change went up for the 2 people in version A (about +15% on average), and went down for all 4 people in version B, where the upper limit was removed (about −50% on average).
• However, this does not tell us whether it works. The paper itself states that because there is no control condition, none of the changes can be attributed to the intervention, and it cannot be said that things improved or worsened overall. The paper's conclusion is that the symptom outcomes remain "descriptive" and that a comparative trial with a sham procedure is needed. The preprint calculates the difference between version A and version B exploratorily, but it is a comparison of 2 people with 4 people, at different times, and not a comparison divided by lot. All that may be read from putting version A and version B side by side is that "the direction of change appeared to differ depending on whether there was an upper limit"; whether that is real is a question to be tested in the next comparative trial.
• Key point 4: Candidate reasons for the decrease (cannot be distinguished without a control):
• Regression to the mean: people who happened to be measured on a high day tend to move closer to their usual value when measured next (you will experience this in the Python below).
• Natural fluctuation: the paper states that some people showed changes of a similar size even between visits before SES.
• Expectation: because both the participants and the researchers know the procedure, the expectation that "it should work" can enter the answers.
• Difference in setting: the effect of spending a night in the hospital itself. The paper states that even when sleep structure on the hospital night is compared with that on a home night, there was no hospital night without sound, so the difference cannot be attributed either to the sound or to the setting. It also states that this study cannot tell whether slow-wave sleep was preserved or reduced.
• The sound itself: whether the content of the sound (that it relates to the trauma) matters, or whether it is listening to a sound in itself, is also not yet known, because there was no silent condition and no unrelated-sound condition.
• Key point 5: What can be said (following the paper's wording): What this study supports is that "delivering sounds related to the person's trauma during slow-wave sleep in people with PTSD can actually be carried out (feasibility)" and that "it provides grounds for proceeding to a comparative trial with a sham procedure." Anything beyond that (that it works, that it cures, that safety has been established) cannot be said.

Hands-on task (Python, 5 minutes, optional)

This is made-up data. For 1000 people, we add that day's fluctuation to their "usual distress score" and measure twice. Nothing is done in between.

1. Although nothing was done, the scores of the people who were high the 1st time fell by about 10 points. Why?
2. Looking at everyone, there is hardly any change. From this difference, what should you be careful about when "selecting people with high scores for a study"?
3. To distinguish regression to the mean from a real effect, what kind of group should be added?

Checklist for reading papers (use at the end of this session)

In a clinical study paper you are reading for the first time, look for the following 6 things. The answers for the SES paper are also given.

Check questions

1. In one line, the difference between a feasibility study and a comparative trial that tests whether something works.
2. Two reasons why, even though some people's distress scores fell in the SES paper, we cannot say "it worked."
3. In one line, following the paper's wording, what can be said from this study.

Summary

• A new method is tested in the order feasibility → comparative trial with a sham procedure. The SES paper is the former.
• Before-and-after changes without a control can also arise from regression to the mean, natural fluctuation, expectation, or the effect of the setting. The paper itself states that "none of the changes can be attributed to the intervention."
• What can be said goes only as far as "it can be carried out (feasibility)" and "grounds for proceeding to a comparative trial." It cannot be said that it works, that it cures, or that safety has been established.

Completion criteria, and what to learn next

Completion criteria (you have completed the course if you can do these 3 things)

• State, in one line each in your own words, the flow 1924 → 1994 → 2007 → 2009 → 2013 → 2020.
• Distinguish TMR, CLAS, and SES in terms of "sound content," "when it is played," "what it aims at," and "whether it is combined with waking therapy."
• In a clinical study paper you are reading for the first time, find the 6 items of the "checklist" in Session 10.

What to learn next

• A list of the lab's papers is on the Publications page of the lab website. You can go from the DOI to the abstract.
• For the overall picture of the lab's research, check what is publicly available on the Research page.
• If you want to learn more about reading figures and statistics in papers, also use the courses "Basics of reading the lab's papers" and "The central limit theorem and statistical tests" on the same "For undergraduates" page (if they are available).
• If you want to check things in a textbook, the chapters on sleep and on learning and memory in a standard neuroscience textbook (e.g. Kandel et al., Principles of Neural Science) correspond to this course.
• If you want to read about TMR research as a whole in English, starting from the reviews in Sessions 4 and 7 (Oudiette and Paller 2013, Paller et al. 2021) makes the flow easy to grasp.

Request for feedback (repeated)

This course is revised every time based on the voices of the people who use it. If you used it, please be sure to send us feedback.

• Send to: the address listed on the Contact page of the lab website
• Subject: "Course feedback: History of TMR and SES"
• What we would like you to write (4 points): (1) how many sessions you completed (2) parts that were hard to understand (3) parts that were good (4) how long one session took

If you ask a conversational AI to "Please help me put my feedback into shape," it will turn it into a form you can send as is.

Department of Neurophysiology, Kobe University Graduate School of Medicine (November 2026)

The courses are free for personal study and teaching. Please contact us before distributing modified versions. They are not medical advice.