I'd like to begin with a definition change to assist in understanding the limits of "resting state" fMRI. I'll continue to refer to rs-fMRI as the act of acquiring a block of fMRI data - let's say six minutes' worth - in the absence of any specific externally presented task during the acquisition, with one small exception: we'll assume that the subject is presented with a simple fixation cross and is asked to keep his eyes open. (More on visual and auditory effects on rs-fMRI below.)
Now, though, I'd like to rename the mental activity that is happening during the rs-fMRI acquisition period. The definition of "rest" is tricky because it depends on so many state-dependent factors. What if I'm worried about an upcoming exam? What if I'm hungry and distracted by the need for food? Just because I'm awake during both doesn't make them equivalent periods of "rest," even if I am lying in the scanner staring at a fixation cross in both cases. And, because I want to distinguish between such periods without an explicit task in the discussion to follow, I'm going to term what we do today as "free-thinking state" fMRI, a term used by Cindy Lustig from WashU in a 2003 interview about some of her work.
Okay, so now we have a new definition to work with. How might this state of free thinking be manipulated without fundamentally changing the goal, which is (I assume) to map the largest networks that arise intrinsically, in the absence of explicit, externally driven, goal-directed behavior (in the form of some sort of task presentation)?
Caveats: every fMRI experiment should have some!
Before we look at intentional manipulation of a subject's brain state, let's first review the confounds and limitations that have already been unearthed when it comes to conducting free-thinking state fMRI experiments. These are effects that have been shown to change the networks detected in standard rs-fMRI experiments. (See Note 1.)
I think most people would accept that the resting state - as we presently define it for fMRI - is nothing of the sort. At a minimum it is a visually directed attention task (or is designed to be!). That little cross floats right there in front of you, a permanent reminder to "stay awake, look straight ahead and don't think of anything specific." Yeah, sure, except the part about looking straight ahead! After all, there's a virtual instruction screen, if you will, right here in front of my nose! Surely, then, the fact that a subject is conducting a particular task - visual attention - is biasing all the rs-fMRI results?
Apparently not. van Dijk et al. and Yan et al. have both shown that the default-mode network (DMN) differs when subjects have their eyes closed versus eyes open, but in the latter condition it mattered less whether the subjects were viewing a fixation cross or just looking at whatever happened to be visible inside the scanner bore. Thus, free thinking with eyes closed seems to be significantly different than with eyes open, a result that fits with our subjective experience of the two states. (See Note 2.)
But one might also think that without a fixation target a subject might make saccades all over the place in a desperate attempt to find something interesting! In my personal experience, I find my gaze wanders occasionally whether there's a fixed target or not. But I am an experienced subject so for a naive subject you'd have to think that a fixation target would be prudent, lest they go on a visual search throughout the magnet bore, the end of their nose, etc.
Why then, no major difference between fixation target and passive eyes open (no target)? Perhaps the attention load required for a visual target is so low as to be undetectable, when compared against a similar protocol without a defined visual target, i.e. free viewing, in an awake, alert state. Anyway, perhaps a vision scientist will be able to show that there are network differences between target and non-target rs-fMRI, perhaps using an eye-tracker output as a covariate.
Work by Benjamin et al. showed that attending to the scanner noise, even when instructed not to, affects the default mode network, too. Subjects were told to attend or not to attend to the noise, or were given no specific instruction either way. When told to attend to or ignore the noise, activity in the dorsal medial prefrontal cortex increased compared to no specific instructions at all. (Whatever you do, do NOT think of an elephant.... Doh!)
As the authors point out, instructions might be especially critical for certain clinical populations, for first-time subjects and for children, although my suspicion is that we need to be very careful with our instructions whoever the subjects are. But there is something else going on here, and it's subtle. Like my silly elephant experiment, priming can be very powerful and difficult if not impossible to overcome. I'll return to this point below.
Before I leave the effects of scanner background noise I do want to make one more observation. A result that shows that being told to ignore the scanner noise makes subjects more prone to attend to it tells us little about the effects of the scanner noise itself, i.e. in terms of its bottom-up effects. We can't yet do rs-fMRI in the absence of scanner noise; a silent fMRI scanner hasn't yet been invented. For all we know the background clanking could have a large effect on everything we measure.
Indeed, precisely how any given subject might involuntarily attend to the scanner noise (when given no explicit instructions concerning noise) will likely vary depending on who you're scanning. If you have a penchant for securing the services of your lab mates as "normal volunteers" then you'd better hope that they don't have too much fMRI experience! If they are like me and know a little bit about pulse sequence design you could find them subconsciously (or not) attending to the characteristic sound patterns being played out. I know I do!
Prior task hangovers
The final caveat for today's rs-fMRI studies is the subject's recent history. Turns out if you scare the crap out of your subject and then expect him to act naturally for six minutes and think all sorts of random, uncontrolled thoughts, he's probably not going to do it. He's too busy trying to calm down, not be angry at you, etc. etc.
Waites et al. showed a few years ago that the effects of a prior task will indeed hang over into an rs-fMRI block. This is an immediate limitation for rs-fMRI studies as conducted today. It implies that for tasks with strong emotional valence, or even for tasks that might involve a lot of problem solving, or perhaps any task whatsoever, that the rs-fMRI shouldn't be done last. In many labs the norm is to acquire the so-called resting state experiment at the end of the session, after whatever the main experiment happened to be. Talk about the potential for bias! I see only two alternatives to ameliorate this hangover effect: either you run each of your subjects' rs-fMRI acquisitions in a separate session, or you acquire the rs-fMRI first, before any task blocks.
But isn't there something subtle and interesting here? This hangover effect is reminiscent of the priming effect with the auditory instructions we saw above: the otherwise "free thinking" state is being manipulated, albeit unintentionally. We could do our level best to control all of these confounds - simple instructions without explicit reference to the scanner noise, run the rs-fMRI acquisition first, etc. - or maybe we could use these effects.
What are the limits of free-thinking state fMRI?
Now we get to the interesting bit. How far can we manipulate this free thinking period, during scanning, such that it might allow us to explore intrinsic brain networks in new ways? As we've seen above, we need to be especially aware of confounds, but if we can control for such effects then is it reasonable to attempt to manipulate the content of a subject's otherwise unconstrained thoughts during the rs-fMRI scan? And if we do, shouldn't we refer to this version of rs-fMRI as guided-thinking state fMRI?
If we are attempting to map the networks that arise naturally as people's brains hop from thought to thought in an uncontrolled manner, the one thing we obviously can't do is supply an instruction that would negate the thought-hopping. The moment we tell a subject to, say, think about elephants for the next six minutes, we may well get various networks going on- and offline, and there may indeed be a significant amount of language, memory, visualization, etc. But surely it's a task with an attention load, not true free thinking. Presumably, then, we need to strike a careful balance between limiting thoughts by the imposition of a particular task and allowing free range, subject to a (previously invoked) bias.
If the goal is to map the DMN across traits or disease conditions, e.g. schizophrenics versus normal controls, then we could probably be satisfied with the current approach to rs-fMRI, adopting some best practices: use a fixation cross, don't tell the subject any explicit instructions about scanner noise, and make sure the rs-fMRI acquisition doesn't follow hot on the heels of some emotionally powerful task block. This approach should minimize (known) confounds.
But if we had a hypothesis that certain (powerful?) stimuli could have a "hangover" effect such that they might alter the networks detected in a subsequent "free thinking" period, we might have a different tool with which to probe the brain. Maybe we have subjects play violent video games for an hour before putting them in the scanner, for example. Or we have subjects read a story about someone doing heroic charity work, then put them in the scanner. Would we expect this biased state to interact with the subject's innate traits to modify the free-thinking networks in some measurable way? And what about controls? When we're investigating traits we can compare one group against another. If we are intending to manipulate states, however, then surely we will need some sort of "control priming," against which we hope to interpret the networks detected during the guided-thinking state.
Prior work in guided-thinking state fMRI
There are several examples in the recent literature that can be categorized as what I'm terming 'guided-thinking state' fMRI. In 2008, Harrison et al. compared resting-state networks during a self-induced sad mood to a separate period when the subjects' self-reported mood was neutral. (See Note 3.) A recent article by Eryilmaz et al. demonstrated the differential modulations of a minute's worth of either joyful, neutral or fearful movie viewing on the subsequent ninety seconds of "resting state" data. In the cognitive realm, Grigg and Grady have shown that certain components of resting networks differed for rs-fMRI acquisitions acquired either side of 45 minutes' worth of a set of cognitive tasks, while Barnes et al. assessed the temporal dynamics following an n-back task, finding that "recovery" to prior baseline could take up to six minutes following a block of 2-back working memory tests. (See Note 4.) There are several other papers that have investigated task hangover effects - do a citation search for the Waites reference and you'll dig them all up - but I'm not covering them here because they used rest blocks with much shorter durations (usually less than 30 seconds) than the standard rs-fMRI protocol most of us use. (It's not clear to me that when the resting period being acquired is less than 4 minutes long that the same things are being measured.)
Controls for manipulation of states
Being able to define, and achieve, a suitable control state for subjects is one of the biggest challenges of guided-thinking state fMRI it seems to me. The literature studies cited above took care to develop appropriate controls but there were still issues to deal with. For example, in the Harrison study the neutral mood scan always preceded the sad mood scan. Presumably it was felt that randomizing the scan order across all subjects would have introduced a new confound - the possibility that the sad mood could hang over into the neutral mood state whenever the sad mood scan was to be performed first. (They did take a lot of care to neutralize the effect of the scanner environment itself, having each subject spend 30 minutes in a mock scanner prior to the real scanning session.) One wonders whether separate scan sessions and randomized scan ordering would have been a better approach. However, that would have generated the possibility of different arousal levels, fatigue levels, etc. that could interact with, even overwhelm, the effect of interest! I leave it to you neuroscientists to figure out the best ways to do these experiments! Overall, though, the Harrison study nicely highlights many of the complexities that guided-thinking state fMRI generates.
Some manipulations might not have obvious control states at all. Take introspection. Let's suppose that we are interested in the intrinsic networks associated with first-person cognition. Would it be meaningful to compare a block of rs-fMRI acquired while a subject tries to think in a first-person narrative against another block where he tries to think in a third-person narrative (about family members and friends, for instance)? Would the third-person narrative be too first-person oriented - it is the subject's family, the subject's friends, etc. - to be distinguishable from the other state? And would both manipulations in any case violate the assumption of no task engagement and thus overcome the DMN and other non-task networks...? Lots to think about.
Thoughts for you to take away
I hope this post provokes some thoughts on free-thinking state fMRI, and perhaps some not-so-free-thinking state fMRI. If your goal is to map DMN and other resting networks in the absence of as many state-dependent biases as possible - traditional rs-fMRI, if you will - then here are some best practices to keep in mind:
- Use a fixation cross and give simple instructions, such as "Keep your eyes open, relax, try not to move, try to stay awake and just let your thoughts wander."
- Also, be consistent with your instructions! Why not present the instructions to the subject on a screen? Alternatively, read a prepared instruction cue over the intercom. This will avoid you accidentally saying "Don't pay any attention to the scanner noise," or introducing some other unintentional confound.
- Try not to acquire your rs-fMRI block right after task blocks. Perhaps acquire rs-fMRI after the structural scans and before the task-based fMRI components of your experiment.
- Finally, when you are trying to interpret the meaning of your results, the effects of instructions and prior tasks/events may need to be taken into account; don't overlook the potential confounds.
It would be useful if the methods sections of papers accurately report what was done for rs-fMRI. Let's make that policy at the very least! I would encourage people to state in their entirety the instructions given to subjects, as well as report the order in which the rs-fMRI block(s) were acquired. (Some authors are already doing both of these things - thank you.) Also note whether the subjects came from doing a behavioral task outside the scanner, from a session in a mock scanner, and so on. All of these events will likely affect the mental state of your subject. They should, right? If we believe any of this fMRI malarkey then by inference we must expect the subject's mental state to be reflected in the neurovascular signature we're measuring with BOLD. Otherwise we might as well pack it in and go do something else!
1. Anybody else wonder whether the act of sticking a person into a dark tube that makes a lot of noise might not be the most typical state to measure "normal" brain responses in? For many subjects the stress of being in an MRI for the first time - "What if there's something wrong with my brain?" - and claustrophobic environment must surely be firing off all manner of brain responses that should be considered atypical in a true resting state. I wonder how networks would differ between subjects trained in a mock scanner, say, versus naive subjects? It would be important to measure and control for sympathetic physiological responses (respiration depth/rate, heart rate, maybe even blood pressure). This is one of the big differences for rs-fMRI and task-based experiments: without control events/blocks we have no natural ability to control the effects of the scanner itself on what we're measuring. As a certain Australian liked to put it: "Danger! Danger!"
2. I cover auditory attention later in the post, but it arises here, too. When we want to attend very closely with one of our non-visual senses - think intricate music, a complex wine, the texture of fine silk - we may close our eyes to maximize the sensation. We're very visual animals; shutting off the bottom-up visual input seems to play big dividends across the rest of the sensory pathways. So, when subjects have their eyes closed, is the principal difference to eyes open rest the absence of the visual input or an enhancement of other non-visual sensations? Subjects could immediately be paying more attention to the scanner sounds, for instance, or the scanner vibration, or even the scanner's smell.
3. In the Harrison et al. study, subjects listened to music before the rs-fMRI scan, with the particular music being selected to assist either in the generation of a neutral mood or a sad mood. Fair enough. However, the rs-fMRI acquisitions were then conducted with the subjects' eyes closed. This would have assisted in the recall of neutral or sad events, which is how the mood induction was facilitated, but it presumably changes the networks, as we've already encountered.
4. Barnes et al. used eyes closed rest for their rs-fMRI periods. One wonders whether the recovery of network dynamics would have been faster or slower with eyes open. Also, I have to wonder about sleep. It's hard enough to stay awake in a scanner without being asked to close your eyes. They relied on subjects' self reports - a notoriously unreliable way to monitor sleep - to avoid it. Again, would the recovery dynamics differ if subjects had their eyes open and were monitored to ensure the absence of sleep? If ever there was justification for using an eye tracker or an EEG to monitor alpha band activity, this would be it!
A Barnes et al., "Endogenous human brain dynamics recover slowly following cognitive effort." PLoS ONE 4(8):e6626 (2009).
C Benjamin et al., "The influence of rest period instructions on the default mode network." Front. Hum. Neurosci. 4, 218 (2010).
H Eryilmax et al., "Impact of transient emotions on functional connectivity during subsequent resting state: a wavelet correlation approach." NeuroImage 54, 2481-91 (2011).
O Grigg & CL Grady, "Task-related effects on the temporal and spatial dynamics of resting-state functional connectivity in the default network." PLoS ONE 5(10):e13311 (2010).
BJ Harrison et al., "Modulation of brain resting-state networks by sad mood induction." PLoS ONE3(3):e1794 (2008) .
KR van Dijk et al., "Intrinsic functional connectivity as a tool for human connectomics: theory, properties, and optimization." J. Neurophysiol. 107, 297-321 (2010).
AB Waites et al., "Effect of prior cognitive state on resting state networks measured with functional connectivity." Human Brain Mapp. 24, 59-68 (2005).
C Yan et al., "Spontaneous activity in the default mode network is sensitive to different resting-state conditions with limited cognitive load." PLoS ONE 4(5), e5743 (2009).