Regulation of the Neural Circuitry of Emotion by Compassion MeditationEffects of Meditative Expertise

Antoine Lutz, Julie A. Brefczynski‐Lewis, Tom Johnstone, Richard J. Davidson•View original

OverviewBalancedalloy voice

Imagine you are sitting quietly, with your eyes open, doing nothing but trying to call up a feeling: a warm, steady wish that whoever is suffering out there finds relief. Not a story about someone specific, just that stance. The big question here is simple and bold: if you voluntarily generate compassion, does your brain switch into a different mode for processing other people’s emotions? And if you have trained that capacity for years, does the switch flip harder? Antoine Lutz and colleagues set this up against what we already know about empathy for pain. A pair of regions—the anterior insula and anterior cingulate cortex—light up reliably when we share someone else’s distress. But empathy isn’t just raw feeling; it often recruits a broader social cognition network that helps us infer and imagine minds. These areas include the temporoparietal junction, also known as TPJ, the posterior superior temporal sulcus, known as pSTS, the medial prefrontal cortex, and the posterior cingulate along with the neighboring precuneus. So, the team asked: if compassion meditation is a learned skill, can entering that state dial up both the feeling circuits and the perspective-taking ones? To test it, they brought in two groups. Long-term Buddhist practitioners with more than 10,000 hours of meditation—true experts—and age and gender-matched novices with no prior experience, who received a week of brief training before scanning. In the scanner, everyone alternated between resting and actively generating compassion, keeping their eyes open and fixated. During both states, they heard short, two-second human sounds drawn from a standardized set—some negative, some neutral, and some positive. The beauty of that design is that the same stimulus hits two different internal states. After each run, participants rated how strong their meditation had felt on a nine-point scale, which later allowed the team to compare “good” versus “poor” blocks within each person. The imaging was standard but careful. Functional data came from a three Tesla scanner with a two-second repetition time and roughly four millimeter voxels. Two participants were excluded for motion, and the main imaging comparisons ended up with 15 experts and 15 novices. They modeled the brain’s response to each sound over a 20-second window using a set of sine functions—basically a flexible way to capture the rise and fall of the blood-oxygen signal without assuming a single shape. They also recorded pupil diameter at 60 hertz for most participants—13 controls and 10 experts—to index arousal. That matters because a bigger pupil can mean you are more alert or emotionally engaged, and you don’t want to confuse that with empathy per se. Pupil size was folded into the analyses as a covariate, effectively subtracting out arousal-linked variance from the brain signal. To keep false positives in check, they used Monte Carlo simulations to set cluster size thresholds for their statistical maps. Start with the broadest contrast: compassion versus rest, averaged across everyone. When people entered compassion, the anterior insula and anterior cingulate ramped up. So did a wider network tied to mentation and social cognition—the temporal poles, the posterior superior temporal sulcus, the temporoparietal junction, the medial prefrontal cortex, posterior cingulate, and precuneus. Think of it as both feeling and thinking about minds getting a little more engaged. There was a clear right hemisphere bias to these state effects. Importantly, just comparing experts to novices without considering state didn’t show a global group difference. The interesting action lived in the interaction: how state and stimulus valence played out differently by expertise. That is where the right anterior insula stole the show. In a voxelwise analysis, a sizable cluster there—about 3,667 voxels, corrected—showed a three-way interaction of group, state, and valence. What does that mean in plain terms? Experts, when meditating, reacted more strongly to emotional sounds than to neutral ones, and that boost was bigger than what novices showed. Focus on negative sounds, and you see it cleanly: the right insula response during meditation versus rest was higher in experts than novices, with a group difference that cleared conventional significance, t with 28 degrees of freedom just over 2. The more striking part is that this wasn’t just an “experts versus beginners” story. Within individuals, when they said a block felt like strong compassion, the insula response to emotional sounds was larger than in their own poor blocks. Across a subset, that verbal report effect in the right insula and the cingulate was statistically reliable, with an F around 6.8. So, the brain signal tracked the felt intensity of compassion. Now, you might be wondering about arousal. Pupil diameter did increase during meditation relative to rest—that was a main effect, with an F just over 5 in a 16-person analysis. Experts, as a group, showed a bigger state-related pupil jump than novices, with a state-by-group interaction near F equals 11 with a small p-value. Bigger pupils leading to a bigger insula response—was it all just alertness? The team checked. Even when they regressed out the pupil effects, the right insula still carried a significant meditation-linked boost to emotional sounds. In fact, the correlation between the meditation-related pupil increase and the insula boost was moderate, around r equals 0.54, suggesting that arousal and empathic processing moved together but weren’t the same thing. That is useful. It means the compassion state changes how the brain treats social emotion beyond simply waking you up. The story didn’t stop at the insula. Experts, specifically during compassion, showed heightened responses in a set of regions we use to read social cues and infer intentions. The right temporoparietal junction and the right posterior superior temporal sulcus stood out, along with the posterior cingulate and precuneus. The amygdalae—the almond-shaped structures sensitive to emotional salience—also joined in, as did the right inferior frontal gyrus and a swath of premotor cortex around Brodmann area six. One of the pSTS peaks sat in the right hemisphere near the middle of that sulcus, consistent with classic voice and biological motion processing zones. None of these differences showed up in the primary auditory cortex, which tells us the effect isn’t about hearing the tones more clearly; it’s about assigning social and emotional meaning to them. There was a twist in the timing. Using impulse response functions—the little waves that tell you how the signal evolves over seconds—the team saw that experts at rest sometimes dipped below baseline in these social emotional regions when the sounds played. During compassion, the same regions swung positive, especially on the right and especially in the amygdala, pSTS, and TPJ. In the precuneus and TPJ, the meditation state seemed to set a different baseline and then sustain a response through and after the sounds. That shift in direction—from slight deactivation at rest to activation during compassion—reads like a change in mode. The system is primed to treat human vocal cues as socially meaningful when you are in that compassionate stance. Rightward lateralization kept popping up. The strongest social cognition effects in experts lived in the right hemisphere, and the team even ran formal laterality checks to confirm it. That aligns with a long tradition that puts fast, holistic social cue processing—tone of voice, biological motion, and gaze—more on the right. Training didn’t invent a new network; it seemed to tune an existing one to be more responsive when the person intended to care. Let’s pause on the logic of the analyses, because it matters for interpreting causality. This was a two times two times three design: group, state, and valence. The headline interaction in the right insula and secondary somatosensory cortex shows that expertise amplifies how the compassion state focuses the brain on emotionally rich sounds, especially negative ones. But the study also leaned on within-person variation. Using that nine-point intensity scale, they resorted blocks into “good” and “poor” meditation and looked at brain responses accordingly. The insula and cingulate tracked those self-reports across people. That’s a strong case that the state itself—not just who you are—matters. Does the autonomic system come along for the ride? Yes. Pupils dilated more to all sounds during compassion than during rest, and that dilation was larger in experts. But controlling for that didn’t erase the insula or the right lateralized social cognition effects. That separation is important because it rules out the simplest explanation—that they were just more aroused—and points to a specific reshaping of empathic and mentalizing circuits. There are caveats. The team excluded two participants for excessive motion, and field maps—used to correct distortion—weren’t available for the first three experts, which adds variance. Cultural background and language differed between some experts and novices, which could, in theory, tilt how social sounds are processed. Crucially, this is cross-sectional: experts weren’t randomly assigned at age eighteen to practice for 10,000 hours. So, you can’t say training caused the differences with ironclad certainty. There was also no concurrent behavioral task in the scanner; that was deliberate to avoid breaking meditation, but it means we don’t have trial by trial measures of accuracy or reaction time to anchor the neural changes to performance. However, the team did use pupil measures and self-reports to triangulate the state. Even with those limits, the picture that emerges is compelling. When people enter a compassion state, their brains lean into both the feeling side—insula and cingulate—and the mind reading side—TPJ, pSTS, medial prefrontal, posterior cingulate, and precuneus. In experienced meditators, the lean is stronger, more right lateralized, and flips regions that would otherwise drift downward at rest into an engaged, responsive mode. The most concrete anchor for that story is the right anterior insula: a big, statistically corrected cluster, a clear expert by state by valence effect, sensitivity to negative over positive sounds during compassion, and a dose response relationship with how strong the meditation felt. It’s hard to ask for a cleaner signature. What does that mean outside the scanner? At the very least, it shows that compassion isn’t just a warm, fuzzy word. It’s a trainable state that reaches into neural systems we rely on to share feelings and to imagine other minds, and it does so in a way you can dial up and down. That’s a scientific foothold. Turning it into everyday impact—on helping behavior, on burnout in caregiving professions, on how we handle conflict—will need longitudinal studies that track brains and behavior together, and that randomize people to real training versus controls. Lutz and colleagues point exactly there. For now, the data give us a satisfying, mechanistic take-home: when you practice caring on purpose, the brain’s empathy engine and its social radar both pick up the signal, and, with expertise, they hum a little louder and a little more to the right.

Antoine Lutz and colleagues set this up against what we already know about empathy for pain. A pair of regions—the anterior insula and anterior cingulate cortex—light up reliably when we share someone else’s distress. But empathy isn’t just raw feeling; it often recruits a broader social cognition network that helps us infer and imagine minds.

These areas include the temporoparietal junction, also known as TPJ, the posterior superior temporal sulcus, known as pSTS, the medial prefrontal cortex, and the posterior cingulate along with the neighboring precuneus. So, the team asked: if compassion meditation is a learned skill, can entering that state dial up both the feeling circuits and the perspective-taking ones?

To test it, they brought in two groups. Long-term Buddhist practitioners with more than 10,000 hours of meditation—true experts—and age and gender-matched novices with no prior experience, who received a week of brief training before scanning. In the scanner, everyone alternated between resting and actively generating compassion, keeping their eyes open and fixated.

During both states, they heard short, two-second human sounds drawn from a standardized set—some negative, some neutral, and some positive. The beauty of that design is that the same stimulus hits two different internal states. After each run, participants rated how strong their meditation had felt on a nine-point scale, which later allowed the team to compare “good” versus “poor” blocks within each person.

The imaging was standard but careful. Functional data came from a three Tesla scanner with a two-second repetition time and roughly four millimeter voxels. Two participants were excluded for motion, and the main imaging comparisons ended up with 15 experts and 15 novices.

They modeled the brain’s response to each sound over a 20-second window using a set of sine functions—basically a flexible way to capture the rise and fall of the blood-oxygen signal without assuming a single shape. They also recorded pupil diameter at 60 hertz for most participants—13 controls and 10 experts—to index arousal. That matters because a bigger pupil can mean you are more alert or emotionally engaged, and you don’t want to confuse that with empathy per se.

Pupil size was folded into the analyses as a covariate, effectively subtracting out arousal-linked variance from the brain signal. To keep false positives in check, they used Monte Carlo simulations to set cluster size thresholds for their statistical maps.

Start with the broadest contrast: compassion versus rest, averaged across everyone. When people entered compassion, the anterior insula and anterior cingulate ramped up. So did a wider network tied to mentation and social cognition—the temporal poles, the posterior superior temporal sulcus, the temporoparietal junction, the medial prefrontal cortex, posterior cingulate, and precuneus.

Think of it as both feeling and thinking about minds getting a little more engaged. There was a clear right hemisphere bias to these state effects. Importantly, just comparing experts to novices without considering state didn’t show a global group difference.

The interesting action lived in the interaction: how state and stimulus valence played out differently by expertise.

That is where the right anterior insula stole the show. In a voxelwise analysis, a sizable cluster there—about 3,667 voxels, corrected—showed a three-way interaction of group, state, and valence. What does that mean in plain terms?

Experts, when meditating, reacted more strongly to emotional sounds than to neutral ones, and that boost was bigger than what novices showed. Focus on negative sounds, and you see it cleanly: the right insula response during meditation versus rest was higher in experts than novices, with a group difference that cleared conventional significance, t with 28 degrees of freedom just over 2. The more striking part is that this wasn’t just an “experts versus beginners” story.

Within individuals, when they said a block felt like strong compassion, the insula response to emotional sounds was larger than in their own poor blocks. Across a subset, that verbal report effect in the right insula and the cingulate was statistically reliable, with an F around 6.8. So, the brain signal tracked the felt intensity of compassion.

Now, you might be wondering about arousal. Pupil diameter did increase during meditation relative to rest—that was a main effect, with an F just over 5 in a 16-person analysis. Experts, as a group, showed a bigger state-related pupil jump than novices, with a state-by-group interaction near F equals 11 with a small p-value.

Bigger pupils leading to a bigger insula response—was it all just alertness? The team checked. Even when they regressed out the pupil effects, the right insula still carried a significant meditation-linked boost to emotional sounds.

In fact, the correlation between the meditation-related pupil increase and the insula boost was moderate, around r equals 0.54, suggesting that arousal and empathic processing moved together but weren’t the same thing. That is useful. It means the compassion state changes how the brain treats social emotion beyond simply waking you up.

The story didn’t stop at the insula. Experts, specifically during compassion, showed heightened responses in a set of regions we use to read social cues and infer intentions. The right temporoparietal junction and the right posterior superior temporal sulcus stood out, along with the posterior cingulate and precuneus.

The amygdalae—the almond-shaped structures sensitive to emotional salience—also joined in, as did the right inferior frontal gyrus and a swath of premotor cortex around Brodmann area six. One of the pSTS peaks sat in the right hemisphere near the middle of that sulcus, consistent with classic voice and biological motion processing zones. None of these differences showed up in the primary auditory cortex, which tells us the effect isn’t about hearing the tones more clearly; it’s about assigning social and emotional meaning to them.

There was a twist in the timing. Using impulse response functions—the little waves that tell you how the signal evolves over seconds—the team saw that experts at rest sometimes dipped below baseline in these social emotional regions when the sounds played. During compassion, the same regions swung positive, especially on the right and especially in the amygdala, pSTS, and TPJ.

In the precuneus and TPJ, the meditation state seemed to set a different baseline and then sustain a response through and after the sounds. That shift in direction—from slight deactivation at rest to activation during compassion—reads like a change in mode. The system is primed to treat human vocal cues as socially meaningful when you are in that compassionate stance.

Rightward lateralization kept popping up. The strongest social cognition effects in experts lived in the right hemisphere, and the team even ran formal laterality checks to confirm it. That aligns with a long tradition that puts fast, holistic social cue processing—tone of voice, biological motion, and gaze—more on the right.

Training didn’t invent a new network; it seemed to tune an existing one to be more responsive when the person intended to care.

Let’s pause on the logic of the analyses, because it matters for interpreting causality. This was a two times two times three design: group, state, and valence. The headline interaction in the right insula and secondary somatosensory cortex shows that expertise amplifies how the compassion state focuses the brain on emotionally rich sounds, especially negative ones.

But the study also leaned on within-person variation. Using that nine-point intensity scale, they resorted blocks into “good” and “poor” meditation and looked at brain responses accordingly. The insula and cingulate tracked those self-reports across people. That’s a strong case that the state itself—not just who you are—matters.

Does the autonomic system come along for the ride? Yes. Pupils dilated more to all sounds during compassion than during rest, and that dilation was larger in experts.

But controlling for that didn’t erase the insula or the right lateralized social cognition effects. That separation is important because it rules out the simplest explanation—that they were just more aroused—and points to a specific reshaping of empathic and mentalizing circuits.

There are caveats. The team excluded two participants for excessive motion, and field maps—used to correct distortion—weren’t available for the first three experts, which adds variance. Cultural background and language differed between some experts and novices, which could, in theory, tilt how social sounds are processed.

Crucially, this is cross-sectional: experts weren’t randomly assigned at age eighteen to practice for 10,000 hours. So, you can’t say training caused the differences with ironclad certainty. There was also no concurrent behavioral task in the scanner; that was deliberate to avoid breaking meditation, but it means we don’t have trial by trial measures of accuracy or reaction time to anchor the neural changes to performance.

However, the team did use pupil measures and self-reports to triangulate the state.

Even with those limits, the picture that emerges is compelling. When people enter a compassion state, their brains lean into both the feeling side—insula and cingulate—and the mind reading side—TPJ, pSTS, medial prefrontal, posterior cingulate, and precuneus. In experienced meditators, the lean is stronger, more right lateralized, and flips regions that would otherwise drift downward at rest into an engaged, responsive mode.

The most concrete anchor for that story is the right anterior insula: a big, statistically corrected cluster, a clear expert by state by valence effect, sensitivity to negative over positive sounds during compassion, and a dose response relationship with how strong the meditation felt. It’s hard to ask for a cleaner signature.

What does that mean outside the scanner? At the very least, it shows that compassion isn’t just a warm, fuzzy word. It’s a trainable state that reaches into neural systems we rely on to share feelings and to imagine other minds, and it does so in a way you can dial up and down.

That’s a scientific foothold. Turning it into everyday impact—on helping behavior, on burnout in caregiving professions, on how we handle conflict—will need longitudinal studies that track brains and behavior together, and that randomize people to real training versus controls. Lutz and colleagues point exactly there.

For now, the data give us a satisfying, mechanistic take-home: when you practice caring on purpose, the brain’s empathy engine and its social radar both pick up the signal, and, with expertise, they hum a little louder and a little more to the right.

Related lectures