communications biology Article

A Nature Portfolio journal

https://doi.org/10.1038/s42003-025-09088-3

Neural and molecular changes during a

mind-body reconceptualization,
meditation, and open label placebo
healing intervention
Check for updates
1,2 1 1 1 1
Alex Jinich-Diamant , Sierra Simpson , Juan P. Zuniga-Hertz , Ramamurthy Chitteti , Jan M. Schilling ,
Jacqueline A. Bonds1, Laura Case 1, Andrei V. Chernov1,3, Joe Dispenza4, Jacqueline Maree5,
Natalia Esther Amkie Stahl1, Michael Licamele1, Narin Fazlalipour1, Swetha Devulapalli1,
Leonardo Christov-Moore6, Nicco Reggente 6, Michelle A. Poirier4, Tobias Moeller-Bertram5 &
1234567890():,;
1234567890():,;

Hemal H. Patel 1,3

Mind-body interventions offer promising avenues for improving physical and mental health, yet the
comprehensive biological effects of increasingly popular mind-body retreat interventions remain
poorly understood. The neural and molecular effects of a 7-day retreat intervention combining
meditation, reconceptualization, and open-label placebo healing rituals are investigated in an
observational study on 20 healthy human participants randomly selected from 561 retreat
participants. BOLD fMRI functional connectivity during rest and meditation and whole plasma
proteomics, metabolomics, exosome-specific miRNA transcriptomics, and neurite growth and real-
time metabolism cellular assays are compared pre- and post-intervention. Meditation decreases
functional integration in the default mode (p = 0.00009) and salience networks (p = 0.000003) and
decreases whole-brain modularity (p = 0.001). Compared to pre-intervention plasma, post plasma
increases in vitro neurite outgrowth (p = 0.01), enhances glycolytic metabolism (p = 0.008), induces
upregulation of BDNF (p = 0.001), inflammatory (p = 0.0001), anti-inflammatory (p = 0.03), and
endogenous opioid (p = 0.03) pathways, and modulates tryptophan metabolism (pFDR = 0.03) and
neurotransmission-associated exosome miRNA transcripts. This intensive non-pharmacological
mind-body intervention produces broad short-term neural and plasma-based molecular changes
associated with enhanced neuroplasticity, metabolic reprogramming, and modulation of functional
cell signaling pathways, highlighting the potential of mind-body techniques to modulate neural circuits
and pathways important to health and well-being.

Mind-body interventions–structured practices that harness the interaction reduced pain three times more than placebo and six times more than usual
between psychological processes and physiological systems–can sig- treatment1. Placebo effects, another mind-body intervention based on
nificantly improve human physical and mental health, yet the neural and healing-centered rituals, symbols, and behaviors, affect every major organ
molecular effects of increasingly popular mind-body retreat interventions system2,3 and sometimes surpass routine surgical outcomes4–6. Interestingly,
remain poorly understood. In a recent RCT, reconceptualizing pain as the open-label placebos (administered without concealment such that the
product of plastic brain activity rather than of peripheral tissue injury subject is aware of the placebo) conserve their effectiveness against many

1
Department of Anesthesiology, University of California San Diego, La Jolla, CA, USA. 2Department of Cognitive Science, University of California San Diego, La
Jolla, CA, USA. 3Veterans Affairs San Diego Healthcare System, La Jolla, CA, USA. 4Metamorphosis LLC, Ranier, WA, USA. 5VitaMed Research, Palm Desert, CA,
USA. 6Institute for Advanced Consciousness Studies, Santa Monica, CA, USA. e-mail: [email protected]

Communications Biology | (2025)8:1525 1

https://doi.org/10.1038/s42003-025-09088-3 Article

conditions7–9, demonstrating that placebo responses do not require decep- The 7-day retreat combined lectures, meditation, and healing rituals.
tion, expectations, or conditioning. Meditation, yet another intervention Daily lectures (25 total hours) emphasized the body’s self-healing abilities,
based on self-regulated attentional practices, can produce subjective the mind’s capacity to shape lived reality, and the healing power of present-
mystical-type experiences10, mental health improvements11,12, and can alter centeredness and mystical-type experiences. All meditations (33 total hours)
neural13–15, immune16, autonomic17,18 gene expression19–21, proteomic22–24, were guided, delivered with atmospheric music, and taught Kundalini
and metabolomic25,26 activity. techniques, which combine conscious meta-awareness and conscious
Each of these interventions operates through partially distinct cognitive breathing exercises with slow, ascending, focused interoceptive attention on
mechanisms, raising the possibility that they may complement one another purported energetic centers along the midline (e.g., brow, throat, heart)
to impact the brain and body synergistically. Reconceptualization operates which, according to practitioners, can reprocess embodied trauma and
through conscious, discursive, and volitional alteration of core beliefs, while catalyze adaptive mental and physical changes27,28. Guidance also empha-
meditation involves a willful yet non-discursive alteration of consciousness, sized sustaining a heart-centered state devoid of thinking or judgment and
and open-label placebo involves conscious awareness but operates uncon- focusing awareness on a void beyond one’s normal sense of space and time
sciously. While research on each technique exists, their combined neural —a common theme in some contemplative practices29. Guided healing
and molecular effect has never been studied. To do so, we conducted an rituals (5 total hours) brought 6–8 “healers” around one “healee” in which
exploratory observational study with functional magnetic resonance ima- the former were instructed to practice loving-kindness compassion medi-
ging (fMRI) and blood plasma-based high-throughput proteomics, meta- tation while focusing attention on their heart, hands, and on the latter’s
bolomics, exosome-specific miRNA transcriptomics; and neurite growth body. A healing mechanism was not presented, but the possibility that
and real-time metabolism cellular assays (Fig. 1A) on 20 healthy adult healing could occur on either party because of the ritual was mentioned,
participants (14 females, age = 46.35 ± 10.06 (SD) years) (Fig. 1B) sampled similarly to how open-label placebos are presented in trials30. All study
before and after a 7-day mind-body retreat (Fig. 1C). While logistical lim- subjects participated as healers. Of the 20 participants, 11 were “advanced”
itations prevented us from including age and gender-matched controls to meditators who had practiced the techniques taught for at least six months,
isolate and mechanistically describe the neural and physiological pathways while 9 were “novices” who had not. No pharmacological substances,
engaged by each separate mind-body technique–an equally important but including any psychedelics, were involved in the retreat.
separate task–, our study provides evidence for the breadth and depth of We characterized the resting and meditation states experienced by
physiological effects following a holistic, multifaceted experience commonly participants pre- and post-intervention. Consistent with previous medita-
described by participants as personally transformational. tion studies31,32, we observed higher meditation-associated whole-brain

A Outcome Measures

Whole-brain network

Meditation BOLD resting state Functional connectivity Resting state networks

Mind MRI
Reconceptualization Structural scan Anatomical differences A priori regions of interest
Healing Ritual
Neuron assay

ELISA
ML
Blood plasma Real time metabolism

Body Metabolomics

Proteomics

Exosomes miRNA

1-week

Pre Post

B Study Participants C Data Collection

Day -1 Day 0 Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8 Day 9
Interventions

0.4 hrs. 5.0 hrs. 4.0 hrs. 4.7 hrs. 4.0 hrs. 6.0 hrs. 4.0 hrs. Meditation

2.0 hrs. 3.5 hrs. 3.0 hrs. 4.0 hrs. 4.5 hrs. 3.5 hrs. 2.5 hrs. Reconceptualization

0.8 hrs. 0.8 hrs. 1.0 hrs. Healing ritual

time point collection

n=1 n = 11 n=9 n=6 n = 10 n=5

Analysis Sample

Blood draw

n = 10 n = 10 n = 10 n = 10 fMRI + MEQ

Pre Post

D CONSORT Flow Diagram

Enrollment

Invited to participate Assessed for eligibility Eligible Consented & enrolled Sampled
(n = 1,444) (n = 561) (n = 65) (n = 27) (n = 20)

BDNF index upper median BDNF index lower median Completed Excluded (n = 496) Randomly selected, Dropped (n = 7)
SCFA index upper median SCFA index lower median screening form Not good health (n = 399) given written consent Scheduling conflict (n = 5)
HDAC index upper median HDAC index lower median Not MRI safe (n = 19) (n = 36) Masking non-compliance (n = 1)
Schedule conflict (n = 78) Failed MRI pre-screening (n = 1)

Fig. 1 | Study design, participants, data collection, and recruitment. A Outcome bottom median scores for BDNF, short-chain fatty acid (SCFA) metabolism, and
measures to capture biological changes associated with brain and body. Created with HDAC proteomic pathways. C Intervention and data collection timeline. (fMRI
BioRender. Simpson, S., Jinich, A. (2025). BioRender.com/ryzs1cd. B Participant functional magnetic resonance imaging, MEQ mystical experience questionnaire).
characteristics, including age, subject, gender, meditation experience level, and top/ D CONSORT flow diagram.

Communications Biology | (2025)8:1525 2

https://doi.org/10.1038/s42003-025-09088-3 Article

A Meditation vs. Rest Contrast B Meditation vs. Rest Contrast

Resting State Networks Resting State Network Regions
Pre Post Pre Post
ECN DMN DAN SN SMN VN AN ECN DMN DAN SN SMN VN AN

* * dorsal medial PFC dorsal medial PFC

left anterior PFC

ECN
left anterior PFC

ECN
right anterior PFC right anterior PFC
left superior parietal left superior parietal
* right superior parietal
PCC
right superior parietal
PCC
mPFC mPFC
left lateral parietal left lateral parietal
* *

DMN
right lateral parietal right lateral parietal

DMN
left inferior temporal left inferior temporal
right inferior temporal right inferior temporal
medial dorsal thalamus medial dorsal thalamus
right posterior cerebellum right posterior cerebellum
left posterior cerebellum left posterior cerebellum
left frontal eye field left frontal eye field
right frontal eye field right frontal eye field
left posterior IPS left posterior IPS

DAN
DAN
right posterior IPS right posterior IPS
left anterior IPS left anterior IPS
right anterior IPS right anterior IPS
* left MT
right MT
left MT
right MT
dACC dACC
left anterior PFC left anterior PFC
dorsal medial PFC dorsal medial PFC

SN
SN
left insula left insula
right insula right insula
left lateral parietal left lateral parietal
Cohen’s d right lateral parietal right lateral parietal

AN VN SMN
left motor cortex

AN VN SMN
left motor cortex
right motor cortex right motor cortex
SMA SMA
-2 -1 0 1 2 left V1 left V1
right V1 right V1
left A1 left A1
right A1 right A1

dorsal medial PFC

left anterior PFC
right anterior PFC
left superior parietal
right superior parietal
PCC
mPFC
left lateral parietal
right lateral parietal
left inferior temporal
right inferior temporal
medial dorsal thalamus
right posterior cerebellum
left posterior cerebellum
left frontal eye field
right frontal eye field
left posterior IPS
right posterior IPS
left anterior IPS
right anterior IPS
left MT
right MT
dACC
left anterior PFC
dorsal medial PFC
left insula
right insula
left lateral parietal
right lateral parietal
left motor cortex
right motor cortex
SMA
left V1
right V1
left A1
right A1

dorsal medial PFC

left anterior PFC
right anterior PFC
left superior parietal
right superior parietal
PCC
mPFC
left lateral parietal
right lateral parietal
left inferior temporal
right inferior temporal
medial dorsal thalamus
right posterior cerebellum
left posterior cerebellum
left frontal eye field
right frontal eye field
left posterior IPS
right posterior IPS
left anterior IPS
right anterior IPS
left MT
right MT
dACC
left anterior PFC
dorsal medial PFC
left insula
right insula
left lateral parietal
right lateral parietal
left motor cortex
right motor cortex
SMA
left V1
right V1
left A1
right A1
Meditation vs. Rest Contrast
A Priori Regions of Interest
Cohen’s d
Pre Post
mPFC -2 -1 0 1 2
left dlPFC left dlPFC

left insula left insula

central precuneus
D Whole Brain Functional Connectivity Network

left angular gyrus left angular gyrus Modularity Characteristic Path Length Global Efficiency

** * **** ***
Rest

Harvard-Oxford Atlas
Meditation
Cohen’s d

-1 -.5 0 .5 1

E Novice vs. Advanced *** ** ****

****
Structural Differences
Power Atlas

L A R ***
L R

Pre Post Pre Post Pre Post

P
Modularity: The extent to which a network can be divided into distinct, interconnected communities.
Characteristic path length: The average shortest path length between all pairs of nodes in a network.
Global efficiency: A measure of how well information can travel across the entire network.

Fig. 2 | Functional magnetic resonance imaging (fMRI) (n = 19 participants). comparisons). C Significant (p < 0.05) meditation-induced functional connectivity
A Meditation vs. rest RSN functional connectivity pre- and post-intervention. Cell changes between a priori ROIs pre- and post-intervention. Color scale represents
values and color scale represent Cohen’s d effect size values from paired t-tests. Cohen’s d effect sizes from paired t-tests. D Whole-cortex/brain network measures
Asterisks denote pFWE < 0.05 significance (28 multiple comparisons). ECN (execu- for 48-region Harvard-Oxford cortical atlas and Power (2011) 264-region brain atlas
tive control network), DMN (default mode network), DAN (dorsal attention net- (n = 19). Asterisks denote * (p < 0.05), ** (p < 0.01), *** (p < 0.001), ****
work), SN (salience network), SMN (somatomotor network), VN (visual network), (p < 0.0001). E Anatomical Structure. Coronal and axial views with superior parietal
and AN (auditory network). B Meditation vs. rest RSN-region functional con- lobule cluster (yellow), where advanced practitioners had greater gray matter
nectivity pre- and post-intervention. Cell values and color scale represent Cohen’s d volume than novices at baseline.
effect size values from paired t-tests. Asterisks denote pFWE < 0.05 (630 multiple

functional integration and reduced intra-network connectivity in the default framewise displacement on a two-way (task × time) repeated measures
mode and salience networks. Comparing pre- to post-intervention whole ANOVA (n = 19, F(1,18) = 25.1, p = 0.00009, η²p = 0.58) (Supplementary
plasma, we report evidence of enhanced neuroplasticity and glycolytic Tables 1 and 2).
metabolism, as well as activation of endogenous opioid and neuromodu-
latory pathways. Meditation versus rest
We examined functional connectivity in seven resting state networks
Results (RSNs) (Fig. 2A, B), eight regions of interest (ROIs) (Fig. 2C) (coordinates in
Functional brain activity Supplementary Tables 3 and 4), and two whole-brain networks (Fig. 2D),
To characterize the neural signature of the meditative state, participants comparing meditation with rest (n = 19, pFWE = 0.0018 for 28 network
underwent structural and blood-oxygenation-level-dependent (BOLD) pairs). All resting state networks had higher intra- than inter-network
functional MRI scans during rest (5 min) and meditation (15 min). Mystical connectivity (Supplementary Fig. 1A). Compared to rest, meditation
experience questionnaire (MEQ) scores reflective of the scanner meditation reduced intra-network connectivity in the salience network pre- and post-
increased significantly following the intervention (n = 20; pre = 2.37 ± 1.26 intervention (pre: t = −4.32, p = 0.0004, Cohen’s d = −0.87; post: t = −6.71,
(SD); post = 3.02 ± 1.44) (Wilcoxon signed rank test W = 35.0, p = 0.03), p = 0.000003, Cohen’s d = −1.76), default mode network post-intervention
indicating a deepening of the meditative state between sessions. One par- (DMN) (t = −5.04, p = 0.00009, Cohen’s d = −1.78), and dorsal attention
ticipant’s data was excluded from further analysis due to corrupt T1w data. network (DAN) post-intervention (t = −3.71, p = 0.002, Cohen’s
Participants moved more during meditation than rest, a potential confound d = −1.06) (Fig. 2A). Meditation also showed reduced inter-network con-
revealed by the significant effect of task (meditation, rest) on mean nectivity post-intervention between the salience and dorsal attention

Communications Biology | (2025)8:1525 3

https://doi.org/10.1038/s42003-025-09088-3 Article

Table 1 | Statistically significant (pFWE < 0.05) rest-vs-meditation differences on resting state network regions (n = 19
participants)

Session RSN component region pair RSN(s) t p Cohen’s d

Pre R post. Cerebellum Medial PFC DMN −4.44 0.0003 −1.15
Post R post. Cerebellum Medial PFC DMN −5.81 0.00002 −1.30
Post L post. Cerebellum Medial PFC DMN −6.30 0.000006 −1.87
Post L post. Cerebellum L inferior temporal DMN −4.56 0.0002 −1.13
Post R anterior PFC L anterior PFC SN −7.41 0.0000007 −1.38
Post L insula L posterior IPS SN −4.32 0.0004 −1.01
Post L insula L anterior PFC SN −5.30 0.00005 −1.57
Post R insula L insula SN −5.58 0.00003 −1.44
Post L lateral parietal L anterior PFC SN −4.62 0.0004 −0.88
Post L lateral parietal L insula SN −7.30 0.0000009 −1.30
Post L lateral parietal R insula SN −6.01 0.00001 −1.02
Post R lateral parietal L lateral parietal SN −4.41 0.0003 −1.18
Post L insula PCC SN-DMN 4.50 0.0003 1.47
Post R insula L frontal eye field SN-DAN −4.24 0.0005 −0.89
Post L lateral parietal L posterior IPS SN-DAN −4.66 0.0002 −1.15
Post L A1 R insula AN-SN −4.68 0.0002 −1.47
Post L A1 L motor cortex AN-SMN −4.67 0.0002 −1.19
Post L A1 SMA AN-SMN −4.55 0.0002 −1.22

Table 2 | Significant (pFDR < 0.05) pre-to-post differences on RSN component regions (n = 19 participants)

Task RSN Component Region Pair RSN(s) t p Cohen’s d

meditation L posterior IPS L A1 DAN-AN −4.72 0.0002 −1.01
meditation L posterior IPS L lateral parietal DAN-SN −4.44 0.0003 −0.80

networks (t = −4.09, p = 0.0007, Cohen’s d = −1.18) and between the d = −1.07; post: W = 45.0, p = 0.04, Cohen’s d = −0.77) and increased global
somatomotor and auditory networks (t = −5.30, p = 0.00005, efficiency (pre: W = 3.0, p = 0.00002, Cohen’s d = 2.04; post: W = 10.0,
Cohen’s d = −1.38). p = 0.0002, Cohen’s d = 1.26) as compared to rest (Fig. 2D). These results
For individual network-component regions (Fig. 2B and Table 1), both were robust to excluding BOLD runs with mean framewise displacement
pre- and post-intervention, meditation reduced connectivity between the > 0.3 mm, indicating they were not due to higher meditation-associated
right posterior cerebellum and medial prefrontal cortex (pre: t = −4.44, head motion.
p = 0.0003, Cohen’s d = −1.15; post: t = −5.81, p = 0.00002, Cohen’s Meditation thus reduced functional connectivity in the default mode
d = −1.30) and bilateral connectivity between insular cortices, lateral par- and salience networks and induced a whole-brain functional state less
ietal cortices, and anterior prefrontal cortices (all p ≤ 0.0003; Table 1), net- segregated into distinct modules, that allows for more efficient informa-
work hubs associated with sensory, affective, and cognitive functions that tion flow.
shape the prediction-driven conscious state33. Interestingly, the only
meditation-driven connectivity increase was post-intervention between the Pre versus post
left insula and posterior cingulate cortex (t = 4.50, p = 0.0003, Cohen’s Pre- to post-intervention, functional connectivity during meditation
d = 1.47), a finding previously reported during absorptive trance states34. decreased between executive control and salience networks (t = −2.61,
ROIs also revealed lower functional connectivity between key network hubs, p = 0.02, Cohen’s d = −0.62) but did not survive correction for multiple
including medial prefrontal cortex (mPFC), precuneus, bilateral insular comparisons (pFWE = 0.0018) (Supplementary Fig. 1B). For network com-
cortices, and bilateral angular gyri (Supplementary Table 5), serving as an ponent regions, connectivity during meditation decreased significantly pre-
internal replication and returning results agreeable with the RSN data- to-post between the left posterior intra-parietal sulcus and auditory cortex
driven approach. and between the left posterior intra-parietal sulcus and left lateral parietal
To observe meditation-induced changes at the whole-brain/cortex cortex (Table 2). No significant changes were found between a priori ROIs
level, we parcellated the brain into 48 cortical regions using the Harvard- (Supplementary Table 8).
Oxford atlas35–37 and into 264 brain regions using the Power atlas38 and
calculated whole-brain/cortex network measures (Supplementary Table 6). Neuroanatomical differences
For the cortical parcellation, a 2 (pre/post) by 2 (rest/meditation) repeated No significant anatomical changes were observed pre-to-post intervention,
measures ANOVA (Supplementary Table 7) (n = 19) revealed a significant but advanced practitioners showed greater gray matter volume in the right
effect of meditation on network modularity (F(1,18) = 15, p = 0.001, η²p = superior parietal lobule at baseline (n = 19, p = 0.049) (Fig. 2E), a region
0.45) and global efficiency (F(1,18) = 48, p = 0.000002, η²p = 0.73), and no linked to spatial awareness and body representation, which has been
significant effect on characteristic path length, with similarly significant reported to have greater cortical thickness in experienced Zen meditators39.
whole-brain results. Post-hoc Wilcoxon signed rank tests confirmed that This difference warrants further investigation with larger samples and
meditation decreased modularity (pre: W = 18.0, p = 0.001, Cohen’s comparisons with age-matched non-meditator controls.

Communications Biology | (2025)8:1525 4

https://doi.org/10.1038/s42003-025-09088-3 Article

A Neurite Outgrowth Assay B Day 10 PC12Neurites

Pre Post
2500 Pre Post
Neurite length Δ%

****
****
2000
1500 **
****
****
1000
****
500 *

2 3 4 5 6 7 8 9 10 50um 50um
Day

C D Proteomic BDNF Pathway Heat Map

Proteomic BDNF PathwayIndex

Advanced Novice

Advanced
Novice

RAF1
JUN
NRAS
SLITRK1
BDNF
SLITRK1
AKT1
BRAF
SHC1
HRAS
MECP2
KRAS
CREB1
GSK3B
BDNF
MAPK3
AKT1
RAC1
MAPK1
GSK3B
CAMK2A
CAMK2B
NTRK1
NTF4
NTRK2
KRAS
SHC1
GRB2
DLG4
NGFR
PIK3R1
SOS1
PTPN11

PTPN11
All

-0.2 0.0 0.2 0.4 0 0.2 0.4

Fold change Fold change

Fig. 3 | Neuronal growth (n = 20 participants). A PC12 neurite growth: Percent index pre-to-post fold change. Error bars denote SEM. Fold change levels sig-
change in mean neurite length from baseline (day 2 post-NGF treatment). Asterisks nificantly above zero signal upregulated pathway or protein cluster. D BDNF
denote * (p < 0.05), ** (p < 0.01), *** (p < 0.001), **** (p < 0.0001). B Phase con- Pathway Heat Map. Fold change per protein for advanced and novice participants.
trast 20× microscopy images of PC12 cells on day 10 post-NGF treated with pre- and
post-intervention plasma with longest neurite per cell body traced. C BDNF pathway

Whole plasma changes in plasma produced during the intervention induced a compen-
Having characterized the neural state induced by meditation, we turned our satory shift from mitochondrial to glycolytic ATP production, with post-
attention to the broader metabolic and molecular changes in whole plasma. plasma-treated cells showing significantly higher basal glycolytic rate than
pre-plasma-treated cells (t = 2.95, p = 0.008, Cohen’s d = 0.36) despite no
Enhanced neuroplasticity significant differences in mitochondrial (t = −1.04, p = 0.31, Cohen’s
Participants’ anecdotal reports consistently emphasize radical psychological d = −0.28) or total ATP production rate (t = 1.46, p = 0.16, Cohen’s
breakthroughs, and previous meditation studies have reported increased d = 0.25). A glycolytic rate assay (Fig. 4B) confirmed that, compared to pre-
BDNF (brain-derived neurotrophic factor) levels22 consistent with plasma, post-plasma increased basal glycolytic rate (t = 3.39, p = 0.003,
enhanced neuroplasticity. To investigate whether the intervention affected Cohen’s d = 0.63) and lowered basal respiration rate (t = −4.80, p = 0.0001,
circulating plasma factors conducive to neuroplasticity, we treated cultured Cohen’s d = −1.26). Mitochondrial stress parameters (Fig. 4C) did not
glutamatergic PC12 neuroendocrine cells with NGF (nerve growth factor) significantly differ between cells exposed to pre- and post-plasma, while,
and 1% pre- and post-intervention plasma and quantified neurite out- compared to cells exposed to growth medium only, plasma-exposed cells
growth length. Starting on day 4 post-NGF treatment, post-plasma-treated displayed significantly higher ATP production and glycolytic rates.
cells exhibited significantly longer neurites than pre-plasma-treated cells As before, to see if the plasma proteome reflected these changes, we
(t = −2.52, p = 0.01, Cohen’s d = 0.59) and continued to do so until the end pre-selected 19 proteins involved in glycolysis and oxidative phosphoryla-
of the experiment (Fig. 3A, B). tion and calculated pre-to-post foldchanges and an index average
To investigate proteomic factors driving this effect, we quantitatively (Fig. 4D–G). The glycolysis index increased significantly pre-to-post
measured 7596 protein targets using the high-throughput SomaScan assay (t = 3.37, p = 0.0008, Cohen’s d = 0.23), with 12 upregulated targets led by
and constructed a BDNF pathway 26-protein pre-to-post foldchange index ENO2 (enolase 2, a neuron-specific converter of 2-phosphoglycerate into
(Fig. 3C, D), which was significantly upregulated (1-sample t-test: t = 3.21, phosphoenolpyruvate) (W = 48.0, p = 0.03, CLES = 0.33) and LDHA (lac-
p = 0.001, Cohen’s d = 0.12). While BDNF itself was not significantly tate dehydrogenase A, converter of pyruvate into lactate in anaerobic gly-
affected, SLITRK1 (SLIT and NTRK-like family member 1), a protein that colysis (W = 38.0, p = 0.01, CLES = 0.22). Oxidative phosphorylation
promotes excitatory synapse development and glutamatergic neurite associated proteins trended upwards (t = 1.38, p = 0.17, Cohen’s d = 0.13)
outgrowth40,41, increased significantly pre-to-post (W = 39.0, p = 0.01, but did not reach statistical significance.
CLES = 0.35). NGFR (nerve growth factor receptor), a TNF receptor that
binds to NGF and BDNF and plays an essential role in neural cell differ- Functional cellular signaling
entiation and survival, also increased (W = 58.0, p = 0.08, CLES = 0.35). Having investigated specific pathways of interest, we performed an
exploratory, hypothesis-free analysis of proteomic and metabolomic results.
Metabolic reprogramming
Previous studies have characterized meditation as a hypometabolic state42 Proteomics
and reported enhanced glycolysis in Tibetan Buddhist monks24. To test the Volcano plot analysis (Fig. 5A) revealed 21 significantly altered proteins.
intervention’s effect on real-time metabolism, we treated BE(2)M17 human Cofilin-2 (COF2) and Enoyl-CoA hydratase were significantly upregulated,
neuroblastoma cells with 1% plasma from all participants for 60 min and which suggests enhanced cellular processes related to cytoskeletal regulation
performed Seahorse XF assays. An ATP Rate assay (Fig. 4A) revealed that and fatty acid metabolism. IL1-F6 (interleukin-36 alpha), MYPC1 (myosin

Communications Biology | (2025)8:1525 5

https://doi.org/10.1038/s42003-025-09088-3 Article

A Seahorse B Seahorse C Seahorse

ATP Production Rate Assay Glycolytic Rate Assay Mitochondrial Stress Assay

****
n.s. Total ATP
**** Pre
****
Pre
Glyco ATP **** ** Post
**** ** Post

O 2 consumption rate
Proton efflux rate
n.s.

(pmol/min)
Mito ATP

(pmol/min)
n.s. n.s. compensatory ATP-
glycolysis linked respiration maximal
basal respiration
basal respiration
glycolysis **
Oligo proton FCCP Rot/AA
leak
****
pmol/min/1000 cells

Rot/AA 2DG ** non-mitochondrial oxygen consumption

42% 44%
35%

Measurement Measurement

O2 consumption rate
Pre Oligomycin: inhibits ATP synthase, reveals ATP synthesis OCR
65% 58% 56% *** **** FCCP: reveals maximal respiratory capacity (OCR)
***
(pmol/min)
Post Rotenone/Antimycin A: inhibits ETC, reveals non-mitochondrial OCR
2-deoxy-glucose (2DG): inhibits glycolysis, reveals non-glycolytic acidification

Control Pre Post basal

respiration Rot/AA
Glycolytic ATP
Total ATP
Mitochondrial ATP
Measurement

D Glycolysis Proteomic E Glycolysis Proteomic Heat Map F OxPhosProteomic G OxPhosProteomic Heat Map
Index Index

Novice Novice
Advanced Advanced
Advanced
Advanced
Novice Novice

COX5A

NDUFA2

ATP5PF

ATP5PO
ATP5F1B

COX7A2L
ENO2
ENO2
ENO1
HK2
HK1
LDHA
ENO3
PFKM
ALDOB
HK3
GAPDH
LDHB
PGAM1
PKM
GPI
TPI1
PGK1
ALDOA
LDHA
LDHC
ALDOC

All All

-0.2 0.0 0.2 0.4 0.6 -0.2 0.0 0.2 0.4 0.6 0 0.1 0.2 0.3
0 0.5
Fold change Fold change Fold change Fold change

Fig. 4 | Metabolic effects (n = 20 participants). A–C Seahorse XF analyzer real time denote * (p < 0.05), ** (p < 0.01), *** (p < 0.001), **** (p < 0.0001). D Glycolysis
cellular metabolic assays with BE(2)M17 human neuroblastoma cells treated with pathway index pre-to-post fold change. Error bars denote SEM. Index fold change
1% pre- and post-retreat plasma for 60-min or assay buffer (control). Mean ± 95% CI levels significantly above zero signal an upregulated pathway or protein cluster.
error bars. A ATP production rate assay with total ATP production rate (mean ± E Glycolysis pathway heat map. Fold change level per component protein for
95% CI) broken into glycolytic and mitochondrial ATP production rates. advanced and novice participants. F, G Oxidative phosphorylation pathway index
B Glycolytic rate assay, including basal and compensatory glycolytic rates (top) and with pre-to-post fold change (F) and heat map for advanced and novice participants
basal mitochondrial respiration rate (bottom). Mean ± 95% CI error bars. (G). Error bars denote SEM.
C Mitochondrial respiration stress assay. Mean ± 95% CI error bars. Asterisks

binding protein C1), LDHA, and FGF-19 (fibroblast growth factor 19) Inflammation, anti-inflammation, and cellular turnover
exhibited moderate upregulation. To assess whether the intervention elicited inflammatory or anti-
Protein-protein interaction networks (Fig. 5B) revealed three sig- inflammatory cascades, we examined a panel of 23 inflammatory and 21
nificantly altered protein clusters related to mitochondrial energy produc- anti-inflammatory proteins (Fig. 5E). We found significant upregulation of
tion (ATP5F1, ATP6V1F); fatty acid metabolism (ECHS1, POP7, and inflammatory markers (t = 3.81, p = 0.0001, Cohen’s d = 0.15), driven by
ACAT2); nucleosome organization (HDAC1, RANGAP1, RBL2, C3, increases in S100A8 (calgranulin A) (W = 30.0, p = 0.004, CLES = 0.33) and
and MYOM2). CCL2 (C-C motif chemokine 2) (W = 47.0, p = 0.03, CLES = 0.33) and
Pathway enrichment analysis (Fig. 5C) showed upregulated proteins trending increases in IL-6, S100A9 (S100 calcium-binding protein A9;
associated with muscle cell apoptosis, amyloid precursor protein catabolism, calgranulin B), and PTGS2 (prostaglandin G/COX-2). These findings align
and mitochondrial proton transport pathways (padjusted < 0.05), suggesting with the role of S100A8 and S100A9 as alarmins—endogenous molecules
alterations in metabolic pathways and muscle-related cellular processes. released in response to cellular damage or stress known to induce secretion
Interestingly, pathways related to butanoate, propanoate metabolism, and of inflammatory mediators IL-6, IL-8, and CCL243,44.
tryptophan biosynthesis were enriched, suggesting shifts in key metabolic Interestingly, we also observed a significantly upregulated anti-
routes. (See Supplementary Figs. 3 and 4 for g:Profiler enrichment and inflammatory markers index (t = 2.25, p = 0.03, Cohen’s d = 0.09), with
pathway heatmaps.) positively trending levels of TGF-b1 (transforming growth factor beta-1),
We also created proteomic indices to examine inflammation, NFKBIA (NF-kappa-B inhibitor), STAT6 (signal transducer and acti-
epigenetic regulation, and short-chain fatty acid (SCFA) metabolism vator of transcription 6), CEBPB (CCAAT/enhancer-binding protein
pathways, comparing across timepoint (pre/post) and experience beta), IL1, SOCS3 (suppressor of cytokine signaling 3), and TNFAIP3
level (novice/advanced) (Fig. 5E) (Supplementary Fig. 5). The epi- (TNF alpha-induced protein 3). Concurrent activation of both pathways
genetic (HDAC) index demonstrated elevated sirtuins and HDAC suggests a dynamic process of immune modulation, possibly reflecting
expression in the advanced group, reflecting increased epigenetic enhanced cellular turnover or repair mechanisms. We also measured
regulation and possible stress resilience. In contrast, the novice group plasma nanoparticles and found no significant change in total nano-
showed differences in mitochondrial function and chromatin remo- particle concentration (t = −0.17, p = 0.87, Cohen’s d = −0.03), but did
deling. SCFA index showed increased expression of acyl-CoA dehy- find a significant decrease in the percentage of particles in the exosome
drogenase short-chain (ACADS) and fatty acid synthase (FASN) in range (20–120 nm diameter) (t = −2.09, p = 0.04, Cohen’s d = −0.65)
the advanced group, suggesting a shift toward enhanced fatty acid (Supplementary Fig. 2A, B) consistent with both higher cellular turnover
oxidation and improved metabolic efficiency. and metabolic suppression.

Communications Biology | (2025)8:1525 6

https://doi.org/10.1038/s42003-025-09088-3 Article

A Hypothesis-Free Plasma Proteome B Protein-Protein Interaction Networks

Fold change > 0.5 cluster

i ii iii iv

p < 0.05 clusters

Fold change
interactor
-3.2 -1.6 0 1.6 3.2

1580
4015

Hierarchical Clustering
4013
4025
4016
213
4023
2119
4027
4018
4026
4019
1960
4017
214
4014
4020
4021
4099
4022

SERPINB1
MNDA
STMN4
SNRPB2
SMC3
APEX1
HNRNPA0
SPI1
RNASE3
UBE2G1
PADI4
NCF1
IFI16
CELF2
VIM
S100A8|S100A9
MMACHC
H2BU1
H2BC21
H2AW
H2BC12
H2AC1
H2AC11
SFTPD
COQ9
H1-10
FGR
H1-2
MMP8
BPI
CAP1
PIP
LEG1
IGFBP1
NIF3L1
GRIA4
CD2
MYBPC1
ACTN2
SAA2
SAA1
C1QC
CRP
LDHA
IL36A
CFL2
ATP6V1F
ADSL
NTS
GCG
CCT8
CCT7
FGF19
GSTA2
ECHS1
HMGCS2
GLYAT
FTCD
CYP2C19
MAT1A
AKR1C4
ADH1C
ADH1A
ALDH2
AKR7A3
ADH4
Fold change
C 1580
1960
2119

K-Means Clustering
213
Gene Ontology: 214
4013
4015
4016
4017
4018
4023
4025
4026
# of genes 4027
4019
4014
4020
4021
4022
4099

MAT1A
HMGCS2
GLYAT
FTCD
ECHS1
CYP2C19
ALDH2
AKR7A3
AKR1C4
ADH1C
SAA2
SAA1
PIP
NTS
NIF3L1
MYBPC1
LEG1
LDHA
IL36A
IGFBP1
GSTA2
GRIA4
GCG
FGF19
CRP
CFL2
CD2
CCT8
CCT7
C1QC
ATP6V1F
ADSL
ADH4
ADH1A
ACTN2
VIM
UBE2G1
STMN4
SPI1
SNRPB2
SMC3
SFTPD
SERPINB1
S100A8|S100A9
RNASE3
PADI4
NCF1
MNDA
MMP8
MMACHC
IFI16
HNRNPA0
H2BU1
H2BC21
H2BC12
H2AW
H2AC11
H2AC1
H1-2
H1-10
FGR
COQ9
CELF2
CAP1
BPI
APEX1
is

ap sis

sis

rp n

ule ein

en

AT le

se

es

ity

oa etab r
sm

sis

ab n
sm

re ion

se

sis

rt
rte

to

te
tio

as

io

tio

ns

po
ho ptos

ic

bio olis
tiv
ha

ell

on
m

ep

r
ce osta

to

he

ta
t

lig ulat

oli

oli

t
s

po

po
la

ro

an ddic

ula

ns
lu

Ve

ac
op

a n gan

s
nt

sp

sp
c

ab

nt
gu

eo
po

ns

ns
ac

tra
g

up ran
sy
e

sy

re
e

et

et
re

C - l re
m

so

tra

tra

m
e
ll a

or

as

F
2

d
yp

TP

m
an

2+

ho
ll
ylo tosis

O
ur

an
yp otin
em ed

tia

de

k
ce

e
t

B u ate

te
gr

ta
A

Ca
V-
ec

an
s,

ne
pid

id
n
d

ke
N
cle

le

po let
an ing

Nic
or

te

ph
un
y
pr

lip

β-endorphin Dynorphin
am o p

br

br

o
g

bo

lic
sc

or
oli

ta
po

an

oid tan
rtin

on

te
us

to
tra -bo

em

&
t
p

id

in
or
et

ck

so

up
u
ph

la
a

op
rd

rt
m

e
po
cr

ba

rs
m

o
e
ell

an
os

n
Tr
Pr

yt
Se

lfu
th

of

rte
ns

ns

ns
r m ran

300
Iro
c

ns
Ph

br

tc
oo

su
br
n

tra

tra
cle

po

10000
em
t

tra
tio

le
of

m
Sm

en
+

on

ns
te
us

H+

ea
lH
la

e
n

rp

Ps
tio

pla

tra
tio
S m gu

se

Ur
tra dria

Te
tic
ca

Ele FB
la
re

8000
th

la

ta

C
d
ap
gu

te
oo

llu

IG
ic

uc

AB
n
g.

200
yn
o

va
re

ce
Ne

ed
ch

ga

pg/ml
es

pg/ml
IG
g.

or
ito

or

6000
Pr
Ne

xid
Ex
M

In
O

4000 100
2000
0 0
Pre Post Pre Post

E Short Chain Fatty EndogenousOpioids Leptin Pathway Anti-inflammatory Pro-inflammatory

HDAC Index Index Index Index
Acids Index Index

Advanced

Novice

All

-0.2 -0.1 0.0 0.1 0.2 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 -0.2 0.0 0.2 0.4 -0.1 0.0 0.1 0.2 0.3 -0.2 0.0 0.2 0.4 0.6
Fold change Fold change Fold change Fold change Fold change Fold change

Fig. 5 | Plasma proteome (n = 20 participants). A Plasma Proteome. Volcano plot Enrichment analysis (Enrichr) with pre/post significantly altered proteins (p < 0.01)
of pre/post fold changes (x-axis) and paired t-test –log(p-values) (y-axis) for 7596 and assay targets as background. D Proteome Cluster Analysis. Hierarchical and
proteins. Vertical yellow dashed lines mark a linear fold change of 0.25 in either K-means clustering of |fold changes| > 0.5 proteins. Dendrogram branch lengths
direction; horizontal yellow dashed line marks p = 0.05 cutoff. Proteins in top right/ indicate degree of similarity between expression profiles. Nodes represent protein
left quadrants are significantly up/downregulated. B Protein-Protein Interactions. targets; colors indicate fold change. E Proteomic A Priori Gene Indices. Mean pre/
Heat-diffused network clusters with ≤ 20 interactors for proteins with pre/post (i–iii) post fold change for protein clusters or pathways. Index proteins in Supplementary
p < 0.05 and (iv) |foldchange| > 0.5. Top gene enrichment terms for each cluster are: Table 15. Mean ± SEM error bars. F ELISA Assay Biomarkers. Mean ± 95% CI error
(i) fatty acid metabolism, (ii) proton-transporting two-sector ATPase complex, (iii) bars. Asterisks denote * (p < 0.05), ** (p < 0.01), *** (p < 0.001).
complement system, and (iv) nucleosome organization. C Gene Enrichment.

Endogenous opioids (PLS-DA) (Fig. 6A) revealed distinct metabolic profiles between time
Placebo effects are known to engage endogenous opioid and endocrine points, with key metabolites involved in synaptic plasticity, metabolism,
systems. We assessed levels of 15 proteomic targets within the endogenous RNA modulation, neurotransmitter availability, and inflammation and
opioid pathway and found the pathway index to be significantly upregulated over 25 metabolites with VIP (variable importance in projection) scores
(t = 2.14, p = 0.03, Cohen’s d = 0.12), driven by positively trending increases > 2 (Fig. 6B, C).
in opioid peptide precursor PENK (proenkephalin-A), opioid peptide MetaboAnalyst pathway enrichment detected 53 impacted pathways
PDYN (dynorphin A), GABR2:CD (GABBA B receptor subunit 2: cyto- (Fig. 6D and Supplementary Table 16). Six showed significant pre-to-post
plasmic domain), and CAMK2B (calcium/calmodulin-dependent protein changes (p < 0.05), led by tryptophan metabolism (pFDR = 0.03) (Fig. 6E and
kinase type II), which is known to increase after opioid administration Supplementary Fig. 6), with decreases in upstream and downstream
(Fig. 5E). ELISA assays to confirm these findings (Fig. 5F and Supplemen- metabolites, including L-tryptophan (p = 0.03), tryptamine (p = 0.04),
tary Table 11) revealed significant increases of beta-endorphin (W = 14.0, L-kynurenine (p = 0.04), indole-3-acetate (p = 0.001), and
p = 0.0002, Cohen’s d = 0.42) and dynorphin (W = 37.0, p = 0.009, 5-methoxyindoleacetate (p = 0.001). The steroid hormone biosynthesis
Cohen’s d = 0.27). pathway showed lower androstenediol (p = 0.04), testosterone (p = 0.19),
cortisol (p = 0.06), cortisone (p = 0.07), 21-deoxycortisol (p = 0.77), corti-
Metabolomics costerone (p = 0.07), and 11-dehydrocorticosterone (p = 0.02). Lower cor-
We performed liquid chromatography-mass spectrometry-based meta- tisol levels reported in previous meditation studies were interpreted as signs
bolomic analysis on plasma. Partial least squares discriminant analysis of improved stress response regulation45.

Communications Biology | (2025)8:1525 7

https://doi.org/10.1038/s42003-025-09088-3 Article

A Scores Plot B VIPScores C Top 25 Compounds

0 1
Gly-Gly-Phe Gly-Gly-Phe
Pre N-(3-Hydroxypropyl)phthalimide N-(3-Hydroxypropyl)phthalimide

15
Methoxyindoleacetic Acid Methoxyindoleacetic Acid
Post Androstenediol Androstenediol
3-Indolebutyric Acid S1P(d18:1)
10

S1P(d18:1) N6-Methyladenosine
1-Methyladenosine 1-Methyladenosine
Hi
Component 2 (7.1%)

N6-Methyladenosine Purine
Purine Benzaldehyde
5

Benzaldehyde Phenethylamine
Phenethylamine Pyrocatechol
Gly-Lys Indole-5-carboxylic acid
0

Pyrocatechol (R)-2-Hydroxy-3-phenylpropionic acid

Indole-5-carboxylic acid Leu-Tyr-Thr-Lys
Gln-Gly L-Phenylalanine
-5

Gly-Gln 5-Carboxyvanillic Acid

(R)-2-Hydroxy-3-phenylpropionic acid 4-Chloroaniline
Val-Val Gly-Glu
-10

Leu-Tyr-Thr-Lys Lo
Glu-Gly
N-acetyl-D-phenylalanine N-acetyl-D-phenylalanine
L-Phenylalanine Val-Val
-15

Glu-Gly Gly-Gln
Gly-Glu Gln-Gly
5-Carboxyvanillic Acid Gly-Lys
-15 -10 -5 0 5 10 15
4-Chloroaniline 3-Indolebutyric Acid
Component 1 (7.8%)
2.4 2.6 2.8 3.0 3.2 3.4 3.6 -1.0 -0.5 0.0 0.5 1.0
Correlation coefficients

D E Tryptophan Pathway F Important Features Heat Map

Pathway Analysis

C00078 Pre Post

Tryptophan
4-Chloroaniline
p = 0.03 Pyrocatechol
p = 0.06
0.2
Methoxyindoleacetic Acid

Concentration
0.2 N-(3-Hydroxypropyl)phthalimide

Concentration
0.1
0.0 0.1 Gly-Gly-Phe
C05660 -0.1 0.0 C00398 Androstenediol
C00328
5-Methoxyindole- -0.2 -0.1 Tryptamine Leu-Tyr-Thr-Lys
acetate L-Kynurenine -0.3 5-Carboxyvanillic Acid
-0.2
1-Methyladenosine
p = 0.001 p = 0.04 Pre Post Pre Post N6-Methyladenosine
0.3
(R)-2-Hydroxy-3-phenylpropionic acid
Concentration

0.2
Concentration

0.1 0.1 S1P(d18:1)

0.0 0.0 C05653 C00954 Phenethylamine
-0.1 Formylanthranilate Indole-3-acetate Benzaldehyde
-0.1 -0.2 Purine
-0.3 L-Phenylalanine
-0.2 p = 0.001
p = 0.42 Indole-5-carboxylic acid
Pre Post Pre Post
Gln-Gly

Concentration
0.4
0.20
Gly-Gln
Concentration
0.15 0.2
0.10 Gly-Lys
0.0
0.05 3-Indolebutyric Acid
KEGG ID 0.00 -0.2 Glu-Gly
-0.05
Compound Name
-0.10 Gly-Glu
Pre Post N-acetyl-D-phenylalanine
-0.15
Pre-to-Post Δ Concentration Pre Post Val-Val
Pathwayimpact
– 0 +
Class Pre Post
Value
2.7 0 -2.8

Fig. 6 | Metabolomics (n = 20 participants). A PLS-DA Scores Plot. Scatter plot metabolites highlight changes in neurotransmitter precursors, lipid signaling, and
shows 2 principal components with greatest variation. Ovals show 95% confidence amino acid metabolism. D Pathway Analysis. Metabolite pathways’ impact scores
intervals. Similar observations cluster together. Component 1 (x-axis) contains 7.8% (x-axis) and p-values (y-axis, -log10 transformed). Larger impact values indicate
of total variation; component 2 contains 7.1%. Oval cluster spatial separation greater contribution to the pathway. Bubble size reflects the pathway’s enrichment
indicates systematic differences between timepoints. B PLS-DA Variable Impor- score, and bubble color corresponds to significance level (darker indicating lower p-
tance in Projection (VIP) Scores. VIP scores indicate metabolite’s contribution to value). E Tryptophan Pathway. Colored boxes represent detected metabolites.
timepoint separation, ranked in descending order. Top metabolites are involved in Color scale represents magnitude of pre-to-post change. Box plots show pre and post
neurotransmitter regulation, lipid signaling, and RNA modification, suggesting the concentration distributions and paired t-test values with raw p-values. F Important
intervention induced broad metabolic and molecular adaptations. C Top PLS-DA Features Heat Map showing top differentially expressed metabolites, including
Compounds. Correlation coefficients between metabolites and PLS-DA dis- several phenylalanine-related metabolites involved in neurotransmitter synthesis
criminant function. Each metabolite’s correlation coefficient indicates strength and and metabolic regulation. Color scale indicates relative abundance (red = higher;
direction of association with timepoint separation (positive = direct, negative = blue = lower). Clustering highlights metabolic pathway shifts, including phenyla-
inverse, closer to ± 1 = more influential in driving group separation). Top lanine metabolism and lipid signaling.

Among top-ranking metabolites, three were related to phenylalanine indicating significant shifts in non-coding exRNA expression during the
metabolism, including CAN-2-Hydroxy-3-phenylpropionic acid, N-acetyl- intervention. At least 5.99% of the variance was attributed to experience
D-phenylalanine, and phenethylamine, which can act as neurotransmitter levels, with partial separation between novice and advanced meditators
storage, contribute to stress adaptation, and enhance dopamine, nor- (Fig. 7D). Correlations between principal components and key variables
epinephrine, and serotonin release. Two RNA-related metabolites, such as timepoint and experience reveal that PC1 and PC2 strongly correlate
1-methyladenosine and N6-methyladenosine, were linked to inflammatory with timepoint (R² = 0.35 and 0.28, respectively), while PC9 moderately
responses, with roles in RNA metabolism, methylation, and cellular sig- correlates with experience (R² = 0.31), indicating that both influence exRNA
naling. 3-indolebutyric acid, associated with tryptophan metabolism, may expression (Fig. 7E).
influence neurotransmitter synthesis and synaptic plasticity. Several We also identified exRNAs mapped to at least 66 annotated protein-
dipeptides were identified as enhancers of gut microbial activity, promoting coding mRNAs upregulated or downregulated post-intervention (Fig. 7F, G).
SCFA production, which is crucial for regulating inflammation and immune Some of the upregulated exRNAs could be mapped to ras/rab interactor 1
function. S1P (d18:1), a signaling lipid, may reflect gut barrier integrity and (RIBC1), synapsin 3 (SYN3), and glutamate ionotropic receptor kainite type
inflammatory pathway alterations. Figure 6F shows a distinct clustering of subunit 3 (GRIK3), genes linked to synaptic function and neurotransmission.
metabolites by relative abundance, highlighting these key contributors. The enrichment of exRNAs specific to solute carrier family 27 member 1
(SLC27A1) and proprotein convertase subtilisin/kexin type 9 (PCSK9) could
Exosome-specific transcriptomics enhance neural signaling and metabolic regulation. Downstream functional
We analyzed differentially expressed exosome-specific extracellular analysis of exRNAs in pre-vs-post exosome fractions against the Reactome
microRNAs, non-coding RNAs, and RNAs mapping to protein-coding database46 (Fig. 7H) predicted pathways related to neurotransmission (Ser-
mRNAs. Data from 6 participants was excluded at preprocessing (n = 16). otonin/Dopamine Neurotransmitter Release Cycle), further highlighting the
At least 18 non-coding exRNAs (p < 0.1, log2FC > ±0.58) exhibited distinct enhancement of synaptic activity post-intervention. Metabolic pathways,
expression profiles on pre and post timepoints (Fig. 7A, B). Principal including Transport of Vitamins and Nucleosides, could reflect broader
components (PC) explain 46.3% of the total timepoint variation (Fig. 7C), metabolic changes.

Communications Biology | (2025)8:1525 8

https://doi.org/10.1038/s42003-025-09088-3 Article

A B C
Differential non-coding exRNA Small non-coding exRNA PCA Time Point
Pre Post
EPIST
LINC01962 MIR200B
4
LOC124902988
1.5 EPIST
ATP11A−AS1 MIR30A LINC01962
MIR200B

PC2 14.56% variation

LINC02048
MIR106B
LOC101929200 CT69 TTTY14 TTTY14
MIR215
LINC00578 MIR215 0
-log10p

STAM−DT MIR455 LOC124902988

MIR194−1
IRF2−DT LOC102546299 MIR455 MIR30A
1.0 LINC02048
MIR194−1 LOC102546299
CT69 -4
LOC101929200
STAM−DT
MIR106B
IRF2−DT
0.0 LINC00578 -8
−5 0 5
ATP11A−AS1 -10 -5 0 5
log2foldchange

F_1580_Adv
F_2119_Nov
F_4018_Adv
F_4019_Adv
F_4021_Adv
F_4026_Nov
F_4027_Nov
M_1960_Adv
M_4014_Adv
M_4020_Adv
M_4022_Nov
M_4023_Nov
M_4099_Adv
F_1580_Adv
F_2119_Nov
F_4018_Adv
F_4019_Adv
F_4021_Adv
F_4026_Nov
F_4027_Nov
M_1960_Adv
M_4014_Adv
M_4020_Adv
M_4022_Nov
M_4023_Nov
M_4099_Adv
PC1 31.8% variation

Pre Post
D PCA Experience
log2(TMM-normalized counts) F
4 Differential protein-coding exRNAs

10
11
5
6
7
8
9
3
E PCA Pearson R2 Correlates
PNMA8A
PC10 2.51% variation

10
1

9
PC

PC

PC

PC

PC

PC

PC

PC

PC

PC
RIBC1

0
Age 0.09 0 0.08 0.07 0 0.02 0.19* 0 0.02 0.03
HMCES
Biological sex 0.01 0 0.02 0.04 0.12 0.05 0.02 0 0.04 0.02 2

-log10p
Level 0.01 0.03 0.02 0 0.04 0.02 0 0 0.31** 0 FBXO45
-2 IL31 ZNF532
Meditation 0.35** 0.28** 0.03 0.04 0 0.01 0.07 0.01 0 0.04 SDK2 SLCO2A1 TOB2
TSTD1 COX19
ARFIP1 PCSK9 SGCEEVCUBE2HPPFIA4
NUP98GRIFIN
MATN4TTC22 O
-2 0 2 PODXL2 HMGXB3FAM178B CD58 PCDHGB3
R-squared SYT12 OGFOD NLRP7 GPRC5C NUCB2ADCY6
PC9 3.48% variation KNOP1 SETD3
ZBTB38 3 STA SLC11A2CADM2SPX SH3TC1
OR1N1
0 .05 .10 .15 .20 .25 .30 .35 .40
SNORC SNCA CEBPD ZNF875 MACROD1
Novice Advanced RADX DAZAP1B1 PHIP GRIK3 PRRT4 REC114
1 MED16 RPL8 GCC2 GGPS1

G
exRNAs mapped to genes
0
ARHGAP23

−5 0 5 10
MACROD1

FAM171A1
PCDHGB3

DCLRE1C

SLCO2A1
DNAJC15
COL16A1

FAM178B
SLC27A1

SLC11A2

HMGXB3

OGFOD3
GPRC5C
PNMA8A

PODXL2
CYB5D2

AKR1C3

DAZAP1
SH3TC1

FBXO45
REC114

SLC7A9

ZBTB38

log2foldchange
ZNF532

NIBAN3

ZNF875

SNORC

HMCES
C8orf90

CADM2

ARFIP1
PPFIA4

GGPS1

CEBPD
NUCB2

GRIFIN

SBNO2

KNOP1
UBE2H

ADCY6

OR1N1

CARS1

MED16
ENOX1
MATN4

PCSK9
PRRT4

NUP98

NLRP7
COX19

SETD3
TSTD1

STAB1
TTC22

SYT12
GRIK3
RIBC1

SGCE

GCC2

SNCA

PITX3
RADX
SYN1

SYN3

SDK2
TOB2

CD58

PHIP
GRP
EVC

SPX

IL31

M_4099_POST_E
M_4023_POST_N
M_4022_POST_N
M_4020_POST_E
H Upregulated Reactome Pathways (p < .01)
M_4014_POST_E
M_1960_POST_E
Post

F_4027_POST_N
F_4026_POST_N Transmission across
F_4021_POST_E chemical synapses
F_4019_POST_E
F_4018_POST_E Neuronal system
F_2119_POST_N
F_1580_POST_E
Serotonin release cycle
M_4099_PRE_E
M_4023_PRE_N Dopamine release cycle
M_4022_PRE_N
M_4020_PRE_E
M_4014_PRE_E Neurotransmitter release cycle
M_1960_PRE_E
SLC transporter disorders
Pre

F_4027_PRE_N
F_4026_PRE_N
F_4021_PRE_E Vitamins, nucleosides &
F_4019_PRE_E
F_4018_PRE_E
related molecules transport
F_2119_PRE_N
F_1580_PRE_E 0.09 0.12 0.15
Count p-value Gene ratio
log2(TMM-normalized counts)
2 3 4 5 .05 .10 .15
6 8 10 12 14

Fig. 7 | Exosome transcriptomics (n = 16 participants). A Differential non-coding metadata and test significances (** denotes p < 0.01; * denotes p < 0.05). Values
exRNAs (miRNAs, ncRNAs) on volcano plot of -log10p versus log2FC (pre/post correspond to Pearson R2 values and are colored by significance. F Differential
difference). Blue/red dots denote exRNAs prevalent in pre/post-intervention exo- protein-coding exRNAs (excluding ncRNAs) on volcano scatter plot of -log10p
somes, respectively; green dots denote exRNAs without significant difference. versus log2FC (pre/post difference). Blue/red dots denote exRNAs prevalent in pre/
Dashed orange lines indicate p < 0.1 and log2FC > ±0.58 thresholds. Full list of post exosomes, respectively; green dots denote exRNAs without significant differ-
exRNAs on Supplementary Table 17. B Differential expression (log2-transformed ence. Dashed orange lines indicate p < 0.1, log2FC > ±0.58 thresholds. G TMM-
TMM-normalized counts) of 18 non-coding exRNAs. Colors indicate counts. normalized counts of log2-transformed protein-coding exRNAs with highest var-
Sample name: (M male, F female)_participantID_(pre; post)_(E, advanced; N, iance across samples. Color scale indicates counts. Sample name: (M male, F female)
novice). C, D Principal component analysis of top normalized exRNA counts. _participantID_(pre; post)_(E, advanced; N, novice). H Reactome pathway analysis
Encircled are clusters corresponding to C pre/post, and D novice/advanced. of protein-coding exRNAs enriched in pre-vs-post intervention exosomes. Color
E Eigenvector plot showing the correlation of principal components to variables’ indicates p-value; circle size indicates exRNA count per pathway.

Machine learning and feature-MEQ correlation Focusing first on timepoint classification, both models achieved strong
To identify features differentiating timepoints and experience levels discrimination between pre- and post-meditation states (XGBoost AUC =
across metabolomic, proteomic, fMRI, and RNA outcomes, we eval- 0.86; Random Forest AUC = 0.90). Top XGBoost predictors included Gly-
uated the data using Random Forest and XGBoost models. After nor- Gly-Phe, 3-indolebutyric acid, Gly-Lys, and connectivity between the
malization and dimensionality reduction, performance was assessed anterior and salience networks, default mode–salience, and precuneus
with tenfold cross-validation using F1 score, precision, recall, and area regions. Random Forest highlighted overlapping features alongside addi-
under the receiver operating characteristic (AUROC). Both models tional metabolites such as Androstenediol, S1P(d18:1), N6-methyladeno-
demonstrated robust classification across conditions, with AUC values sine, and purine derivatives. Many of these metabolites are linked to amino
ranging from 0.70 to 0.93. SHAP analyses revealed key contributors acid metabolism, neurotransmitter balance, and cellular stress signaling,
spanning metabolites, fMRI connectivity features, proteins, and non- while sphingolipids such as Androstenediol and S1P(d18:1) implicate lipid
coding RNAs (Fig. 8A, B). signaling and neuroendocrine function. Connectivity differences involving

Communications Biology | (2025)8:1525 9

https://doi.org/10.1038/s42003-025-09088-3 Article

A Top Machine Learning Features - Time Point B Top Machine Learning Features - Experience Level

XGBoost Random Forest XGBoost Random Forest

1.0 1.0 1.0 1.0

0.8 0.8 0.8 0.8

True Positive Rate

True Positive Rate

True Positive Rate

True Positive Rate

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

ROC curve (AUC = 0.86) ROC curve (AUC = 0.90) ROC curve (AUC = 0.70) ROC curve (AUC = 0.93)
0.0 0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
False Positive Rate False Positive Rate False Positive Rate False Positive Rate

Gly-Gly-Phe Androstenediol Fibroblast Growth Factor 19 AN – SN (rest)

3-Indolebutyric Acid Gly-Gly-Phe Interferon Gamma Myocyte Enhancer Factor 2C
AN – SN (rest) 3-Indolebutyric Acid AN – DMN (meditation) Lactate Dehydrogenase A
Gly-Lys Gly-Lys VN – DAN (rest) DAN – DMN (meditation)

DMN – SMN (rest) S1P(d18:1) AN – SN (rest) Myocyte Enhancer Factor 2D

N-acetyl-D-phenylalanine N-acetyl-D-phenylalanine left AI - left AG (meditation) left AI – left AG (meditation)

Pseudouridine Synthase N6-methyladenosine C-C Motif Chemokine Ligand 5 Interferon Gamma

Precuneus – right dlPFC (rest) AN – SN (rest) Sirtuin 3 VN – ECN (meditation)

Precuneus – left AG (rest) Indole-5-carboxylic Acid Fibroblast Growth Factor 19 SMN – SN (rest)

AN – ECN (rest) Gln-Gly Sirtuin 2 ATP Synthase ATP5PB

Precuneus – right AG (rest) Leu-Tyr-Thr-Lys Phosphatase & Tensin Homolog Prodynorphin

Sidekick 2 Protein Benzaldehyde Neurotrophin 4 Sirtuin 2

Fibrinogen-like Protein 1 Sidekick 2 Protein Toll Like Receptor 2 Neurotrophin Receptor Tyrosine Kinase 2

AN – SMN (rest) Purine GAPDH SN – DAN (meditation)

Sum of 380 other features Sum of 380 other features Sum of 380 other features Sum of 380 other features

-2 -1 0 1 -.15 -.10 -.05 0 .05 .10 .15 .20 -.4 -.2 0 .2 .4 .6 -.100 -.075 -.050 -.025 0 .025 .050
SHAP value SHAP value SHAP value SHAP value

Lo Feature Value High Lo Feature Value High Lo Feature Value High Lo Feature Value High

C ML Features - MEQ Correlations

XGBoost Time Point Features Random Forest Time Point Features XGBoost Experience Features Random Forest Experience Features

Pre Post Pre Post Pre Post Pre Post

t)
cid * nin
e
Ac
id * ine ( res in
4
se
A
r 2C
ic A lala ric lan AN op
h na cto
tyr ny uty yla ge Fa
bu st) he s leb ly en s –D t) uro
tr
H. dro ) E.
ole (r e D-p -Ly do -G -ph -Ly VN re s Ne hy ati
on
yte P5
PB
3-I
nd G tyl- Gly 3-In Gln l-D Gly N( ns
in De dit oc AT
RA st) ce s cety –S 4 Te 2 tate ( me My ase )
s –
(re N -a st) -Ly N-
a ly AN e& tor La
c N PB on
n eu N N (re Gly ln -G
op
h in
ata
s
ep – EC T P5 S ynth dit
ati
cu M M G tr ph ec A P me
Pre –S st) –S st) l uro r2 os eR VN ase t) AT ( A
AN in dio Ne
H es N
C (re AN (re rote ne ce
pto Ph ll Lik A PD ynth SN (r EC n ase
F AG 2P ste To G PS – ge
dlP –R st) ick dro Re t) AT – C VN dro D
–R st) us (re ek ys An cid ike DH res AN r2 hy r2
(re ne r-L ic A ll L AP N( cto De cto
ne
us AG u A G Sid r-T
h
uty
r To G a Fa
cu –L Pre
c
–R y
leb –S E.
F 2 cta
te
E.
Pre us Lys us cid u-T res
t)
do :1) nd
5
AN
3 TK La TK
2
ly- ne ic A Le 18 iga tuin yte NR ocy
te
u ne G cu bu
tyr N( 3-In P(d .L S ir
My
oc y NR
Pre
c
Pre st) –S S1 n) t) M
in ole (re AN rin
e
res
t) em tio
2 res res
t)
rote 3-I
nd
CN Pu N( Ch ita tuin N( N(
2P E S M e d Sir S S
ick se – st) – e C-
C (m 3 t) N– n) – )
ek tha AN (re ine AN s in G tuin res SM tio AN on
Sid yn SN lan no LA Sir N( ed
ita itati
S – yla e I- n)
A (m ed
din
e
es
t) AN (re
st)
ph
en id la d id LA tio –D on
) G r2
D (m 2
uri N (r D- Ac thy Ac ita VN ati LA cto AN tuin
eu
do
–S MN tyl- ylic me ylic ed dit I– Fa –D Sir
Ps
S 1 e o x l N6
-
rbo
x ( m H. me ) A . N
AN (re
st) N– in
N-
ac arb dio ys MN N( on L te
E hin S st)
DM rote 5-c ne -5-
ca r-L sin DM ati ocy orp (re
MN eP le- ste Th –D en dit My dyn SN
S -lik se o r o ole yr- AN &T AN
– e ro
N– )
N– ine en tha Ind An
d :1) Ind u-T Ph
e se in
2 (m
nd
5 P on
) on
DM lan og yn 18 Le ly-
G
itati SM ati
yla rin eS P(d -G ph
a ta
Sir
tu LA iga ed dit
en Fib in t) S1 ly os I- .L (m ( me n)
-ph res
t)
uri
d res Ph
e G de Ph
a LA em 9 2
tio
l-D N( do C( ly- hy mm Ch F1 AN tuin M N a
ety eu lPF -G lde Ga CM FG –D Sir D
ed
it
ac EC Ps Rd
e Gly za n - SN N– (m
N -
AN
–
in
1
s– ly-
Ph
hy
de Be
n
rin
e
rfe
r o
F1
9 * C
F1
9
mm
a DA
LA
G
orp
hin *
rote eu -G lde Pu Inte FG FG Ga
eP un Gly za ron I– dyn
n- li k h e
Pre
c in n
in e * in 9 * a * rfe on
) LA Pro a *
ge ly-
P rote Be
os rote F1 mm ati mm
o G P G Ga Inte dit Ga
rin Gly
-
ick
2 en 2P F
ron me on
Fib ek la d kic
k (
rfe
r
Sid thy e rfe MN
e Sid Inte D Inte
-m N–
N6 DA

-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
Spearman R Spearman R Spearman R Spearman R

Fig. 8 | Machine learning and MEQ-features correlations (n = 20 participants). (somatomotor network), ECN (executive control network), VN (visual network),
A ROC curves and SHAP plots for XGBoost and Random Forest models predicting DAN (dorsal attention network), dlPFC (dorsolateral prefrontal cortex), AI (ante-
pre/post classification: AUC = 0.86 (XGBoost) and 0.90 (Random Forest) indicate rior insula), AG (angular gyrus), GAPDH (glyceraldehyde-3-phosphate dehy-
good classification performance. SHAP plots display the top contributing features drogenase), ATP5PB (ATP synthase peripheral stalk membrane subunit B), FGF
ranked by impact on model output. B ROC curves and SHAP plots for XGBoost and (fibroblast growth factor), NTRK2 (neurotrophin receptor tyrosine kinase 2).
Random Forest models predicting novice/advanced classification: AUC = 0.70 C Heatmaps of Spearman R correlations between Mystical Experience Ques-
(XGBoost) and 0.93 (Random Forest). Abbreviations: R (right), L (left), AN tionnaire (MEQ) scores and the top machine learning features per model and time
(auditory network), SN (salience network), DMN (default mode network), SMN point, with * denoting FDR-adjusted statistical significance (pFDR < 0.05).

salience, executive and default mode networks (SN, ECN, DMN, PC) align Sirtuin 2. Across both models, large-scale connectivity differences involving
with prior literature that mediation can reorganize large-scale brain net- salience, executive, and default mode networks (e.g., AN–SN, VN–DMN,
works that govern salience47, attention48, interoception49, and self-referential SN–DAN) emerged as consistent drivers of classification. These findings
processing47. Taken in concert, timepoint differences identify shifts that indicate that long-term meditation experience is associated with distinct
integrate across neurofunctional and metabolic layers, reinforcing the molecular and network-level adaptations that support resilience and energy
concept that mediation acts as a systemic regulator of mind-body regulation.
physiology. Finally, to explore links between biological features and subjective
To evaluate the influence of meditation experience, post–pre deltas outcomes, we correlated model-identified predictors with Mystical
were compared between novice and advanced practitioners. XGBoost Experience Questionnaire (MEQ) scores. Several metabolites and con-
achieved modest classification performance (AUC = 0.70), while Random nectivity measures showed nominal associations (p < 0.05), with a subset
Forest reached higher discrimination (AUC = 0.93). SHAP analysis indi- surviving FDR correction (asterisks, Fig. 8C). In the timepoint analysis,
cated that immune- and stress-related proteins, including fibroblast growth baseline 3-indolebutyric acid correlated positively with MEQ (r = 0.72,
factor 19, interferon-γ, C-C motif chemokine ligand 5, and Toll-like q < 0.05), whereas N6-methyladenosine correlated negatively (r = −0.67,
receptor 2, were major contributors in XGBoost, whereas Random Forest q < 0.05). At the experience level, immune markers including FGF19 and
emphasized metabolic and mitochondrial features such as myocyte interferon-γ were negatively associated with MEQ deltas (r = −0.59 to
enhancer factors, lactate dehydrogenase A, ATP synthase ATP5PB, and −0.73, q < 0.05), and reductions in network connectivity (DAN–DMN,

Communications Biology | (2025)8:1525 10

https://doi.org/10.1038/s42003-025-09088-3 Article

Cortical hierarchy

high level
reconceptualization
priors

+
ascending descending
prediction predictions
errors

expected
meditation precision
low level
priors
neuromodulatory cell

– superficial pyramidal cell

deep pyramidal cell

sympathetic/
healing ritual sensory input autonomic set points parasympathetic physiology
activation

Fig. 9 | Potential cortical implementation. Hierarchical predictive coding scheme new predictive avenues; and meditation weakens descending predictions to facilitate
showing how the three mind-body techniques may synergistically facilitate a more their replacement with new beliefs and present-centered sense data and activates
flexible and adaptive predictive system: Reconceptualization remodels priors and sympathetic and parasympathetic autonomic responses. New predictions alter
hyper-priors; the healing ritual, acting as an open-label placebo, opens the system to allostatic setpoints and autonomic regulation.

LAI–LAG) were inversely correlated with MEQ outcomes in experienced needs, with experience and behavior largely shaped by the system’s priors
meditators (r = –0.64 to −0.68, q < 0.05). Together, these exploratory and predictive architecture56. Predictive brain regions project efferent copies
findings suggest that both peripheral metabolic and immune markers and to the subcortical nuclei that regulate metabolic, endocrine, immune, and
central network dynamics are tied to reported mystical experience levels other physiological setpoints33,57 based on current and predicted energy
during meditation. needs. Thus, physiological and psychological health can be targeted by
shifting the priors and functioning of the predictive system.
Discussion Each mind-body technique in the study may be thought of as acting on
Our study shows how an intensive non-pharmacological mind-body a different part of the predictive architecture (Fig. 9): Reconceptualization
intervention produced broad short-term neural and plasma-based mole- changes how participants believe their minds construct reality and affect
cular changes associated with enhanced neuroplasticity, metabolic repro- their bodies, which is equivalent to reconfiguring the priors (beliefs about
gramming, and modulation of functional cell signaling pathways. the causes of symptoms or sense data) and hyperpriors (beliefs about beliefs
fMRI data showed that this meditation style functionally disrupts the about the causes of sense data) that structure how interoceptive and
default mode and salience networks (responsible for self-referential thought exteroceptive sensory signals are interpreted and experienced. The open-
and allostatic regulation32,50) and cerebellum-prefrontal predictive proces- label placebo-like healing ritual enacts a known health-promoting behavior
sing circuits involved in integrating internal models with external sensory (healing) that does not conform to rational norms, creating a mismatch
data51. This complements studies showing meditation-induced changes in between motor/affective behavior and the experience-driven self-fulfilling
DMN connectivity32,52. Mindfulness meditation has been linked to DMN prediction of sense data, which opens the system to new predictive (sensory
deactivation32,53 and stronger connectivity between DMN-Salience and and allostatic) paths30. Finally, meditation, by focusing attention on the
DMN-Executive Control networks47, likely reflecting an increased capacity present, weakens the predictive processes themselves to produce a state of
to switch in/out of default mode dominance. Here, the main effect we pure awareness that is free from the predicted self 50 in which priors are more
observed was a meditation-driven decrease in intra-network DMN con- easily replaced with present-centered sense data. The three techniques may
nectivity and a broad desynchronization of whole-brain connectivity. therefore work synergistically to facilitate a more flexible and adaptive
Meditation also reduced prefrontal-cerebellar connectivity both pre- and prediction system, accounting for the personal transformations anecdotally
post-intervention, suggesting a state-dependent suppression of self- reported by participants and the downstream neuroplastic and molecular
referential and evaluative processing consistent with focused interoceptive changes. The observed changes in blood biomarkers may have resulted from
awareness, non-judgmental awareness, and a state that transcends the self— the endocrine, immune, and other regulatory changes evoked by the
all features of the guided meditations. DMN-cerebellar connectivity intervention’s effects on these predictive allostatic mechanisms, as well as
alterations have been reported for depressive disorder and during previous from meditation-induced effects on both sympathetic and parasympathetic
meditation studies, albeit with opposite (higher connectivity) effects54,55. Our branches of the autonomic nervous system22, a finding common to other
machine learning analyses further reinforced these findings by demon- meditation studies58, which may also account for the concomitant upre-
strating that meditation-related shifts in fMRI connectivity were embedded gulation of inflammatory and anti-inflammatory proteins during the
within broader molecular changes across metabolomic, proteomic, and intervention.
transcriptomic layers. The convergence of neural and peripheral predictors, MEQ-30 scores revealed that participants experienced mystical-type
including amino acid metabolites, lipid signaling molecules, immune reg- experiences during meditation, reflected in neural activity by strengthened
ulators, and network-level connectivity, underscores that meditation connectivity between the left insula and the PCC—a hallmark sign of trance
engages a systemic mind-body axis rather than isolated pathways. states—and diminished DMN connectivity and whole-brain modularity—
We can interpret these findings with a Bayesian brain framework, consistent with previous meditation studies32,59. Elevated post-intervention
which posits that the brain probabilistically predicts incoming sensory plasma levels of SLITRK1 and NGFR and the inducement of greater glu-
information based on prior beliefs to allostatically regulate the body’s energy tamatergic dendritic growth by post-intervention plasma are both similar to

Communications Biology | (2025)8:1525 11

https://doi.org/10.1038/s42003-025-09088-3 Article

effects by serotoninergic entheogens, which some have compared to med- sample size and partial reliance on experienced meditators may limit the
itative states60,61, and consistent with neuroplastic changes mediated by the generalizability of our findings, and future research should include larger,
BDNF pathway. While follow-up mechanistic studies are needed to estab- more diverse cohorts. Additional mechanistic studies should investigate the
lish causality, we provide converging evidence at multiple scales to suggest molecular targets implicated by this study to causally link the neuroplastic
that the mind-body intervention produced neuroplastic changes mediated and phenomenological changes to the molecular pathways implicated by
by these BDNF-associated proteins. our work.
Beyond neural changes, we found that post-intervention plasma Specific methodological limitations include the following. Although
enhanced glycolysis in treated cells and contained higher levels of glycolysis- data on meditation experience and practice frequency were collected for
associated proteins ENO2 and LDHA. While the in vitro results may be retreat-based meditations within the study cohort, equivalent information
interpreted as a Warburg effect on neuroblastoma cells, our proteomic data regarding other forms of meditation practice—apart from the binary indi-
suggests that glycolysis became enhanced in participants. Notably, a similar cation of their presence or absence—was not obtained. Potential circadian
phenomenon was observed in experienced Tibetan monks24, for whom the and metabolic confounds were introduced by variable blood collection
enhanced glycolytic phenotype was associated with a cardioprotective times (up to 8-h range on Day 8), pre-collection fasting durations, and time
plasma proteome, improved oxygen release, and decreased atherosclerosis. elapsed after the meditation intervention ended (up to 48 h for blood col-
This underscores the complex interplay between neural and non-neural lections and 24 h for fMRI acquisition) between participants. Dietary factors
mechanisms in mind-body interventions and warrants further investiga- and fasting may also have introduced confounds in proteomic and meta-
tion. Since the brain’s predictive architecture is evolutionarily geared bolomic measures since no standardized diet was implemented and fasting
towards regulating metabolic outlays, we speculate that the present-cen- times beyond 30-min pre-blood collection were not controlled during the
tered, sensory-driven brain state induced by meditation requires a more intervention. The short duration of the resting state scan can limit our ability
dynamic energetic profile made possible by glycolysis’ fast response times. to attribute connectivity-derived features solely to the experimental
The significant pre-to-post intervention increases in beta-endorphin variables64,65. The open eyes paradigm during fMRI BOLD acquisition
and dynorphin point to the engagement of the endogenous opioid system chosen to reduce the risk of drowsiness likely introduced potential visual
without deception. We suggest that reconceptualization altered high-level confounds. Finally, denoising with ICA-AROMA + WM/CSF regression is
priors about the mind’s ability to influence the body, leading to new pre- limited in removing physiological noise in regions with strong cardiac or
dictions that modulated endorphin levels. Since reconceptualization creates respiratory activity, which may have resulted in additional confounds.
a steady belief state, this effect is likely sustained for longer than deception-
based placebos. In a world in which up to 50% of American physicians Conclusions
regularly and knowingly prescribe placebo medications62, interventions that This observational study is the first to investigate the joint effects of three
activate a self-mediated, conscious, and non-deceptive pain relief mind-body interventions–meditation, reconceptualization, and open-label
mechanism carry great potential, especially for pain conditions without a placebo healing rituals–on neural activity and plasma physiology. Our
well-described physical etiology. Future research should investigate which findings show these techniques may act synergistically to produce mystical-
specific beliefs associated with health, disease, mind, and body sustainably type experiences, enhance neuroplasticity, reprogram metabolic pathways,
optimize endogenous opioid modulation, as well as if and how meditation and modulate endogenous opioids, highlighting the potential of mind-body
aids in initiating and sustaining these changes. interventions to effect profound changes in neural activity and physiology.
Our transcriptomic analyses reveal significant changes in circulating
exRNAs implicated in neural activity, metabolic processes, and cellular Methods
signaling. Specifically, key upregulated transcripts such as RIBC1 (RIB43A Participant recruitment
domain with coiled-coils 1) and MIR455 (microRNA 455) mapped to The study was advertised to all 1444 registered retreat attendees via
pathways involved in neurotransmission and glycolysis. Pathway enrich- email invitation, of which 561 expressed interest by answering an
ment analysis highlighted upregulated Reactome pathways related to online questionnaire to determine eligibility (Fig. 1D). Inclusion
synaptic transmission and neurotransmitter cycling, consistent with the criteria included being an English speaker; being at least 21 years of
observed neural connectivity and neuroplasticity changes. Together, these age; being in good general health; being able to provide blood sam-
results provide multi-level evidence that the mind-body intervention ples before, during, and after the retreat; and agreeing to undergo
induced metabolic reprogramming, neuroplastic changes, and endogenous fMRI neuroimaging before and after the retreat. Exclusion criteria
opioid modulation, potentially mediated by dynamic exRNA activity and included inability to consent, inability to follow or comply with study
glycolytic adaptation. procedures, current use of psychoactive medications, contra-
Interestingly, simultaneous activation of inflammatory and anti- indications for phlebotomy, and contraindications for MRI (preg-
inflammatory protein pathways is reminiscent of prior findings showing nancy, history of seizures, electronic or ferromagnetic medical
mind-body techniques can enhance resilience to environmental stressors implants or devices, claustrophobia). Of 65 eligible participants, 36
via gene expression63. Cytokines like IL-6 are known to have both pro-and were randomly selected to participate and provided written informed
anti-inflammatory roles depending on the context acting as both acute- consent. Of 27 who consented, five were dropped due to scheduling
phase reactants during stress, but also stimulating IL-1 receptor antagonist conflicts; one participant was dropped for not complying with MRI
and IL-10 in exercise and repair settings64. While inflammation is often facility masking requirements; and one participant failed MRI
viewed as a harmful response, increased cellular turnover and tissue screening due to a heart stent. The study size was determined by the
remodeling could occur as part of an adaptive response to these interven- scanning time available at the fMRI facility (1 MRI scanner for 2 days
tions – another area worthy of further investigation. pre- and post-intervention). Our study included male (n = 6) and
While our study provides valuable insights into the broad effects of female (n = 14) participants (gender identity), and similar findings
mind-body interventions, we acknowledge several important limitations. are reported for both genders. Of the 6 male participants, 2 had been
First, the uncontrolled observational design limits our ability to infer practicing the meditations carried out in the workshop daily for more
causality or disentangle the relative contributions of meditation, placebo, than 1 year, 3 had been practicing the meditations daily for less than
reconceptualization, and other factors such as expectations, diet, and 1 year (one of whom practiced a different form of meditation), and 1
relaxation or distance from routine stressors. Future studies should imple- participant had not previously practiced the meditations before
ment more rigorous, controlled designs to test how these elements interact attending the retreat, though he had a different meditation practice.
and whether the non-predictive meditation state facilitates the replacement Of the 14 female participants, 8 had been practicing the retreat
of maladaptive priors during reconceptualization. Additionally, the small meditations for over 1 year (most being a daily practice), and 1 of the

Communications Biology | (2025)8:1525 12

https://doi.org/10.1038/s42003-025-09088-3 Article

8 also practiced a different form of meditation. Six of the 8 females estimated with Topup70. Anatomical T1-weighted (T1w) images were
had been practicing the meditations for less than 1 year, with only 3 corrected for intensity non-uniformity using ANTS 2.3.371, skull-strip-
reporting an almost daily practice, and 4 of the 8 reported having a ped, segmented using Fast (FSL 6.0.5)72, and normalized to standard
different meditation practice. One of the 8 females had not previously space (MNI152Nlin2009cAsym) with nonlinear registration (ANTs
practiced the meditations before attending the retreat, though she 2.3.3). Brain surfaces were reconstructed using recon-all (FreeSurfer
had a different meditation practice. The retreat began at 5 pm on Day 6.0.1)73. Due to a corrupted T1w image for participant 4023, co-
1 and ended at 1:30 pm on Day 7 and was conducted in April 2022 at registration failed, and 4023 was excluded from fMRI analysis.
the Manchester Grand Hyatt (San Diego, CA, USA). The diet was BOLD head motion parameters were estimated using MCFLIRT
provided by the hotel. There were no adverse events. (FSL 6.0.5)74. Functional runs were co-registered to anatomical T1w
The study was conducted in accordance with the Declaration of Hel- references using boundary-based registration (FreeSurfer) and resampled
sinki principles and all relevant ethical regulations. Experimental protocols to standard space. Confounding time-series were calculated for CSF/WM
were approved by the Western Institutional Review Board (WIRB; now region-wise global signals, and motion artifacts were identified using
WCG-IRB; Protocol MED02#20211477) and registered on clinicaltrials.gov independent component analysis (ICA-AROMA)75. Non-steady state
(NCT 06615531). Written informed consent was obtained from all parti- volumes were removed and spatial smoothing with an isotropic Gaussian
cipants prior to study inclusion, and clinical records are housed at VitaMed kernel of 6 mm FWHM (full-width half-maximum) was applied.
Research (Palm Desert, CA) as mandated by federal laws. All ethical reg- Denoising using Nilearn 0.9.276 was performed by detrending, standar-
ulations relevant to human research participants were followed. dizing, and bandpass filtering (0.01–0.1 Hz) time series data, and by
regressing ICA-AROMA motion artifacts and mean CSF + WM
Functional magnetic resonance imaging (fMRI) signals75. Physiological noise was addressed via ICA-AROMA + CSF/
Acquisition. fMRI was performed off-site at the UCSD Center for Func- WM regressions rather than with physiological recordings. Additional
tional MRI pre-intervention (Day 0: 8 am–8 pm and Day 1: 8 am–4 pm) and denoising for whole-brain network analyses involved regressing global
post-intervention (Days 8 and 9: 8 am–6 pm). Participants were positioned signal and excluding BOLD runs with mean FD > 0.3 mm to obtain
in the MRI scanner (Siemens Prisma 3T with standard 32-channel head coil) functional connectivity distributions mean-centered around zero, which
with a respiratory transducer placed around the chest, a pulse oximeter were confirmed by visual inspection.
placed on the left index finger, and MRI-safe headphones. A structural scan,
two functional blood oxygen level dependent (BOLD) scans (resting state Functional connectivity analysis. Functional connectivity analyses
and meditation), and a diffusion tensor imaging (DTI) scan (not described were performed on seven canonical resting state networks (RSNs), eight a
here) were acquired. priori defined regions of interest (ROIs), and whole-brain networks
(Nilearn 0.9.2) (Brain Connectivity Toolbox77). To understand how this
Structural scan. Participants were instructed to “not move and keep eyes meditation style influences large-scale neural dynamics, we examined
closed.” High resolution structural images were acquired with a T1- functional connectivity and whole-brain network measures, capturing
weighted magnetization-prepared rapid gradient-echo (MPRAGE) the integration, segregation, and reorganization of brain networks.
sequence with TR = 2400 ms, TE = 2.22 ms, TI = 1000 ms, flip angle = 8°,
FOV = 224 mm, voxel size = 0.7 mm isotropic, 320 slices, slice thickness Resting state networks and regions of interest. RSNs examined
0.80 mm, saggital orientation, bandwidth = 210 Hz/Px, and acquisition included the default mode network (DMN), dorsal attention network
time = 7 min. 40 s. (DAN), executive control network (ECN), salience network (SN), sensor-
imotor network (SMN), visual network (VN), and auditory network (AN).
Functional scans. During the 5-min BOLD resting state scan, partici- The Montreal Neurological Institute (MNI) coordinates for the 36 regions
pants were instructed to “not move, keep eyes open, stay awake, and think that comprise them were extracted from the Raichle (2011) atlas78 (Sup-
about whatever you want, but do not meditate.” During the subsequent plementary Table 3). Mean denoised BOLD time series were extracted from
15-min BOLD meditation scan, an auditory recording of a guided 10 mm-radius (523-voxel) spheres centered on the MNI coordinates.
meditation from the retreat was played and participants were instructed Between-region Pearson correlations were calculated and z-transformed to
to “not move, listen to the guided meditation soundtrack, and meditate as obtain a measure of connectivity strength. Within-network connectivity was
suggested by the audio while keeping your eyes open.” An eyes open calculated as the mean connectivity between each unique pair of ROIs within
paradigm was chosen to reduce the risk of drowsiness or sleep during the a given network, and between-network connectivity was calculated as mean
meditation run, which was deemed a potentially greater risk than the connectivity between all ROIs from two networks, with each ROI pair
presence of visual confounds to accurately comparing active meditation containing one region per network. Eight a priori ROIs were additionally
with passive rest66,67. One rest and one meditation run were collected per selected for a hypothesis-driven analysis based on task-induced and con-
scan. The audio consisted of continuous expansive atmospheric music nectivity changes in other meditation studies79–82: medial prefrontal cortex
with an intermittent ethereal voice repeatedly instructing listeners to (mPFC), right and left dorsolateral prefrontal cortices (r/l dlPFC), right and
“tune into nothingness”, “no time”, “nowhere”, and “love”, combining left insular cortices (r/l IC), right and left angular gyri (r/l AG), and central
loving-kindness meditation with focused awareness of the experiencing precuneus, with MNI coordinates extracted from the DiFuMo atlas83 (Sup-
self and its potential dissolution into pure awareness. BOLD images were plementary Table 4). Mean denoised BOLD time series per region were
obtained with a gradient-recalled echo-planar imaging (EPI) sequence extracted from 10 mm-radius (523-voxel) spheres centered on the MNI
with TR = 800 ms, TE = 37 ms, flip angle = 52°, FOV = 208 mm, voxel coordinates, and between-region Pearson correlations were calculated.
size = 2 mm isotropic, 2 mm slice thickness, 72 slices, number of
volumes = 1200, matrix size = 104 × 90, bandwidth = 2290 Hz/Px, and Whole-cortex/brain networks. Networks were constructed from the
acquisition time = 5 min. (rest) and 15 min. (meditation). 25%-thresholded 48-region Harvard-Oxford 2 mm cortical atlas37 and
from the 264-region whole-brain Power (2011) atlas38 by extracting mean
Post-scan questionnaire. Immediately after each scanning session, denoised BOLD time series per region per parcellation and calculating
participants were asked to assess their experience meditating by com- between-region Fisher z-transformed correlations for all possible pairs.
pleting the Mystical Experience Questionnaire (MEQ-30)68. We constructed weighted undirected graphs and computed modularity,
global efficiency, and characteristic path length per network. Functional
fMRI preprocessing and denoising. Anatomical and functional data connectivity value distributions were visually inspected, and one outlier
were preprocessed using fMRIPrep v21.0.269. B0-field maps were run not mean-centered around zero was excluded from Power atlas

Communications Biology | (2025)8:1525 13

https://doi.org/10.1038/s42003-025-09088-3 Article

analysis. Network measures were recalculated while excluding runs with ELISA kits listed in Supplementary Table 12 following manufacturers’
mean framewise displacement > 0.3 mm to check if any significant effects instructions. Absorbance was measured with a Tecan Spark 10 M micro-
were due to motion artifacts. plate reader, and concentrations were calculated by interpolating absor-
bance values against the standard curve, previously fit with a 3-parameter
Statistics. We investigated scanner head motion per session and con- logistic curve. Pre- and post-intervention concentrations were compared
dition with a 2 (pre/post) × 2 (meditation/rest) repeated measures with Wilcoxon signed rank tests.
ANOVA on mean and maximum framewise displacement per BOLD run
with post-hoc effects confirmed with Wilcoxon signed rank tests. MEQ Plasma proteomics
score differences were compared between sessions with Wilcoxon signed To investigate the intervention’s effects on the plasma proteome, 7596
rank tests (n = 18, two subjects missing data). proteins were quantified with the SomaScan Assay v4.1 (SomaLogic,
After confirming normality and heteroskedasticity, paired t-tests Boulder, CO, USA).
compared RSNs, their component regions, a priori ROIs, and whole-cortex/
brain networks pre- and post-intervention (separately for rest and medi- SomaScan assay. Samples were thawed on ice, diluted in SomaLogic
tation scans), plus rest and meditation (separately for pre- and post-scans) to plasma diluent, and loaded onto SomaScan 96-well plates containing
investigate effects of time and of meditation, respectively. On rest vs. capture SOMAmer reagents for 7596 unique human protein targets.
meditation comparisons, the last 5 min of the 15-min meditation scan were Plates were incubated for protein capture, unbound material was washed
used to compare equal-length scans. P-values were corrected for multiple away, and biotinylated capture antibodies were added to hybridize with
comparisons using Bonferroni–Holm family-wise error correction and SOMAmer-bound proteins. Streptavidin-conjugated Cy3 dye was added
Benjamini–Hochberg false discovery rate when no results reached to label the antibodies, plates were washed to remove unbound dye, and
pFWE < 0.05. protein-bound SOMAmer reagents were eluted and hybridized to cus-
tom DNA microarrays containing complementary sequences to each
Anatomical data. Preprocessed skull-stripped segmented T1w data was SOMAmer. Cy3 fluorescence intensity in relative fluorescence units
resampled to the MNI fsaverage template using mris_preproc (Free- (RFU) measured protein abundance. Hybridization control normal-
Surfer 7.3.2) and cortical surface was inflated to an average spherical ization removed sample variance between microarrays and scanners;
surface. Hemispheres were automatically parcellated into regions of median signal normalization removed within-plate inter-sample differ-
interest, and vertex-wise volumetric and cortical thickness calculations ences; and calibration normalization removed variance across assay runs.
were performed. Images were spatially smoothed with a 5 mm FWHM Performance and quality were monitored with blank wells, technical
Gaussian kernel. Pre-vs-post differences were assessed by repeated replicates, and spiked protein controls, and precision was assessed by
measures ANOVA, and group differences between novice and advanced calculating coefficients of variation for all measurements. Background
participants at each timepoint were assessed by fitting a general linear subtraction RFU calculations were performed with SomaLogic Discovery
model using mri_glmfit-sim. Total intracranial volume and age were Server software.
included as covariates, and results were cluster corrected with a cluster-
wise p-threshold = 0.01. Data processing. RFU values were normalized, log10-transformed, and
auto-scaled (mean-centered and divided by each variable’s standard
Human plasma collection deviation), and outliers above two SDs were flagged. Foldchange and
Human plasma was collected and processed, as described in ref. 84, via paired t-tests tested pre/post-intervention differences, and hierarchical
venous puncture by registered nurses, physicians, and phlebotomists, at clustering and k-means clustering revealed correlated variation expres-
pre-intervention either off-site (“Day -1”, 2 days pre-retreat, 2–4 pm) or sion patterns. Functional enrichment analysis (Enrichr-KG) on differ-
on-site (“Day 0”, 1 day pre-retreat, 11 am–4 pm) and post-intervention entially expressed proteins (pre/post p < 0.01), with all SOMAmer targets
(“Day 8”, on-site, 9–5 pm). Participants were offered the same food as background, assessed enriched biological processes, molecular path-
options for all breakfast, lunch, and snacks throughout the week, ways, and cellular components. Enriched pathways were considered
although food choices were not monitored. All participants were significant at pFDR < 0.05. Protein-protein interaction subnetworks of
required to fast for at least 30 min prior to blood collection. Blood was significant proteins (p < 0.05 or |fold change| ≥ 0.5) were created in
collected in EDTA-coated tubes (BD, Franklin Lakes, NJ) and kept at Cytoscape with ≤30 interactor nodes and heat diffusion using the
4 °C (wet ice) for less than 30 min. Plasma was isolated by centrifuga- STRING database85.
tion at 3000 RPM for 15 min in an E8 Touch tabletop centrifuge (LW
Scientific, Lawrenceville, GA), aliquoted into 1.5 mL Eppendorf tubes, Plasma metabolomics
and immediately frozen on dry ice. At the end of the retreat, samples 640 metabolites in whole plasma were detected via liquid chromatography-
were shipped to UCSD and stored at −80 °C. mass spectrometry analysis using the widely-targeted Metware platform
(Metware Biotechnology, Woburn, MA, USA).
Plasma nanoparticle tracking analysis
Plasma Nanoparticle Tracking Analysis (NTA) measured circulating Metabolite detection. Plasma samples were extracted with a 1:4
plasma nanoparticle size and concentration with a NanoSight NS300 ACN:methanol solution and centrifuged at 12,000 RPM (10 min, 4 °C)
instrument (Malvern Instruments Ltd., UK). Samples were thawed on ice and again at 12,000 RPM (3 min, 4 °C) after 30 min. at −20 °C. Super-
and diluted in PBS (100×) to prevent aggregation. Measurements were natants were analyzed by ultra-HPLC (UPLC) (ExionLC 2.0, Sciex,
performed at room temperature under continuous video recording Framingham, MA, USA) on a UPLC column with a gradient elution
(16 × 30-s. acquisitions/sample) and 532 nm laser illumination. Mean system for metabolite separation (UPLC conditions listed in Supple-
hydrodynamic diameter, size distribution, and particle concentration were mentary Table 13). Eluted metabolites were detected based on mass and
obtained per sample with NTA 3.3 software. fragmentation patterns by Tandem Mass Spectrometry (MS/MS) using a
quadrupole-time of flight mass spectrometer (QTRAP®6500+, Sciex)
ELISA with an electrospray ionization (ESI) Turbo Ion-Spray interface, oper-
To investigate plasma concentrations of oxytocin, beta-endorphin, dynor- ating in positive and negative ion mode. Untargeted qualitative meta-
phin, anandamide, cocaine, and amphetamine-regulated transcript bolite identification was performed by matching ion features to
(CART), c-reactive protein (CRP), and neuropeptide Y (NPY), pre- and references from Metware, HMDB86, METLIN87, and KEGG88 databases.
post-intervention plasma samples were tested with commercially available Identified metabolites were then quantified using triple quadrupole mass

Communications Biology | (2025)8:1525 14

https://doi.org/10.1038/s42003-025-09088-3 Article

spectrometry with multiple reaction monitoring (mass spectrum con- Neurite differentiation
ditions in Supplementary Table 14). Mass spectrometry chromato- To investigate the effects of human plasma on neurite growth, PC12 neu-
graphic peaks were corrected with MultiQuant software (Sciex). roendocrine cells were differentiated to their neuronal phenotype, treated
Reproducibility, cross-contamination, and inter-sample quality were with plasma, and live-imaged following published methods92.
controlled with control samples, internal standard peaks from blank
samples, and internal standards, respectively. Cell culture and differentiation. PC12 cells (ATCC, Manassas, VA,
USA) were cultured in RPMI 1640 medium supplemented with 10%
Metabolite analysis. Quantified metabolite features were variance-filtered horse serum, 5% FBS, and 1% penicillin/streptomycin under standard
via inter-quartile range (max 25% filtered out), normalized, and log10- conditions (37 °C, 5% CO2). On Day 0, differentiation was induced by
transformed. Outliers above 2 standard deviations were flagged. Dietary plating cells onto poly-D-lysine-coated plates at a 1.0 × 104 cells/cm2
factors and fasting may have introduced confounds in proteomic and density and culturing them with Opti-MEM medium supplemented with
metabolomic measures, as no standardized diet was implemented and 0.5% FBS, 1% penicillin/streptomycin, 50 ng/mL nerve growth factor
fasting times beyond 30 min pre-blood collection were not controlled. (NGF), and either 1% human plasma pooled from pre- and post-
Exogenous compounds annotated as dietary or drug-related by HMDB were intervention or no plasma (control cells) with 2 technical replicate wells
monitored using the MetaboAnalyst pipeline and were flagged or removed if per condition. Medium was replaced every 48 h. Media was supple-
observed among the top 25 features. PCA, hierarchical cluster analysis mented with 1% Culture One from Day 2 onwards to support neuronal
(HCA), partial least squares discriminant analysis (PLS-DA), and FDR- differentiation.
corrected pre/post paired t-tests (Scikit-learn89) identified differentially
expressed metabolites. Correlation heatmaps were generated using Pearson r Live-cell imaging and neurite analysis. Differentiated cells were live-
distance measures. Metabolic pathway analysis comparing pre- to post- imaged every 24 h for 10 days at 20× in a live cell imaging microscope
samples was performed with MetaboAnalyst v5.0 based on the KEGG (Keyence BZ-X700) equipped with a phase contrast objective and an
database. Pathway enrichment was first assessed using MetaboAnalyst, and incubation chamber at 37 °C, 5% CO2, and controlled humidity. Images
false discovery rate (FDR) correction was applied to control for multiple were acquired under consistent illumination across groups and time
comparisons. Significant pathways (pFDR < 0.05) were then selected for points, and background subtraction and image thresholding were applied
follow-up analysis. For these pathways, we examined the distribution of to enhance neurite visualization. The single longest neurite per cell, for
normalized feature values and report the corresponding p-values for group cells for which it was longer than the cell body diameter, was manually
comparisons after normalization. traced in FIJI software (v1.0). Neurite lengths were compared with two-
tailed independent samples t-tests between plasma-treated and control
Real time cellular metabolic analysis groups at each time point. Treatment wells remained blinded until
To investigate the effects of human plasma on cellular metabolism, we completion to avoid bias.
performed Seahorse XF mitochondrial stress, glycolytic rate, and ATP
production assays on BE(2)M17 cells (ATCC), a human neuroblastoma cell Exosome-specific small RNA transcriptomics
line, exposed to plasma followings90,91 methods. Exosome isolation. Exosomes were isolated from plasma samples
(300–500 µl) by centrifugation at 300 × g for 10 min at 4 °C to remove cell
Cell culture and plasma treatment. BE(2)M17 cells were cultured in debris. Supernatants were transferred to microcentrifuge tubes, filled
complete DMEM/F12 medium supplemented with 10% FBS and 1% with 1× PBS, balanced for mass, placed in a pre-cooled Beckman Coulter
penicillin/streptomycin under standard conditions (37 °C, 5% CO2). Type 70.1 rotor, and centrifuged at 10,000 × g for 30 min at 4 °C with
Cells were harvested, washed, and seeded onto microplates pre-coated maximum acceleration and gradual deceleration. Supernatants were
with poly-L-lysine at a 40,000 cells/well density and incubated overnight. transferred to new tubes, exosome pellets were resuspended in 100 µl 1×
Cells were resuspended in XF assay medium and treated with 1% plasma PBS, and another round of ultracentrifugation at 100,000 × g for 2 h. at
for 1 h at 37 °C, with 4 technical replicate wells per plasma sample, and 4 °C was performed. Final exosome pellets were resuspended in 150 µl 1×
stained with DAPI for post-assay cell counting. PBS, analyzed by nanoparticle tracking using the same protocol as whole
plasma, and stored at −80 °C.
Seahorse XF assays. Seahorse XF96 Analyzer (Agilent, Santa Clara,
CA, USA) was calibrated according to manufacturer’s instructions, and Exosome RNA sequencing. Total RNAs were purified from exosome
the following injection protocols were implemented: Mitochondrial pellets using Direct Zol RNA mini kit (Zymo Research). RNAs were
Stress Test (kit #103015-100): Oligomycin (1.5 µM), FCCP (2.0 µM), and eluted in 20 µl of RNAse-free water and concentrated to 5 µl using
Rotenone/Antimycin A (0.5 µM) were sequentially injected to measure SpeedVac Vacuum Concentrator (Thermo Fisher Scientific). 44 to
basal respiration, proton leak, maximal respiration, and spare respiratory 523 ng of exosome-specific extracellular small RNAs (exRNAs) were
capacity, with oxygen consumption rate (OCR) monitoring. Glycolytic obtained from 28 human plasma samples. exRNA quantity and integrity
Rate Test (kit #103344-100): Rotenone/antimycin A (0.5 µM) and were determined using the NanoDrop ND-1000 spectrophotometer
2-deoxyglucose (50 mM) were injected to assess basal glycolytic rate, (Thermo Fisher Scientific) and the Bioanalyzer 2100 (Agilent, Santa
glycolytic reserve, and non-glycolytic acidification, monitored by extra- Clara, CA, USA), respectively. Small exRNA libraries were generated
cellular acidification rate (ECAR). ATP Production Rate Test (kit using NEXTFlex Small RNA-seq Kit v4 with UDIs (Revvity Inc, Wal-
#103592-100): Injections of oligomycin (1.5 µM) and rotenone/anti- tham, MA, USA) according to modified manufacturer’s protocol to
mycin A (0.5 µM) enabled real-time calculation of ATP production rates account for low exRNA concentrations in exosome samples. The strategy
from OCR and ECAR. for small RNA libraries aimed at a range of 16–120 nucleotide (nt)-long
transcripts for accurate identification and quantification of exosome-
Data analysis and statistics. We used Seahorse XF software (v.2.6) to specific extracellular microRNAs, ncRNAs, and protein-coding RNAs.
analyze raw OCR and ECAR data, normalizing results by cell counts, NEXTFLEX® 3′ Adenylated Adapter v4 was ligated at 25 °C for 1 h.
and compared metabolic parameters between control, pre-, and post- Following 3′ Adenylated Adapter inactivation, NEXTFLEX 5′ Adapter v4
intervention groups with paired t-tests (p < 0.05 considered sig- was ligated at 20 °C for 1 h. Reverse Transcription-First Strand Synthesis
nificant). We assessed cell viability pre- and post-treatment and from 5′ and 3′ NEXTFLEX® Adapter Ligated RNA templates was con-
included four non-cellular blank well controls per plate to ensure assay ducted at 50 °C for 1 h, followed by inactivation at 90 °C for 5 min. First
quality. Strand Synthesis products were purified using NEXTFLEX Cleanup

Communications Biology | (2025)8:1525 15

https://doi.org/10.1038/s42003-025-09088-3 Article

Beads by two washes in 80% ethanol. RNA templates were combined with homoskedasticity were ascertained, in which case two-tailed paired t-tests
NEXTFLEX UDI Barcoded Primer Mixes v4 diluted 1:4 and amplified by were employed. Given that proteomic and metabolomic analyses were
PCR for 24 cycles. PCR products size selection and cleanup were done hypothesis-free and data driven, False Discovery Rate (FDR) correction was
using NEXTFLEX Cleanup Beads. Small exRNA libraries were eluted applied using the Benjamini–Hochberg procedure, and reported p-values
from beads in 15 µl and concentrations were determined using Qubit are labelled as FDR-adjusted, with nominal p-values included for trans-
dsDNA HS Assay (Life Technologies). DNA fragment profile in small parency. Seahorse XF plates were prepared with 4 technical replicates (wells)
exRNA libraries was analyzed on the Bioanalyzer 2100. RNA-seq was per participant and time point. PC12 assay plates were prepared with pooled
conducted on the NovaSeq × Plus Short-Read Sequencer (Illumina, San plasma (n = 20) with 2 technical replicates (wells) per treatment (pre-
Diego, CA) at the Genomics Research and Technology Hub, University plasma/post-plasma/no treatment).
of California, Irvine.
Ethics and inclusion statement
RNA-seq data processing. Sequencing read data were converted Data collection and analysis was performed locally in San Diego, CA, USA
to FASTQ format for bioinformatic processing using and included local researchers. Roles and responsibilities were agreed
Lonestar6 supercomputer at the Texas Advanced Computing Center and amongst collaborators ahead of the research, and clinical practice and
bioinformatically processed according to guidelines developed for biospecimen handling trainings were provided to all researchers. The study
NEXTFLEX Small RNA-Seq (Revvity Inc., Waltham, MA, USA). In brief, was approved by a local research ethics committee and did not result in
flanking forward (TGG AAT TCT CGG GTG CCA AGG) and reverse personal risk to study participants. Researchers who handled biospecimens
(AGA TCG GAA GAG CGT CGT GTA GGG AAA GA) adapter during collection and analysis wore adequate personal protective equip-
sequences were trimmed prior to alignment with Cutadapt93. Paired reads ment. fMRI researchers and participants were thoroughly screened for MRI
above 16 nucleotides with quality scores above 20 were mapped to the safety before being allowed inside the fMRI magnet room.
human genome primary assembly (Release 47, GRCh38.p14). Reads-to-
genome alignment was conducted using STAR aligner94. Mapping quality Data availability
was assessed using MultiQC tool95. Values for all data points in plots and reported means are available (Sup-
plementary Data 1). RNA-seq data is available from GEO repository
Detection of differential exosomal RNAs. The differential analysis of (accession #GSE291700). Proteomics data is available from PRIDE repo-
sequence read count data was done using a generalization of a paired t- sitory (dataset PAD000010). Any other data is available from the corre-
test using EdgeR software96–98 in which exosome RNA pools in pre- and sponding author on reasonable request.
post-intervention samples per participant were compared separately and
the baseline differences between participants were subtracted. Tran- Code availability
scripts with fewer than 10 counts were removed from analysis. Trimmed Analytic code can be obtained from the corresponding author on reasonable
Mean of M-values (TMM) normalization99,100 was applied to account for request.
compositional difference between samples. exRNA were annotated using
org.Hs.eg.db101 and miRbase102 databases. Sequences corresponding to Received: 1 May 2025; Accepted: 17 October 2025;
non-coding RNAs (ncRNAs) and exhibiting the highest variance across
all samples were analyzed by PCA of TMM-normalized counts using
PCAtools Bioconductor package. Functional analysis of differential References
exRNA was done using ClusterProfiler103 and the Reactome Pathways 1. Ashar, Y. K. et al. Effect of pain reprocessing therapy vs placebo and
Knowledgebase46. A paired design in the negative binomial (NB) gen- usual care for patients with chronic back pain: a randomized clinical
eralized linear model (GLM) detected the effect of the intervention on the trial. JAMA Psychiatry 79, 13–23 (2022).
composition of exRNAs from paired post-intervention (n = 14) vs pre- 2. Benedetti, F., Carlino, E. & Pollo, A. How placebos change the
intervention (n = 14) samples, while adjusting for differences between patient’s brain. Neuropsychopharmacology 36, 339–354 (2011).
participants. Participants’ experience level (9 advanced, 5 novices), 3. Benedetti, F., Piedimonte, A. & Frisaldi, E. How do placebos work?.
gender (8 females, 6 males), and age were included as additional Eur. J. Psychotraumatol. 9, 1533370 (2018).
covariates. 4. Jonas, W. B. et al. To what extent are surgery and invasive
procedures effective beyond a placebo response? A systematic
Machine learning and MEQ-features correlation analyses review with meta-analysis of randomised, sham controlled trials.
We applied machine learning to identify the most biologically relevant BMJ Open 5, e009655 (2015).
features across time point (pre/post) and experience level (novice/advanced) 5. Probst, P. et al. Placebo-controlled trials in surgery: a systematic
datasets. Each dataset was preprocessed (log-transformed and auto-cen- review and meta-analysis. Med. (Baltim.) 95, e3516 (2016).
tered) and missing data was imputed to ensure feature scaling, normal- 6. Beecher, H. K. Surgery as placebo: a quantitative study of bias.
ization consistency, and data integrity and comparability across JAMA 176, 1102–1107 (1961).
modalities104. Post-preprocessing, ELISA, metabolomics, transcriptomics, 7. Hoenemeyer, T. W., Kaptchuk, T. J., Mehta, T. S. & Fontaine, K. R.
and proteomics datasets were concatenated into a single feature matrix used Open-label placebo treatment for cancer-related fatigue: a
as input for an eXtreme Gradient Boosting (XGBoost)105 classifier chosen for randomized-controlled clinical trial. Sci. Rep. 8, 2784 (2018).
its ability to handle high-dimensional data. To obtain a unified impact 8. Carvalho, C. et al. Open-label placebo treatment in chronic low back
metric per feature, the dataset was passed through a Shapley Additive pain: a randomized controlled trial. Pain 157, 2766–2772 (2016).
exPlanations (SHAP)106 explainer model. Spearman correlations were then 9. Schaefer, M., Harke, R. & Denke, C. Open-label placebos improve
calculated between MEQ scores and the top 14 features per model and symptoms in allergic rhinitis: a randomized controlled trial.
corrected for multiple comparisons to explore whether the top molecular Psychother. Psychosom. 85, 373–374 (2016).
changes were linked to subjective outcomes. 10. Zanesco, A. P., King, B. G., Conklin, Q. A. & Saron, C. D. The
occurrence of psychologically profound, meaningful, and mystical
Statistics and reproducibility experiences during a month-long meditation retreat. Mindfulness
Given the n = 20 sample size (n = 19 for fMRI), nonparametric Wilcoxon 14, 606–621 (2023).
signed rank statistics were performed throughout when comparing paired 11. Kuyken, W. et al. Efficacy of mindfulness-based cognitive therapy in
samples, except when n ≥ 30 (e.g., protein targets) or when normality and prevention of depressive relapse: an individual patient data meta-

Communications Biology | (2025)8:1525 16

https://doi.org/10.1038/s42003-025-09088-3 Article

analysis from randomized trials. JAMA Psychiatry 73, 565–574 33. Barrett, L. F. & Simmons, W. K. Interoceptive predictions in the brain.
(2016). Nat. Rev. Neurosci. 16, 419–429 (2015).
12. Goyal, M. et al. Meditation programs for psychological stress and 34. Hove, M. J. et al. Brain network reconfiguration and perceptual
well-being: a systematic review and meta-analysis. JAMA Intern. decoupling during an absorptive state of consciousness. Cereb.
Med. 174, 357–368 (2014). Cortex 26, 3116–3124 (2016).
13. Fox, K. C. R. et al. Functional neuroanatomy of meditation: a review 35. Makris, N. et al. Decreased volume of left and total anterior insular
and meta-analysis of 78 functional neuroimaging investigations. lobule in schizophrenia. Schizophr. Res. 83, 155–171 (2006).
Neurosci. Biobehav. Rev. 65, 208–228 (2016). 36. Frazier, J. A. et al. Structural brain magnetic resonance imaging of
14. Afonso, R. F., Kraft, I., Aratanha, M. A. & Kozasa, E. H. Neural limbic and thalamic volumes in pediatric bipolar disorder. Am. J.
correlates of meditation: a review of structural and functional MRI Psychiatry 162, 1256–1265 (2005).
studies. Front. Biosci. Sch. 12, 92–115 (2020). 37. Desikan, R. S. et al. An automated labeling system for subdividing
15. Lomas, T., Ivtzan, I. & Fu, C. H. Y. A systematic review of the the human cerebral cortex on MRI scans into gyral based regions of
neurophysiology of mindfulness on EEG oscillations. Neurosci. interest. NeuroImage 31, 968–980 (2006).
Biobehav. Rev. 57, 401–410 (2015). 38. Power, J. D. et al. Functional network organization of the human
16. Oyler, D. L. et al. The influence of meditative interventions on brain. Neuron 72, 665–678 (2011).
immune functioning: a meta-analysis. Mindfulness 14, 1815–1851 39. Grant, J. A. et al. Cortical thickness, mental absorption and
(2023). meditative practice: possible implications for disorders of attention.
17. Davidson, J. M. The physiology of meditation and mystical states of Biol. Psychol. 92, 275–281 (2013).
consciousness. Perspect. Biol. Med. 19, 345–380 (1976). 40. Aruga, J. & Mikoshiba, K. Identification and characterization of Slitrk,
18. Tang, Y.-Y. et al. Central and autonomic nervous system interaction a novel neuronal transmembrane protein family controlling neurite
is altered by short-term meditation. Proc. Natl. Acad. Sci. USA 106, outgrowth. Mol. Cell. Neurosci. 24, 117–129 (2003).
8865–8870 (2009). 41. Beaubien, F., Raja, R., Kennedy, T. E., Fournier, A. E. & Cloutier, J.-F.
19. Álvarez-López, M. J. et al. Changes in the expression of Slitrk1 is localized to excitatory synapses and promotes their
inflammatory and epigenetic-modulatory genes after an intensive development. Sci. Rep. 6, 27343 (2016).
meditation retreat. Compr. Psychoneuroendocrinol. 11, 100152 42. Jevning, R., Wallace, R. K. & Beidebach, M. The physiology of
(2022). meditation: a review. A wakeful hypometabolic integrated response.
20. Buric, I., Farias, M., Jong, J., Mee, C. & Brazil, I. A. What is the Neurosci. Biobehav. Rev. 16, 415–424 (1992).
molecular signature of mind–body interventions? A systematic 43. Cremers, N. A. J. et al. S100A8/A9 increases the mobilization of pro-
review of gene expression changes induced by meditation and inflammatory Ly6Chigh monocytes to the synovium during
related practices. Front. Immunol. 8, 670 (2017). experimental osteoarthritis. Arthritis Res. Ther. 19, 217 (2017).
21. Venditti, S. et al. Molecules of silence: effects of meditation on gene 44. Crowe, L. A. N. et al. S100A8 & S100A9: alarmin mediated
expression and epigenetics. Front. Psychol. 11, 1767 (2020). inflammation in tendinopathy. Sci. Rep. 9, 1463 (2019).
22. Cahn, B. R., Goodman, M. S., Peterson, C. T., Maturi, R. & Mills, P. J. 45. Pascoe, M. C., Thompson, D. R., Jenkins, Z. M. & Ski, C. F.
Yoga, meditation and mind-body health: increased BDNF, cortisol Mindfulness mediates the physiological markers of stress:
awakening response, and altered inflammatory marker expression systematic review and meta-analysis. J. Psychiatr. Res. 95, 156–178
after a 3-month yoga and meditation retreat. Front. Hum. Neurosci. (2017).
11, 315 (2017). 46. Milacic, M. et al. The reactome pathway knowledgebase 2024.
23. Morin-Alain, V. et al. Effects of a mindfulness-based intervention on Nucleic Acids Res 52, D672–D678 (2024).
circulating cytokine levels in individuals with amnestic mild cognitive 47. Bremer, B. et al. Mindfulness meditation increases default mode,
impairment: a pilot study. OBM Integr. Complement. Med. 5, 1–24 salience, and central executive network connectivity. Sci. Rep. 12,
(2020). 13219 (2022).
24. Xue, T. et al. The heart-brain axis: a proteomics study of meditation 48. Vishnubhotla, R. V. et al. Advanced meditation alters resting-state
on the cardiovascular system of Tibetan Monks. eBioMedicine 80, brain network connectivity correlating with improved mindfulness.
104026 (2022). Front. Psychol. 12, 745344 (2021).
25. Chen, W. et al. Comprehensive metabolomic and lipidomic analysis 49. Datko, M. et al. Increased insula response to interoceptive attention
reveals metabolic changes after mindfulness training. Mindfulness following mindfulness training is associated with increased body
11, 1390–1400 (2020). trusting among patients with depression. Psychiatry Res.
26. Raman, M. et al. Isha yoga practices, vegan diet, and participation in Neuroimaging 327, 111559 (2022).
Samyama meditation retreat: impact on the gut microbiome & 50. Laukkonen, R. E. & Slagter, H. A. From many to (n)one: meditation
metabolome – a non-randomized trial. BMC Complement. Med. and the plasticity of the predictive mind. Neurosci. Biobehav. Rev.
Ther. 23, 107 (2023). 128, 199–217 (2021).
27. Greenwell, B. Energies of Transformation: A Guide to the Kundalini 51. Van Overwalle, F. & Mariën, P. Functional connectivity between the
Process (Motilal Banarsidass Publishers, 2002). cerebrum and cerebellum in social cognition: a multi-study analysis.
28. Krishna, G. Kundalini: The Evolutionary Energy in Man (Shambhala NeuroImage 124, 248–255 (2016).
Publications, 2018). 52. Guidotti, R. et al. Long-term and meditation-specific modulations of
29. Hopkins, J. Meditation on Emptiness (Simon and Schuster, 2014). brain connectivity revealed through multivariate pattern analysis.
30. Kaptchuk, T. J. Open-label placebo: reflections on a research Brain Topogr. 36, 409–418 (2023).
agenda. Perspect. Biol. Med. 61, 311–334 (2018). 53. Garrison, K. A., Zeffiro, T. A., Scheinost, D., Constable, R. T. &
31. Prakash, R. S. et al. Mindfulness meditation and network Brewer, J. A. Meditation leads to reduced default mode network
neuroscience: review, synthesis, and future directions. Biol. activity beyond an active task. Cogn. Affect. Behav. Neurosci. 15,
Psychiatry Cogn. Neurosci. Neuroimaging https://doi.org/10.1016/j. 712–720 (2015).
bpsc.2024.11.005 (2024). 54. Guo, W. et al. Abnormal resting-state cerebellar-cerebral functional
32. Brewer, J. A. et al. Meditation experience is associated with connectivity in treatment-resistant depression and treatment
differences in default mode network activity and connectivity. Proc. sensitive depression. Prog. Neuropsychopharmacol. Biol.
Natl. Acad. Sci. USA 108, 20254–20259 (2011). Psychiatry 44, 51–57 (2013).

Communications Biology | (2025)8:1525 17

https://doi.org/10.1038/s42003-025-09088-3 Article

55. Shen, Y.-Q., Zhou, H.-X., Chen, X., Castellanos, F. X. & Yan, C.-G. 75. Pruim, R. H. R. et al. ICA-AROMA: A robust ICA-based strategy for
Meditation effect in changing functional integrations across large- removing motion artifacts from fMRI data. NeuroImage 112,
scale brain networks: preliminary evidence from a meta-analysis of 267–277 (2015).
seed-based functional connectivity. J. Pac. Rim Psychol. 14, e10 76. Abraham, A. et al. Machine learning for neuroimaging with scikit-
(2020). learn. Front. Neuroinform. 8, 14 (2014).
56. Friston, K. The free-energy principle: a unified brain theory?. Nat. 77. Rubinov, M., Kötter, R., Hagmann, P. & Sporns, O. Brain connectivity
Rev. Neurosci. 11, 127–138 (2010). toolbox: a collection of complex network measurements and brain
57. Seth, A. K. Interoceptive inference, emotion, and the embodied self. connectivity datasets. NeuroImage 47, S169 (2009).
Trends Cogn. Sci. 17, 565–573 (2013). 78. Raichle, M. E. The restless brain. Brain Connect 1, 3–12 (2011).
58. Sezer, I. & Sacchet, M. D. Advanced and long-term meditation and 79. De Filippi, E. et al. Meditation-induced effects on whole-brain
the autonomic nervous system: a review and synthesis. Neurosci. structural and effective connectivity. Brain Struct. Funct. 227,
Biobehav. Rev. 173, 106141 (2025). 2087–2102 (2022).
59. van Lutterveld, R. et al. Meditation is associated with increased brain 80. Zhang, Z. et al. Longitudinal effects of meditation on brain resting-
network integration. NeuroImage 158, 18–25 (2017). state functional connectivity. Sci. Rep. 11, 11361 (2021).
60. Calder, A. E. & Hasler, G. Towards an understanding of psychedelic- 81. Cristofori, I. et al. Neural correlates of mystical experience.
induced neuroplasticity. Neuropsychopharmacology 48, 104–112 Neuropsychologia 80, 212–220 (2016).
(2023). 82. Josipovic, Z. Neural correlates of nondual awareness in meditation.
61. Smigielski, L., Scheidegger, M., Kometer, M. & Vollenweider, F. X. Ann. N. Y. Acad. Sci. 1307, 9–18 (2014).
Psilocybin-assisted mindfulness training modulates self- 83. Dadi, K. et al. Fine-grain atlases of functional modes for fMRI
consciousness and brain default mode network connectivity with analysis. NeuroImage 221, 117126 (2020).
lasting effects. NeuroImage 196, 207–215 (2019). 84. Zuniga-Hertz, J. P. et al. Multidimensional analysis of twin sets
62. Tilburt, J. C., Emanuel, E. J., Kaptchuk, T. J., Curlin, F. A. & Miller, F. during an intensive week-long meditation retreat: a pilot study.
G. Prescribing “placebo treatments”: results of national survey of US Mindfulness 16, 1634–1655 (2025).
internists and rheumatologists. BMJ 337, a1938 (2008). 85. Szklarczyk, D. et al. STRING v11: protein–protein association
63. Saatcioglu, F. Regulation of gene expression by yoga, meditation networks with increased coverage, supporting functional discovery
and related practices: a review of recent studies. Asian J. Psychiatry in genome-wide experimental datasets. Nucleic Acids Res. 47,
6, 74–77 (2013). D607–D613 (2019).
64. Vadaq, N. et al. Targeted plasma proteomics reveals upregulation of 86. Wishart, D. S. et al. HMDB: the human metabolome database.
distinct inflammatory pathways in people living with HIV. iScience Nucleic Acids Res. 35, D521–D526 (2007).
25, 105089 (2022). 87. Smith, C. A. et al. METLIN: a metabolite mass spectral database.
65. Laumann, T. O., Snyder, A. Z. & Gratton, C. Challenges in the Ther. Drug Monit. 27, 747 (2005).
measurement and interpretation of dynamic functional connectivity. 88. Kanehisa, M., Furumichi, M., Sato, Y., Kawashima, M. & Ishiguro-
Imaging Neurosci. 2, 1–19 (2024). Watanabe, M. KEGG for taxonomy-based analysis of pathways and
66. Garrison, K. et al. Effortless awareness: using real time genomes. Nucleic Acids Res. 51, D587–D592 (2023).
neurofeedback to investigate correlates of posterior cingulate cortex 89. Pedregosa, F. et al. Scikit-learn: machine learning in Python. J.
activity in meditators’ self-report. Front. Hum. Neurosci. 7, 440 Mach. Learn. Res. 12, 2825–2830 (2011).
(2013). 90. Song, Y. et al. Impact of blood factors on endothelial cell metabolism
67. Tagliazucchi, E. & Laufs, H. Decoding wakefulness levels from and function in two diverse heart failure models. PLoS ONE 18,
typical fMRI resting-state data reveals reliable drifts between e0281550 (2023).
wakefulness and sleep. Neuron 82, 695–708 (2014). 91. Kopp, E. L. et al. Modeling and phenotyping acute and chronic type 2
68. Maclean, K. A., Leoutsakos, J.-M. S., Johnson, M. W. & Griffiths, R. diabetes mellitus in vitro in rodent heart and skeletal muscle cells.
R. Factor analysis of the mystical experience questionnaire: a study Cells 12, 2786 (2023).
of experiences occasioned by the hallucinogen psilocybin. J. Sci. 92. Karliner, J. & Merry, D. E. Differentiating PC12 cells to evaluate neurite
Study Relig. 51, 721–737 (2012). densities through live-cell imaging. STAR Protoc. 4, 101993 (2023).
69. Esteban, O. et al. fMRIPrep: a robust preprocessing pipeline for 93. Martin, M. Cutadapt removes adapter sequences from high-
functional MRI. Nat. Methods 16, 111–116 (2019). throughput sequencing reads. EMBnet. J. 17, 10–12 (2011).
70. Andersson, J. L. R., Skare, S. & Ashburner, J. How to correct 94. STAR: ultrafast universal RNA-seq aligner | Bioinformatics | Oxford
susceptibility distortions in spin-echo echo-planar images: Academic. https://academic.oup.com/bioinformatics/article/29/1/
application to diffusion tensor imaging. NeuroImage 20, 870–888 15/272537.
(2003). 95. Ewels, P., Magnusson, M., Lundin, S. & Käller, M. MultiQC:
71. Avants, B. B., Epstein, C. L., Grossman, M. & Gee, J. C. Symmetric summarize analysis results for multiple tools and samples in a single
diffeomorphic image registration with cross-correlation: evaluating report. Bioinformatics 32, 3047–3048 (2016).
automated labeling of elderly and neurodegenerative brain. Med. 96. Tuch, B. B. et al. Tumor transcriptome sequencing reveals allelic
Image Anal. 12, 26–41 (2008). expression imbalances associated with copy number alterations.
72. Zhang, Y., Brady, M. & Smith, S. Segmentation of brain MR images PLoS ONE 5, e9317 (2010).
through a hidden Markov random field model and the expectation- 97. Robinson, M. D., McCarthy, D. J. & Smyth, G. K. edgeR: a
maximization algorithm. IEEE Trans. Med. Imaging 20, 45–57 Bioconductor package for differential expression analysis of digital
(2001). gene expression data. Bioinformatics 26, 139–140 (2010).
73. Dale, A. M., Fischl, B. & Sereno, M. I. Cortical surface-based 98. McCarthy, D. J., Chen, Y. & Smyth, G. K. Differential expression
analysis: I. Segmentation and surface reconstruction. NeuroImage analysis of multifactor RNA-Seq experiments with respect to
9, 179–194 (1999). biological variation. Nucleic Acids Res. 40, 4288–4297 (2012).
74. Jenkinson, M. & Smith, S. A global optimisation method for robust 99. Robinson, M. D. & Oshlack, A. A scaling normalization method for
affine registration of brain images. Med. Image Anal. 5, 143–156 differential expression analysis of RNA-seq data. Genome Biol. 11,
(2001). R25 (2010).

Communications Biology | (2025)8:1525 18

https://doi.org/10.1038/s42003-025-09088-3 Article

100. Risso, D., Ngai, J., Speed, T. P. & Dudoit, S. Normalization of RNA- and manuscript edits. M.A.P. contributed to study design, execution, and
seq data using factor analysis of control genes or samples. Nat. manuscript edits. T.M.B. contributed to study design and execution. H.H.P.
Biotechnol. 32, 896–902 (2014). contributed to study design, execution, manuscript edits, and funding.
101. Carlson, M., Falcon, S., Pages, H. & Li, N. org. Hs. eg. db: genome
wide annotation for human. R. Package Version 3, 3 (2019). Competing interests
102. Kozomara, A., Birgaoanu, M. & Griffiths-Jones, S. miRBase: from The authors declare the following competing interest: J.D. is employed by
microRNA sequences to function. Nucleic Acids Res. 47, Encephalon, Inc., which offers the meditation retreats. All other authors
D155–D162 (2019). declare no competing interests.
103. Wu, T. et al. clusterProfiler 4.0: a universal enrichment tool for
interpreting omics data. Innov. Camb. Mass 2, 100141 (2021). Additional information
104. Song, M. et al. A review of integrative imputation for multi-omics Supplementary information The online version contains
datasets. Front. Genet. 11, 570255 (2020). supplementary material available at
105. Chen, T. & Guestrin, C. XGBoost: a scalable tree boosting system. In https://doi.org/10.1038/s42003-025-09088-3.
Proceedings of the 22nd ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining 785–794 https://doi.org/10. Correspondence and requests for materials should be addressed to
1145/2939672.2939785 (2016). Hemal H. Patel.
106. Lundberg, S. M. & Lee, S. A unified approach to interpreting model
predictions. In Proceedings of the 31st International Conference on Peer review information Communications Biology thanks Vaibhav Tripathi
Neural Information Processing Systems (NIPS'17) Vol. 31, and Suruchi Fialoke for their contribution to the peer review of this work.
4768–4777 (2017). Primary Handling Editor: Jasmine Pan. A peer review file is available.

Acknowledgements Reprints and permissions information is available at

Support for this research was provided by the InnerScience Research Fund http://www.nature.com/reprints
(H.H.P.) and a Research Career Scientist Award from the Veterans
Administration (BX005229 to H.H.P.). Publisher’s note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Author contributions
A.J.D. contributed to FMRI design, data collection, and analysis, PC12 Open Access This article is licensed under a Creative Commons
assay, ELISA assays, Seahorse assays, NanoSight measurements, Attribution 4.0 International License, which permits use, sharing,
exosome isolation, proteomics and metabolomics analyses, figure creation, adaptation, distribution and reproduction in any medium or format, as long
and manuscript writing. S.S. contributed to manuscript writing, figure as you give appropriate credit to the original author(s) and the source,
creation, and metabolomics and machine learning analyses. J.P.Z.H. provide a link to the Creative Commons licence, and indicate if changes
contributed to Seahorse and NanoSight analyses. R.C. contributed to were made. The images or other third party material in this article are
sample collection and processing and manuscript edits. J.M.S. contributed included in the article’s Creative Commons licence, unless indicated
to sample collection and processing. J.A.B. contributed to data collection otherwise in a credit line to the material. If material is not included in the
and manuscript edits. L.C. contributed to FMRI design, data collection, and article’s Creative Commons licence and your intended use is not permitted
interpretation. J.D. contributed to guiding the mind-body retreat interven- by statutory regulation or exceeds the permitted use, you will need to
tion. J.M. contributed to participant recruitment, IRB, and data collection. obtain permission directly from the copyright holder. To view a copy of this
N.E.A.S. contributed to PC12 neurite analysis. M.L. contributed to dataset licence, visit http://creativecommons.org/licenses/by/4.0/.
curation and analysis. N.F. contributed to PC12 and ELISA assays. S.D.
contributed to sample collection and processing. A.V.C. contributed to This is a U.S. Government work and not under copyright protection in the US;
RNA-seq, bioinformatic analysis, and transcriptomics figure creation. L.C.M. foreign copyright protection may apply 2025
contributed to analysis. N.R. contributed to methodological consultation

Communications Biology | (2025)8:1525 19