NEXTUP and the Default Mode Network
In the last quarter of the twentieth century, two brain imaging techniques were developed that would dramatically change our understanding of how the brain works. Positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) have allowed scientists to look at brain activity while people are performing a whole range of mental tasks and create detailed 3-D maps of that brain activity. Using these new techniques, we have learned which brain areas are “turned on” during all sorts of mental activity—from staring at geometric patterns to viewing emotional pictures, reading, and memorizing word lists to having out-of-body experiences.
To produce these maps of brain activity, individuals are rolled into the center of a huge doughnut-shaped machine. Pictures of their brain activity are then taken, first while they’re just lying there resting and again while they’re doing some task. If we subtract the activity patterns seen while individuals are resting from the patterns seen when they’re performing the task, we get a picture of which brain regions are activated by the task—a map, if you will, of which parts of the brain actually perform the task. The use of fMRI to map brain activity in real time was an amazing breakthrough, and scientists quickly began mapping out dozens of brain functions.
As more and more brain imaging studies were published, it slowly became clear that something strange was happening. While turning on its own specific set of brain regions, each mental task also turned off other regions. At first this seemed quite reasonable. But as time went on, it also became clear that the regions being turned off were the same, no matter what task the participants were performing. And this made no sense.
Over time, however, Marcus Raichle, along with his colleagues at Washington University in St. Louis, realized what they were seeing. Scientists had been assuming that the activity pattern seen during quiet rest reflected the activity of a brain not doing anything. In retrospect, this was obviously a foolish assumption. Our brains are always thinking about something. Because of this, the brain areas that turn off whenever we start to carry out a mental task are the regions that do whatever the brain does when we’re “not doing anything.” Together these regions make up the default mode network (DMN), whose discovery has helped us appreciate just how true it is that the brain never rests.
When we look at the brain regions that make up the DMN, we find a sub-network that monitors the environment for important changes, watching out for any danger. Keeping us safe is probably one function of the DMN. But we also find a sub-network that helps us recall past events and imagine future ones, another that helps us navigate through space, and yet another that helps us interpret the words and actions of others. And these are the mental functions associated with mind wandering. Much of mind wandering involves hashing over the events of the day or anticipating and planning future events. Indeed, such planning has been proposed as a function of mind wandering. So it’s perhaps not surprising that mind wandering is associated with increased activity in the DMN. This appears to be a second function of the DMN.
The DMN is not a static structure, however. It changes based on what you’ve been doing earlier. Bob and his colleague Dara Manoach looked at how activity in the DMN changed after doing one of Bob’s favorite tasks: his finger- tapping task, which we saw in Chapter 5 involves learning to type the sequence 4–1–3–2–4 as quickly and accurately as possible. Young participants get a lot better in just a couple of minutes of practice, but then they plateau. A period of rest in the same day doesn’t make them any faster, but if they get a night of sleep and then try again, they become 15 to 20 percent faster. It’s another example of sleep-dependent memory evolution.
When Bob and Dara had participants learn the task while having their brains scanned, with periods of quiet rest before and after the training, they found that brain regions involved in performing the task were talking to each other more during the quiet rest after training than during the quiet rest before training. The DMN, which is normally measured during such periods of quiet rest, was altered by performing the task. And more important, the more the DMN was altered, the more improvement participants showed the next day. It was as if this new DMN activity was telling the brain what to work on once it fell asleep.
Indeed, much of the DMN is also activated during REM sleep, suggesting that the term daydreaming may be more appropriate than we thought. William Domhoff and his colleague Kieran Fox have gone so far as to suggest that dreaming, or at least REM sleep dreaming, constitutes a brain state of “enhanced mind wandering.” More recently, Domhoff has proposed that the neural substrate of dreaming lies within the DMN. When you put it all together, you get an exciting extension of our NEXTUP model. Whenever the waking brain doesn’t have to focus on some specific task, it activates the default mode network, identifies ongoing, incomplete mental processes—those needing further attention—and tries to imagine ways to complete them. Sometimes it completes the process shortly after the problem arises, making decisions without our ever realizing it. But at other times it sets the problem aside after tagging it for later sleep-dependent processing, either within or without dreaming. Several dream theories have suggested something like this—that dreaming helps us address areas of concern in our lives. The DRM might provide the mechanism for identifying these concerns, thereby determining what’s NEXTUP.
NEXTUP and Dream Function in Different Sleep Stages
With this added insight from the DMN, we can return to our question of how the function of NEXTUP might vary in different sleep stages. These differences are likely the greatest when looking at sleep onset. The hypnagogic period is a unique link between pre-sleep mind wandering and early sleep dreaming. A “fracture point” often takes place in sleep-onset mentation, wherein rational waking thoughts—inevitably about waking concerns or incomplete mental processes—shift into hypnagogic dreams.
Perhaps, then, it’s not surprising that Silvana Horovitz at the National Institutes of Health outside of Washington, D.C., has found that the DMN is active throughout the hypnagogic period. She also saw a dramatic increase in brain activity in visual processing regions after sleep onset. Other features of these dreams give additional support to the idea that these dreams have a unique role in NEXTUP. Hypnagogic dream reports from the sleep-onset (N1) stage of sleep are dramatically shorter than other nonREM and REM dream reports. They’re often clearly related to the thoughts you were having immediately before falling asleep, frequently evolving smoothly, if unpredictably, from those thoughts. Hypnagogic dreams are usually less bizarre and much less emotional than dreams from later in the night, and they often lack two features that are almost always present in other dreams; namely, self- representation and narrative structure. Much of the time these dreams are just unusual thoughts, or a random geometrical pattern, or a simple picture, like a landscape or a face.
This finding does not feel like fertile ground for the work of NEXTUP. Instead, these brief hypnagogic dreams appear to extend the DMN’s work into the sleep period, identifying and tagging current concerns for further sleep-dependent processing, and perhaps then beginning to identify associated memories for later consideration. But the very briefness of such dreams suggests that they can do little more than tag these memories, leaving more extensive processing for later in the night.
NEXTUP saves its best work for REM sleep. Compared to non-REM dreams, REM dreams are longer and more vivid, emotional, and bizarre, and they have more complex narratives. In addition, when people try to identify the waking sources of the content within these dreams, they report distinctly fewer episodic memory sources—memories of actual events in our lives that we can fully bring back to mind, essentially allowing us to relive the original event. For example, if you saw flying saucers in a nonREM dream, you might identify its source as a related episodic memory, saying, “Oh, those flying saucers looked just like the pizza I had for dinner last night.” In contrast, in a REM dream, you’d be more likely to say, “Oh, they looked just like pizzas; I love pizza,” thereby identifying a general semantic memory (I love pizza) instead of a specific episodic memory (I ate pizza last night). This example aligns with how we imagine NEXTUP working in REM sleep as it tries to use the simulated world of the dream to generalize from these memory sources and create a more integrated understanding of their meaning and importance (Figure 8.5).
By comparison, N2 dreams are shorter and less emotional, bizarre, and vivid. But perhaps the most telling difference is that the waking sources of N2 dream content tend to be more recent and more episodic—what you had for dinner, what your partner told you at dinner, who washed the dishes, and so on—and do not arise from less-specific “semantic” memories such as what you like to eat, what you often talk about with your partner, and which household chores are yours.
The memory sources used in constructing N2 dreams thus lie between the immediate pre-sleep sources of hypnagogic dreams and the very loose, associative links to semantic memories seen in REM dreams. The function of these N2 dreams is probably intermediate as well. Although REM sleep appears to be seeking weak, often unexpected, remote associations that might be usefully related to memories of unresolved concerns from the day, N2 dreams appear to search for more obviously related episodic memories from the recent past.
Presumably, this logic would hold for all sleep- dependent memory processing, both within and outside of dreaming. Anna Schapiro, a former postdoc of Bob’s now at the University of Pennsylvania, made this argument in a 2017 paper. She characterized the role of nonREM sleep (without reference to dreaming) as “an opportunity to recap the details of the day’s events, providing additional exposure to information that was recently acquired from the world,” and that of REM sleep as facilitating the “exploration of cortical networks containing long- term memories,” a description that matches the definition of NEXTUP as network exploration.
This separation of REM and nonREM functions provides a rationale for the normal sequence of sleep stages across the night. Each night begins with N1, moves to N2 and N3, and then moves to REM before cycling between N2/N3 and REM for the rest of the night. As the night progresses, nonREM decreases and REM increases, allowing the brain to seek out weaker and weaker associations and our dreams to become more and more bizarre.
This evolution of dreams across the night can be seen even on a shorter time scale in the hypnagogic period. In another study from Bob’s lab, using the arcade game Alpine Racer II, Erin Wamsley found that some dreams had strong, direct relations to the game, featuring unambiguous depictions of the game specifically, or of skiing in general; other dreams had weaker, more indirect relations to it, containing sensations, locations, or themes related to the game. When reports were collected at the start of the hypnagogic period, within 15 seconds of sleep onset, dreams were eight times more likely to show direct incorporation of the game than indirect incorporation. But after just 2 minutes of sleep, the rates of occurrence had become the same for both direct and indirect incorporations. Another group of participants were allowed to sleep for 2 hours before any reports were collected. After the 2 hours, they were awoken and then allowed to fall back asleep. They were then awoken again within 2 minutes and their sleep-onset reports collected. Now the ratio of indirect to direct incorporations was five times higher than for those collected right at the start of the night.
Interestingly, Stuart Fogel at the University of Ottawa in Canada had participants practice the Nintendo game Grand Slam Tennis before collecting eight sleep-onset dream reports that night. Overnight improvement seemed to depend on how similar dream incorporations were to the actual game for the first four of these sleep-onset dreams, but not for the last four. Perhaps only the earlier, more direct incorporations successfully tagged the game memories for NEXTUP.
This all makes sense. Whether we look at how pre-sleep activities are linked to dreaming in the hypnagogic period or how they’re connected to dreaming in all the sleep stages across the rest of the night, dreaming appears to play an important role in how memories are selected and how they subsequently evolve across the night.
Excerpted from When Brains Dream: Exploring the Science and Mystery of Sleep by Antonio Zadra and Robert Stickgold. Copyright © 2021 by Antonio Zadra and Robert Stickgold. Available from W.W. Norton & Company.