I thought I’d write a post on a topic I love to talk about: dreams. You wouldn’t believe this dream I had last night: there was this nutria and he was riding on a surfboard made of cheetos and…wait sorry I’m getting off topic. What I really want to talk about is the neuroscience (not the usual pseudoscience) behind dreams. I study memory and a lot of work has shown the heavy importance of sleep in forming memories, but dreaming is a bit more of a mystery. Does it serve a purpose, or is it just an accidental offshoot from memory consolidation? To be honest: we really don’t know yet. But we do know some interesting pieces here and there that serve as first steps to unraveling the mystery of dreaming.
To talk about dreaming you have to start with sleep. We don’t even really know the purpose of sleep. For certain you don’t do very well without it and can even die from enough insomnia, but why sleep is so important that our brain uses approximately 1/3 of our lives doing it is a bit of a mystery. Oh, and this isn’t just true for us, as far as I know all higher animals sleep (including sharks and dolphins contrary to popular belief—they just do it half a brain at a time). A recent theory proposed that neurons need to be turned off at night for “prophylactic cellular maintenance,” which seems reasonable considering long-term sleep deprivation appears linked to higher mortality rates, and your brain IS shuttling around all sorts of charged ions all day. If you want to read more about why it’s a good idea for YOU to sleep more, check out my buddy Pascal’s pensees on sleep.
While we haven’t proven why you need to sleep, it is clear that it helps us form memories. As far as I know the first papers that linked sleep to memory were from the mid-90s out of the Weizmann Institute in Israel. Karni et al showed in 1994 that memory consolidation (the process of short term memories being stored) specifically happened during REM sleep. They did this by disrupting people as they entered REM, which led to their memories for a visual task being compromised. Many other studies have gone on to show how sleep improves memory, even when the memory tests are significantly further away in time. If you have subjects learn a finger-tapping task (e.g. press keys in the order 5-2-4-3-1 and remember to do it hours later) and then test them after 8 hours of being awake versus 8 hours awake plus 8 hours sleep, the group that slept will be both more accurate and faster at repeating the correct key order.
Now we can start talking about dreams a little bit. Radiolab had a good episode on sleep with a segment on dreams that I’m referencing some anecdotes from. Robert Stickgold started things off in 2000 by publishing the first (real) scientific study on dreaming in decades (as Wikipedia points out the last notable one from 1977 has ‘compromised factual accuracy’, which sounds about right). Stickgold personally had noticed that when he went hiking when he later fell asleep he would dream of continuously climbing up slopes. Knowing that trucking a bunch of undergraduates to whatever mountains are near Harvard and then bringing them back to test them falling asleep wasn’t feasible, he sat on the idea for a while until someone mentioned Tetris to him. Apparently when you play Tetris a bunch you see falling blocks connecting to each other in your sleep. He quantified this with some undergrads and found that a large percentage of both ‘experts’ (I would have liked to see the signs posted for that: ‘Wanted: Tetris experts’) and novices had dreams about Tetris after playing for a few hours. More anecdotally, Stickgold also described how people had more abstract dreams when they slept longer outside the lab. As opposed to the more literal Tetris dreams measured by Stickgold when people were awoken soon after falling asleep (these quick dreams when falling asleep are referred to as hypnagogic dreams), subjects returning days later would describe things like ‘[I was] thinking about a project I have for work that involves designing a garden space indoors and as I was thinking about it in my mind little Tetris pieces kept falling down into the garden spaces’. This implies that the Tetris dreams are going from being replayed quite literally in the beginning of sleep to being incorporated into abstract episodic memories associated with other aspects of life later in the night. REM sleep is also enriched the longer you’ve been sleeping, which is taken as a reason for increased dreaming right before you wake up. I, for one, have tried on many occasions to fall back asleep in the morning to reenter a sweet dream!
Also interestingly, they got a group of amnesiacs to play. While the amnesiacs didn’t remember playing the game or the experimenter and didn’t get any better at the game like normal subjects did, some of them did still report Tetris-related dreams when awoken right after falling asleep. This implies that dreaming can take place outside the hippocampus/medial temporal lobe, which is the seat of the malfunctioning brain tissue in those amnesiacs. This is a bit of a surprise, since we think of the medial temporal lobe as the initiator of the short to long-term consolidation process of memory (see step 3 in a few paragraphs to understand what ‘consolidation’ is). Plus, studies of rats ‘dreaming’ find many of those seemingly dream-related signals in the hippocampus (the best part of the medial temporal lobe. Just kidding. ALL the brain is cool! But mostly the hippocampus).
Speaking of those rat studies: Matt Wilson’s work was also featured in that Radiolab episode. Wilson discovered a process now termed ‘memory replay’ back in the early 90s. He describes in the episode how he was training a rat to run around a track and the hippocampal neurons would fire spikes in order: one cell after another cell after another cell. As a neurophysiologist you get used to the signature sound of these events as the voltage measurements are attached to a speaker. While recording from these rats he was surprised to hear the cells firing in sequence when he hadn’t put any food in the maze to motivate the rat to run. When he looked over at the rat it had dozed off on the track. The ‘replay’ is these groups of hippocampal cells firing in order while sleeping in the same way as they were when it is awake. Here’s a visualization:
“RUN” is what 9 hippocampal neurons do when the rat is running through a maze. Notice how they fire one after another, with each vertical tick representing a single neuron firing. “SLEEP” is from when the rat is sleeping. When the same 9 neurons are recorded, they fire in the exact same order, an indication what the rat is thinking of when it was previously RUNning. If you look closely you’ll notice the replay timescale is 0.2 s. while the running timescale is 1.0 s. This means the replay is done at approximately 5x the speed of the actual event, which is believable to me since dreams often seem like they’re not in real-time, but counter to some other dream research that I’m not even sure has been published.
If you had the rat run in a different enclosure those same cells might not even be active. And if they were they almost certainly would not be firing in the same sequence. This gives us some idea of why the “dream” is taking place—the order of cell firing must be important for memory and the rat is replaying the order to store them in memory. We don’t know exactly why the rat has to replay them for the memory to hold, but future work has indicated that these sequences of cell firing also takes place in the cortex, an indication that memories are being transferred to this region for long-term storage.
This is in line with the canonical theory of how memories are consolidated, which I’ll describe briefly since it’s pretty interesting and will help explain some dream theory later:
1.) you experience the stimuli with the cortex on the outer parts of your brain (e.g. your auditory cortex first interprets Tubthumping by Chumbawumba).
2.) the stimuli are passed on to your medial temporal lobe (including the hippocampus) for contextualization. This region unites the disparate senses into episodes and also can link them to other important facets of the experience like emotion/reward/anxiety (e.g. you remember hearing Tubthumping at a super awkward but tasty middle school pizza party).
3.) the important stimuli are selected to be kept and the cortical region that originally experienced the stimuli then stores them permanently (i.e. your auditory cortex still remembers what Tubthumping sounds like, sadly, even to this day). This is called ‘consolidation’ of the memory.
Therefore, the replays might work as some kind of signal being sent from the hippocampus indicating which memory was important enough to keep, with the sequence of cellular firing possibly the code itself that represents that specific episode. There’s evidence the hippocampus then later indexes these memories so you can form associations with other memories, sort of like a really, really good Pandora Radio for your experiences that robustly finds the right memory when given a related experience as a suggestion. You know, much like Pandora Radio fails to do for music.
Anyway, in the Radiolab episode Wilson goes on to speculate that the actual physical replay of cell firing that happens when the rat is acting out the experience is a way to combine information from various memories for storage. He claims they’ve seen a rat run around one track, run around a second track and then have replays that combined the sequential firing from both tracks. Frankly, there has been no real scientific evidence of this from what I can find. I actually even emailed him once and he pointed to a Loren Frank paper that didn’t really describe this idea. So, while it’s a nice thought, we’re not really sure if these replays actually are done to help make connections between distinct experiences*. More on this later.
One thing that has confused these ideas is the discovery of more immediate replay. Rats won’t just replay events when they fall asleep, they’ll also do it right after experiencing the event when they’re standing still. Strangely, some of these replays are even in reverse, with the cells firing in the opposite order in the awake replays from when they were running on the track. So while replays are likely happening when you dream, they don’t seem exclusive to dreams themselves but are more ubiquitous in any kind of memory-related process.
A new paper from the Wilson group in 2012 took another step by actually biasing the rat’s ‘dreams’. They trained rats to go left or right in an enclosure for reward (often chocolate sprinkles—I’m told rats love chocolate sprinkles) depending on which of two sounds they heard. One sound would net them reward for going left, the other for going right. They found that while the rat was sleeping, if they played these tones they were able to get the rat to preferentially replay (or think about) running left or right in response to the left and right tones. Again, like in previous work, these replays were sequences of cells firing one after another. They were even able to estimate where the rat was running in its dream by matching up the replay to the sequences from when it was running in the enclosure. Here’s their graph showing this Inception-like dream-reading, possibly even colored to remind you of Inception:
(from http://www.nature.com/neuro/journal/v15/n10/fig_tab/nn.3203_F7.html …and IMDB)
I don’t want to go into details about their model, but the darker colors basically indicate that the neurons predict the rat to be on the right side of the track in this case. Again, this doesn’t tell us a ton about why events need to be replayed during dreams. But it does indicate that dreams can be influenced by external stimuli even while sleeping, which hints at your brain assigning valence to certain memories that it decides are important during sleep. Interestingly, all this ‘dream biasing’ happened during rat’s slow-wave sleep (SWS). This is a bit confusing since people usually remember dreams when awoken from REM sleep, again the reason you tend to dream right before you wake up as you get more and more REM sleep throughout the night. Apparently in humans it has been shown that biasing dreams doesn’t work during REM sleep, and even though you don’t necessarily remember dreams during SWS they’re still happening. This speaks to how ALL stages of sleep are important, and why all that crazytalk about the da Vinci sleeping pattern doesn’t make a lot of sense. But that’s really for another blog post.
Alright, so now that you’re up to speed on some recent dream research, you still probably don’t feel all that satisfied. Sure, dreams could be important for memory, but what’s the purpose of semi-consciously rehearsing only select events? And why are they so freakin’ weird? And what about all those stories of people having insights in dreams (e.g. Mendeleyev claims the layout for the periodic table came to him in a dream), isn’t that related in some way? I’ll start on the speculation train in a moment, but first one more supercool study to help answer this last question.
Jan Born’s group has a paper from 2006 titled simply: “Sleep Inspires Insight.” They had people learn a numbers game where they had to predict the next numbers in a sequence based on previous strings of numbers. What they were not told was there was a ‘hidden rule’ that once discovered would allow them to figure out the next numbers in the sequence considerably faster if they figured it out. And, as the title spoiled for you already, this rule was discovered significantly more often by people that were able to sleep on it for 8 hours versus people that did the task again after 8 hours of being awake. That is so cool—I love it when science backs up anecdotal intuition!
Now you can see where Matt Wilson was going with his idea about rats combining replays of different tracks together into single representations: he was trying to describe how these distinct memories could be combined and analyzed in an offline state while the rat was ‘dreaming’. This still doesn’t explain why the dreams have to be conscious, since a lot of our problem-solving takes place subconsciously, but it does assign them a useful enough purpose for us (and our wall-headbutting pets) to be actively rehearsing daily events during sleep.
I also think the word ‘offline’ is key in that previous paragraph. One idea is that your brain can only process so much remembered information when it’s currently engaged in inputting new information. Since, as I described before, the same bit of cortex that first experiences the memory is the place where the memory is stored long-term, there’s clearly going to be some interference screwing things up as your brain tries to pull double duty. For example, you probably have experienced how hard it is to remember how a song goes when loud music is playing. The reason is your auditory cortex is being engaged by the loud music, likely by many of the same cells that hold the memory of the previous song you’re trying to remember, and this makes pulling the song out of your head significantly more difficult than when everything is silent. It’s actually quite impressive you can do this at all—it speaks to the robustness of the cellular encoding process. But it would certainly be a lot easier for your brain to recall ‘true’ memories without distraction while nothing new is coming in–like when you’re sleeping. Further, it would be particularly useful for there to be no new interference coming in if you were trying to combine multiple memories together and form new insights. This includes stimuli like songs playing, but could also include such distractions as emotional reactions not necessarily relevant to a memory you experienced earlier.
This is in line with new work by Matt Walker (not Matt Wilson—I know it’s confusing) that postulates that REM sleep serves as ‘emotionally safe’ periods of reactivation. He found that people had reduced responses to emotionally charged photos after sleep, which also showed up as a reduction in activity in their emotional processing centers from fMRI scans. To directly quote him, “We know that during REM sleep there is a sharp decrease in levels of norepinephrine, a brain chemical associated with stress,” Walker said. “By reprocessing previous emotional experiences in this neuro-chemically safe environment…we wake up the next day, and those experiences have been softened in their emotional strength.” Could replaying such events with reduced emotions during dreams make ‘truer’, more objective memories?
Alright, over 2500 words later and maybe dreaming makes a bit more sense to you. As I said in the beginning (and harped on a few more times throughout), we still don’t really know why semi-consciously recalling daily events during sleep is useful. It’s even less clear why crazy, invented dreams occur. But the evidence at least points to some kind of process that both selects for important memories and does some problem-solving to boot while you’re not being distracted by new experiences. Further, more and more evidence seems to indicate that the brain is consistently biasing and predicting future events, so the possibility that these dreams might be some kind of exploration into the vast realms of ideas we form every day seems like a reasonable one.
*I stumbled upon this paper from David Redish’s group recently where they managed to show neuronal signatures of rats envisioning “shortcuts” on parts of a maze that they had never traversed. This is pretty good evidence that the brain is using hippocampal replays to collaborate past experiences to imagine future events, although doesn’t necessarily say that dreams have anything to do with such problem-solving “nexting”.