Training the embodied self in its impermanence: meditators evidence neurophysiological markers of death acceptance

Abstract

Background

Human predictive capacity underlies its adaptive strength but also the potential for existential terror. Grounded in the predictive processing framework of brain function, we recently showed using a magnetoencephalogram visual mismatch-response (vMMR) paradigm that prediction-based self-specific neural mechanisms shield the self from existential threat—at the level of perception—by attributing death to the ‘other’ (nonself). Here we test the preregistered hypothesis that insight meditation grounded on mindful awareness is associated with a reduction in the brain’s defensiveness toward mortality. In addition, we examine whether these neurophysiological markers of death-denial are associated with the phenomenology of meditative self-dissolution (embodied training in impermanence).

Methods

Thirty-eight meditators pooled from a previous project investigating self-dissolution neurophenomenology underwent the vMMR task, as well as self-report measures of mental health, and afterlife beliefs. Results were associated with the previously-reported phenomenological dimensions of self-dissolution.

Results

Meditators’ brains responded to the coupling of death and self-stimuli in a manner indicating acceptance rather than denial, corresponding to increased self-reported well-being. Additionally, degree of death acceptance predicted positively valenced meditation-induced self-dissolution experiences, thus shedding light on possible mechanisms underlying wholesome vs. pathological disruptions to self-consciousness.

Conclusions

The findings provide empirical support for the hypothesis that the neural mechanisms underlying the human tendency to avoid death are not hard-wired but are amenable to mental training, one which is linked with meditating on the experience of the embodied self’s impermanence. The results also highlight the importance of assessing and addressing mortality concerns when implementing psychopharmacological or contemplative interventions with the potential of inducing radical disruptions to self-consciousness.

Introduction

The abstract capacity to consciously predict, simulate, and actively prepare for an uncertain future can be argued to be humanity’s primary adaptive strength. However, and perhaps uniquely, it also carries with it the existential certainty of inevitable mortality coupled with uncertainty about its cause and timing (Pyszczynski et al. 2015, Varki 2019). Recent frameworks of brain function take the notion of “prediction” further, casting it not as a thing brains do, but rather as the thing brains do. Brains continuously generate and update mental models—of the world and of their self—predicting the causes underlying incoming sensory input and minimizing errors between their predictions and actual sensory input (Friston 2010, Clark 2013, Hohwy 2013). Thus, as a predictive system bent on allostatic control, where “hierarchically deep self-modelling mechanisms function to ‘tune’ organisms to opportunities for adaptive action across multiple interlocking timescales” (Deane 2021), the potential for existential terror is omnipresent. More fundamentally, it has been argued that the basic experiences of phenomenal selfhood (Blanke and Metzinger 2009), conscious presence, and even the sense of “existing” arise from such predictive processes (Seth et al. 2011, Hohwy 2016, Deane 2021). It is therefore the very foundation of subjectivity that may come under threat (Metzinger 2024). As no action policies can avert this ultimate outcome, the human brain has evolved inbuilt automatic processes for avoiding awareness of its mortality. Hundreds of empirical studies, mostly within terror management theory paradigm (Greenberg et al. 1986, Pyszczynski et al. 2015), have described transparent cognitive processes which serve as anxiety buffers (Florian et al. 2002, Pyszczynski et al. 2004, Burke et al. 2010), and have outlined predispositions and circumstances which can undermine these buffers and lead to psychological vulnerability (Maxfield et al. 2014, Pyszczynski and Taylor 2016, Yetzer and Pyszczynski 2019, Mikulincer et al. 2020). In line with this, death anxiety has been proposed and empirically demonstrated across numerous psychopathological domains (Menzies and Menzies 2023) to be a transdiagnostic construct, underpinning the development and maintenance of mental disorders (Arndt et al. 2005, Iverach et al. 2014).

A few, mostly functional Magnetic Resonance Imaging (fMRI), studies have explored the brain responses to death-related stimuli. Initial results point at enhanced prefrontal activation, possibly indexing enhanced cognitive control (Yanagisawa et al. 2013, Klackl et al. 2014, Quirin et al. 2019), as well reduced posterior cingulate cortex (Shi and Han 2017, Hirano et al. 2021) and insula activation (Han et al. 2010, Shi and Han 2013, Klackl et al. 2014, Qin et al. 2018, Luo et al. 2019), interpreted as reflecting attenuated embodied self-processing (Craig 2009, Seth 2013, Trautwein et al. 2024). We draw on these latter findings, as well as on findings highlighting the link between mortality salience and self-awareness (Arndt et al. 1998, Guan et al. 2015, 2020, Wisman et al. 2015, Fan et al. 2023). However, motivated by the promise of clinical neuroscience (Huys et al. 2021, Smith et al. 2021), we aim to move beyond reverse-inference interpretations of brain activity to function (Poldrack 2006) toward a mechanistic account embedded in a unified understanding of brain function. As a first step in this direction, our work (Dor-Ziderman et al. 2019) has utilized the principles of predictive processing (Friston and Kiebel 2009, Clark 2013, Limanowski and Blankenburg 2013, Seth and Friston 2016) and self-specificity (Christoff et al. 2011), that is, the human tendency to divide its experiential field into self and nonself, to provide an initial characterization of a mechanistic neural process which upholds the high-order prior that “death pertains to the other,” but not to one’s self, thus avoiding information that could undermine the self model’s integrity (see “The death-denial visual mismatch response paradigm” section for the paradigm description). Based on this conceptual and empirical framework, we set out here to address the outstanding question regarding the neuroplasticity of these death-denying, existential processes. Are they fixed or amenable to change by mental training? And can they be re-wired in a manner conducive to well-being?

To test the plasticity of the neural mechanisms underlying death-denial, we recruited 38 Buddhism-inspired insight meditators (see “Participants” section). Buddhist contemplative traditions emphasize the realization of the self’s impermanence as one of its primary soteriological aims. Thus, a plethora of practices across all Buddhist traditions are geared toward reducing fear and denial and cultivating a recognition and acceptance of one’s finitude (Sogyal 1994, Dalai Lama 1997, Holocek 2013). Notably, the Satipatthana Sutra (Majjhima Nikaya 10: The Discourse on the Establishing of Mindfulness), one of the most widely studied discourses in the Pāli Canon and often cited as the foundation for contemporary mindfulness meditation practice, includes a series of instructions on Maranasati (Shonin and Van Gordon 2014), that is, meditating on various stages of body decomposition while maintaining the awareness that “This body of mine, too, is of the same nature as that body, is going to be like that body, and has not got past the condition of becoming like that body”. Some of the other more colorful death-related meditation practices come from the tantric Tibetan (as well as other) tradition and include practices such as Chod, rituals often set in graveyards which include visualizations of offering one’s body in a tantric feast (Edou 1995); Delogs, who are individuals that can induce near-death experiences (Drolma 1995, Van Gordon et al. 2018); Phowa, the transference of consciousness at the time of death (Nydahl 2012); and Tukdam, maintaining one’s consciousness in a meditative state after death (Rinpoche and Hopkins 1979, Lott et al. 2021). Empirically, little research has been conducted on such practices and their aftereffects. However, preliminary findings suggest that mindful awareness is associated with reduced self-reported death anxiety (Schultz and Arnau 2019, Anālayo et al. 2022, Jain 2024), as well as lower behavioral defensiveness and reduced thought suppression in response to mortality salience (Niemiec et al. 2010, Park and Pyszczynski 2019).

Importantly for the current study, none of the meditator participants had participated in meditation retreats explicitly focused on death-related themes (see Supplementary Figure 2). They were pooled from a previous research project [summarized in Berkovich-Ohana et al. (2020)] investigating the phenomenology (Ataria et al. 2015, Nave et al. 2021) and neural mechanisms (Dor-Ziderman et al. 2013, 2016, Trautwein et al. 2024) underlying meditative embodied self-dissolution.

Self-world boundary meditative dissolution phenomenology results

As part of that project, participants underwent a specially-tailored meditative training [for more details, see Supplementary Material of Nave et al. (2021)], which facilitated undergoing different tasks while maintaining and dissolving their felt sense of their self-world boundaries (SB) while their brains’ activity was measured by magnetoencephalogram (MEG), and complemented by micro-phenomenological interviews (Petitmengin 2006, Petitmengin et al. 2019). Their analysis indicated that SB dissolution was characterized by five experiential dimensions including the sense of agency, self-location, first-person perspective, attentional disposition, and body sensations (Nave et al. 2021). The neurophenomenological method applied to the categories included their mathematization (Berkovich-Ohana et al. 2020), that is, deriving numerical scores for each category’s subcategories based on their apparent order that emerged from the qualitative analysis. This transformed the categorical data into ordinal dimensions and allowed applying quantitative analyses (see “Phenomenological data” section for more detail). As the correlations between these five dimensions were high, a Phenomenological Dissolution Depth (PDD) was quantified as an unweighted sum of the above five dimensions. Interestingly, the PDD was orthogonal to the Phenomenological Affective Valence (PAV) dimension (see Supplementary Table 3), indicating that dissolution depth was not predictive of how blissful-to-terrible experiences would be (Nave et al. 2021). As an explanation for this result, it was suggested that the affective valence attached to the meditation-induced self-dissolution state was related to the fear of nonbeing (craving for existence, Bhava Tanha in Pali). The degree to which the mind could ease into a state of “dissolving” its self was hypothesized to relate to the degree to which the brain’s predictive system allowed associating death with self (see next section). Thus, this unique sample allowed capitalizing on the rich previously collected phenomenological data and investigating putative links between death-denial and practitioners’ experiences during self-deconstructive meditation practice (Dahl et al. 2015).

The death-denial visual mismatch response paradigm

The paradigm (Dor-Ziderman et al. 2019) employed a self-specific (processes differentiating self from other; Christoff et al. 2011) prediction-based (embedded in the predictive-processing framework; Friston 2005, Clark 2013) neuroscience paradigm. Death-denial was defined as an encoded statistical prior belief that “death concerns others, but not our self,” and was operationalized via a MEG visual MisMatch Response (vMMR; Kremláček et al. 2016) paradigm using self and other facial images (Sel et al. 2016) coupled with either death or merely negative linguistic stimuli (see Supplementary Information section SI1). There is growing evidence that vMMR paradigms, and particularly its later component (Poublan-Couzardot et al. 2023), reflect Bayesian learning, and that predictive coding can serve as a unifying framework of MMR generation (Garrido et al. 2009). The vMMR response was obtained by presenting deviant sensory stimuli following 3–6 presentations of standard stimuli consisting, in each trial, of either self or other facial images. Prior to the presentation of these facial sequences, a word appeared above the center of the screen to provide a context which was either death-related or merely-negative. This word remained for the duration of the sequence (Fig. 1).

Using this paradigm, we reported (Dor-Ziderman et al. 2019; Fig. 2a) vMMR effects at ∼250 to 300 ms postdeviant stimuli presentation over posterior-left sensors under a negative context regardless of face identity (negative-self and negative-other conditions, NS and NO respectively) and under a death context when a stranger’s facial image was presented (death-other condition, DO). Crucially, the presentation of self-images displayed under a death-related word (DS condition) resulted in a sharp attenuation of this basic predictive perceptual mechanism as indicated by a lack of a detectable vMMR effect. Other results from that study demonstrated a functional link between how death impacted self-image vs. other-image which manifested as an inverse correlation between the DS and DO condition, and confirmed this effect in an additional behavioral experiment showing that death-related words bias perceptual judgment on facial self and other morphed video clips. Together, these results were interpreted as indexing a prior belief categorizing death as relating to other, and thus incoming death cues tilted the self-other perceptual system toward the other, shielding the self from existential threat. This led to the construction of the death-denial index (DDI), computed by subtracting the DS from the DO condition, which was shown to predict implicit fear of death (Dor-Ziderman et al. 2019).

Aims and hypotheses

Building on these prior results, the present study aimed at examining: [1] whether meditation impacts the brain’s early defenses against mortality, determining whether disrupting these defense mechanisms predicted [2] mental health, as well as [3] the valence of meditative self-dissolution experiences. For the first aim, our preregistered hypotheses included [H1a] a significant response to deviancy in the condition coupling death-related words and self faces, in contrast to the data reported for meditation-naïve participants. However, as a manipulation check, similar to meditation-naïve participants, we hypothesized that [H1b] the temporal and spatial features of the vMMR effect (deviant–standard stimuli), beyond conditions, would correspond to those of the meditation-naïve participants—presenting itself at ∼250 to 300 ms poststimuli onset at left-central-posterior sensors. Additionally, we hypothesized [H1c] higher meditation lifetime experience (total hours) to predict lower DDI scores. Finally, while belief in literal immortality has been shown to diminish mortality defenses (Vail et al. 2010, 2012), we hypothesized that [H1d] the DDI would not be affected by afterlife beliefs as the vMMRs targeted preconceptual rungs of the cognitive hierarchy. The second and third aims were exploratory. We examined whether reduced defensiveness [H2a] predicted well-being while [H2b] did not predict trait anxiety, and whether [H3] the DDI predicted the valence of meditators’ embodied self-dissolution experiences.

Methods

Participants

A preregistered power estimation (G*Power 3.1), based on reported effect sizes (Dor-Ziderman et al. 2019), indicated that n = 39 participants would yield power = 0.95 for a one-tailed t-test at α = 0.05. For individual-difference analyses, n = 37 would achieve power = 0.80 for one-tailed test at α = 0.05 of an effect size r = 0.4. The sample consisted of experienced meditators (n = 38, 20 females; age = 40.4 ± 11.2, range = 26–72 years), recruited through Tovana (Israel Insight Meditation Society), a nondenominational nonprofit offering frameworks for learning the practice of insight meditation grounded on mindful awareness in the Theravada Vipassana tradition. The meditators were pooled from a previous project, where inclusion criteria specified having attended at least one residential meditation retreat (minimum 7 days), and a regular daily practice of minimum 1 year (ranging between 15 and 120 practice minutes per day). Meditation experience was measured as lifetime meditation practice hours (mean = 4146 ± 5111, range = 115–24 837). Phenomenological data were collected between November 2018 to December 2019; neural data were collected between November 2019 to July 2020. Other pre-established exclusion criteria were conditions limiting MEG data quality (dental splints, artificial cardiac pacemakers), active psychiatric disorders, current psychiatric medication, and not having normal or corrected-to-normal vision or hearing. Study was approved by the institutional review board of the Education Faculty, Haifa University. Participants provided written consent and were financially compensated for their time.

Phenomenological data

A detailed description of data collection and analysis can be found in a previous publication (Nave et al. 2021). Briefly, interviews were conducted based the micro-phenomenological interview method (Petitmengin 2006), which facilitates richly detailed descriptions of concrete moments of experience, including prereflective attentional, affective, and sensorial facets, while reducing subjective bias. Micro-phenomenology is recommended for and has been previously used in neurophenomenological meditation research (Petitmengin et al. 2019, Berkovich-Ohana et al. 2020, Lutz et al. 2024). Interviews lasted 21–57 min (mean = 33) conducted in a quiet room. Participants were directed to focus on one specific moment of receiving the audible instruction to enter the meditative self-boundaries dissolution state. Interviews were recorded and transcribed verbatim while preserving paraverbal communications. Data were processed through successive stages of analysis including thematic coding of relevant passages, data-driven generation of categories and subcategories, and categorization of each report into the various categories. Interrater agreement of two independent raters yielded an average fuzzy kappa (Kirilenko and Stepchenkova 2016) of κ = 0.69. Numerical scores were derived for each subcategory, transforming categorical data into ordinal dimensions and facilitating correlations with neural effects.

Self-report measures

To address the hypotheses regarding associations between the DDI and ontological belief, mental health, and self-dissolution valence, several questionnaires were employed. These included the Afterlife belief certainty (ABC) which was measured via a novel tool where participants responded on a numerical scale ranging from −10 to 10, with −10 indicating certainty in death as annihilation (“when the heart and brain stop working the soul/consciousness terminally ends”) to 0 (“I don’t know”) to 10 indicating certainty in continuation after death (“The soul/consciousness continues after death”). Trait mental health was measured using the well-established Trait Anxiety Inventory (Spielberger et al. 1983) and the General Well-Being Scale (Dupuy 1978). Altered state was gauged via the Altered States of Consciousness Questionnaire (5D-ASC; Dittrich 1998).

MEG data

Data collection, preprocessing, analyses, and statistics were preregistered, and are similar to those described in our previous publication (Dor-Ziderman et al. 2019).

Data collection

Ongoing brain magnetic activity was recorded (sampling rate, 1017.23 Hz, online 1–400 Hz band-pass filter) using a whole-head 248-channel magnetometer array (4D Neuroimaging, Magnes 3600 WH) in supine position inside a magnetically shielded room. Reference coils located above the head oriented by the x, y, and z axes were used to remove environmental noise. Five coils were attached to the participant’s scalp for recording the head position relative to the 248-sensor array. The head shape was manually digitized using a Polhemus Fastrak digitizer. A photosensitive diode on the screen recorded the onset time of visual stimuli. A response box was used for collecting manual responses.

Data preprocessing

Data were analyzed using the FieldTrip toolbox (Oostenveld et al. 2011) as well as MATLAB R2020a (MathWorks, Natick, MA, USA) custom-made analysis scripts. External noise (e.g. power-line, mechanical vibrations), jumps in the MEG signal, and heartbeat artifacts were removed from the data using a predesigned algorithm (Tal and Abeles 2013). One bad channel was detected and excluded from further analysis. Data were segmented into 600 ms epochs (100 ms before face presentation to 500 ms after). Data epochs of interest were checked for artifacts using a semiautomatic routine in which data were 60 Hz high-pass filtered for detecting and rejecting trials containing muscle artifacts, then subjected to an independent component analysis (Bell and Sejnowski 1995) to remove from the data any remaining variance related to eye blinks, eye movements, and heartbeat artifacts (Jung et al. 2000), and finally data were visually inspected for any remaining trials with artifacts.

Event-related fields (ERFs) were calculated by low-pass filtering the data using a two-pass Butterworth filter with a filter order of 4 and a frequency cutoff of 40 Hz. ERFs were baseline corrected using an interval of 100 ms before faces presentation. A planar gradient transform was calculated (Bastiaansen and Knösche 2000), as it simplifies the interpretation of the sensor-level data by typically placing the maximal signal above the source (Hämäläinen et al. 1993). We avoided using the grand average data to manually select components’ time-windows and electrode sites, as this has been shown to often yield bogus results (Luck and Gaspelin 2017). Time and sensors of interest were determined using nonparametric cluster-based permutation statistics (see “Statistics” section) using paired-samples t-test between standard and deviant trials, first on the time-domain (averaged over all conditions, sensors, and trials) and then on the sensor domain (averaged over determined time-window, conditions, and trials). Circular analysis (Kriegeskorte et al. 2009) was avoided by using orthogonal contrasts (Litvak et al. 2011, Kilner 2013, Luck and Gaspelin 2017) for data selection (detecting times and sensors of interest stages) and inference (comparing the different conditions). Additionally, the number of trials was equalized for each condition per subject before averaging, in this way keeping the selection stage completely uninformed about what the conditions were and not inflating type I errors (Brooks et al. 2017). Finally, for each subject ERF power values were collapsed over the significant sensor cluster and the significant time-window, and difference waveforms (deviant–standard trials) were computed. These were subjected to repeated-measures ANOVAs, with posthoc one-sampled and paired-sample t-tests, as well as Pearson (rp) and Spearman (rs) correlations (depending on whether data were continuous/ordinal).

Source of evoked activity were estimated on the identified (in the sensor-level analysis) times-of-interest using a time-domain beam-forming approach on the magnetometers sensor data (linearly constrained minimum variance, Van Veen et al. 1997). For each participant, a single shell brain model was built based on a template brain (Montreal Neurological Institute), which was modified to fit each participant’s digitized head shape using SPM12 (Welcome Department of Imaging Neuroscience University College London, www.fil.ion.ucl.ac.uk). The participant’s brain volume was then divided into a regular grid. The grid positions were obtained by a linear transformation of the grid positions in a canonical 1 cm grid. This procedure facilitates the group analysis because no spatial interpolation of the volumes of reconstructed activity is required. For each grid position, the lead field matrix was calculated according to the head position in the system and the forward model. A common spatial filter was then constructed for each grid point using the covariance (of all trials of all conditions) and the lead field matrices. Single trial cortical (deep brain structures were masked out) source power estimates were calculated based on single-trial covariance matrices and the common spatial filter. For each subject, we statistically compared differences between the standard and deviant faces for each condition separately, what resulted in a 3D t-value distribution of the vMMR effect for each condition. These distributions were then pooled over all participants and subjected to second-level two-tailed nonparametric cluster-based permutation statistics on pooled t-values.

Nonparametric cluster-based permutation statistics (Maris and Oostenveld 2007) were used both on the time domain and spatial domain to identify significant spatial clusters of differential ERF activity. This type of test controls the type I error rate in the context of multiple comparisons by identifying clusters of significant differences over space. This approach was chosen as it makes no assumptions on the underlying distribution, and is unaffected by partial dependence between neighboring sensors/voxels. In addition, this approach has been shown to yield nominal (noninflated) false-positive rates for spatial extent (Eklund et al. 2016). The cluster-level statistics, defined as the sum of t-values within each cluster, were evaluated under the permutation distribution of the maximum (minimum) cluster-level statistic. This permutation distribution was approximated by drawing 1000 random permutations of the observed data. The obtained P-values (the minimum being P = .002 for a 1000 permutations tail-corrected distribution) represent the probability under the null hypothesis (no difference between the conditions) of observing a maximum (minimum) cluster-level statistic that is larger (smaller) than the observed cluster-level statistics.

Results

Meditators’ neurophysiological markers indicate acceptance rather than denial of death

Following the preregistered methodology, cluster-based permutation tests yielded a significant vMMR effect across conditions during a 247–288 ms poststimulus time-window (Fig. 2b) localized in sensor-space as a left-posterior cluster (22 sensors, P = .009, Fig. 2c)—thus replicating the original study’s vMMR signal [H1b]. In addition to the hypothesized time-window, a later time-window was detected at 456–500 ms poststimulus (Fig. 2b) over central-posterior electrodes (Supplementary Figure SI1a). However, as it was not sensitive to condition it is not further discussed (see Supplementary SI2 and Figure SI1b).

Confirming our main preregistered hypothesis [H1a], self-specific predictive processes were maintained under existential threat, as indicated by a highly significant vMMR (t(37) = 3.61, P < .001, d = 0.59, 95% CI [0.24–0.93], Fig. 2d) for the meditators during the DS condition. This finding sharply differed from the previous meditation-naïve results which showed a null effect for the DS vMMR (Fig. 2a). To formally quantify meditators vs. nonmeditators group differences, a secondary analysis of the previously published nonmeditators data was conducted. The time and space-resolved cluster-data of each group were normalized (by subtracting the mean and computing z-scores), and a mixed ANOVA was computed with GROUP (meditators/nonmeditators) as a between-subject factor, and IDENTITY (self/other) and PRIME (death/negative) as within-subject factors. The results indicated a significant GROUP by IDENTITY by PRIME interaction effect [F(1,60) = 6.762, P = .012, |$\eta _p^2$|=0.101]. Post-hoc analyses using independent samples t-tests showed that the DS condition was significantly larger for the meditators relative to the nonmeditators (t(60) = 2.658, P < .01, d = 0.693, 95% CI [0.14–1.24]) while the other conditions were not significantly different between the groups (Ps > .17). These data suggest that meditators’ brains functioned differently from nonmeditators specifically when processing the combination of death and self-related information.

Given the dramatic vMMR differences, the DDI (computed by subtracting the DS from the DO condition) was reverse-scored and re-named the Death Acceptance Index (DAI), alongside a control NI (subtracting NO from the NS condition). We then tested whether the DAI was correlated with meditation experience and ontological afterlife beliefs. Unlike our prediction [H1c], the DAI did not correlate with meditation experience (rs = −.07, P = .7, n.s.). Regarding afterlife beliefs, most of the participants endorsed a belief in the continuation of consciousness after death. Only one meditator’s score was negative (−2) indicating an uncertain tendency toward annihilation, five indicated agnosticism (0), and the rest had positive scores (median = 8.5) indicating a very high degree of certainty. As predicted [H1d], the ABC was not correlated with the DAI (rs = −.12, P = .5, n.s.). Additionally, the ABC was not correlated with meditation experience (rs = 0.21, P = .21, n.s.).

To understand the pattern of the meditators’ data, we conducted a 2 × 2 repeated-measures ANOVA with IDENTITY (self/other) and PRIME (death/negative) as within-subject factors. The results indicated a nonsignificant main effect for PRIME [F(1,37) = 1.07, P = .308, n.s.] and for the interaction [F(1,37) = 1.8, P = .188, n.s.]. However, a main effect was observed for IDENTITY [F(1,37) = 4. 5, P = .041, |$\eta _p^2$|=0.11, see Figs 2d and 4a], indexing the well-established self-advantage effect in behavior (Bortolon and Raffard 2018), electro/magneto-physiology (Sel et al. 2016, Trautwein et al. 2016), and fMRI (Platek et al. 2006, Sui et al. 2015). Furthermore, one-sample t-tests indicated that the conditions involving one’s self image (DS and NS) both evidenced significant vMMRs (t(37) = 3.61, P < .001, d = 0.59, 95% CI [0.24–0.93[and t(37) = 2.04, P = .049, d = 0.33, 95% CI [0.002–0.66], respectively), while the conditions involving a stranger’s image (DS and NO) did not (Ps > .24). These results provide evidence that for meditators, death does not pose a uniquely aversive category and is rather processed no different than merely-negative stimuli.

Beamformer source estimation analyses of the self vMMR main effect (Fig. 3a) revealed a significant (tail-corrected) right-lateralized positive (dev > std) cluster (21 voxels, P = .002) spanning mostly visual (cuneus and occipital) and face recognition (fusiform) regions, but extending also to precuneus and middle-temporal regions. These regions are congruent with our own previous study (Dor-Ziderman et al. 2019), other self-face vMMR (Sel et al. 2016), and recognition studies (Platek et al. 2008). Those studies, however, also reported an additional frontal component which was not present here. In addition to the expected vMMR effect, a small cluster (2 voxels, P = .002) in the opposite direction (std > dev) located in the right inferior temporal area was detected (see Table 1 for cluster information). In line with the nonsignificant sensor-level effects for the other vMMR conditions (DO and NO), no significant source-level clusters emerged for the other main effect. Source localization of the DS and NS conditions separately revealed very similar clusters (all significant at P = .002, tail-corrected) topographies (Fig. 3b and c, and Supplementary Table 2). Furthermore, directly comparing the DS and NS conditions did not yield significant clusters (Ps > .3). Overall, these results further support the claim that the meditation-trained brain does not differentiate between merely-negative and death-related (but equally negative) stimuli. Put another way, death reminders do not constitute a uniquely aversive category the brain needs to monitor and defend against.

To examine links between the pattern of results in the meditators’ data and mental health, the self-advantage effect was quantified by subtracting the mean of the DO and NO conditions from the mean of the DS and NS conditions, and Pearson correlation analyses with self-reported trait anxiety and well-being were computed. As predicted [H2a, b], the results showed that the self-advantage effect was not correlated with trait anxiety (rp = 0.14, P = .44, n.s.), but was negatively correlated with well-being (rp = −.39, P = .032, Fig. 4b). Participants whose vMMRs indicated less self prioritization reported higher well-being. We also verified that the self-advantage effect was not associated with age, education, and gender (Ps > .3).

In summary, the findings demonstrated a radically different manner of processing the coupling of death and self-stimuli in meditators, mirroring the well-established self-advantage effect irrespective of death specificity. These results lead us to cautiously suggest that meditators’ brains signal an acceptance of death rather than its denial.

Neurophysiological markers of death acceptance predict positively-valenced embodied self-dissolution experiences

To test the hypothesized [H3] links between death acceptance and embodied self-dissolution phenomenology, we computed correlations between the DAI and the experiential dimensions of the self-boundary dissolution state. Specifically, we looked at the AV experiential dimensions extracted from the phenomenological interviews, as well as the Anxious Ego Dissolution (AED) subscale of 5D-ASC. The AED subscale was of special interest to us as it directly related to our hypothesis regarding the association between self/ego dissolution and the neurophysiology of death-denial/acceptance. Both the phenomenological and self-report measures were obtained shortly after participants volitionally induced self-dissolution states in the lab.

As outlined above [and detailed in Nave et al. (2021)], the phenomenological analysis yielded six ordinal dimensions, five of which were highly inter-correlated and comprised the PDD score, also strongly correlated with lifetime meditation experience. The sixth dimension, the PAV, indexed the valence of the dissolution experiences and was orthogonal to the PDD and unrelated to meditation experience. However, a strong and positive correlation was found between the PAV and the DAI (rs = 0.524, P < .001, Fig. 5a). Similarly, The AED subscale was negatively correlated with the DAI (rs = −.421, P = .008, Fig. 5b), indicating that less death acceptance was linked with more anxious self-dissolution experiences. Both the DAI and the AED were not correlated with the control NI (Ps > .69). Additionally, the PDD, as well as the remaining 5D-ASC categories (oceanic boundlessness, visionary restructuralization, auditory alterations, and vigilance reduction) were not linked with the DAI (Ps > .22, see Supplementary Table 3).

Discussion

The main objective of the study was to confirm the preregistered hypothesis that insight meditation grounded on mindful awareness practice was associated with attenuation of the brain’s automatic tendency to deny death by attributing death to the other but not to one’s self. Our results confirmed that meditators differed from nonmeditators in that their brains displayed a large and significant response to deviancy in the condition coupling death-related words and self faces, in contrast to the data reported for meditation-naïve participants where the coupling of death and self-related information led to a suppression of the vMMR effect (Dor-Ziderman et al. 2019). In predictive processing terms, for meditators, the repeated appearance of self faces together with death-related words was treated by their brains as reliable (high precision) information, thus leading to formation of predictions and consequently to a “surprise” response when they were violated by deviant stimuli. The pattern of results in the meditators’ data indicated no significant differentiation in processing naturalistic self stimuli under merely-negative and death-related contexts, indicating that death did not present a uniquely aversive category for the meditators. Rather, the results mirror the well-established self-advantage effect, aligning with the notion of meditation-induced reduced self-other differentiation (Dambrun and Ricard 2011, Giommi et al. 2023), which predicted well-being and not pathology. These results complement studies showing that meditators display smaller differences in facial self vs. other EEG saliency effects relative to nonmeditators (Trautwein et al. 2016), and thus concur with the notion that meditation training can have a long-lasting impact on maladaptive aspects of self-processes in a manner conducive to enhanced well-being (Dambrun and Ricard 2011).

Overall, the vMMR results are interpreted as indicating an acceptance of death rather than denial. Furthermore, that insight meditation grounded on mindful awareness may be a safe and effective means for confronting one’s mortality. These findings complement the few existing self-report and behavioral studies (Niemiec et al. 2010, Moon 2019, Park and Pyszczynski 2019, Blomstrom et al. 2020) indicating that mindfulness is associated with reduced death-anxiety and defensiveness, but go further in showing that these stabilize as trait effects, and occur much earlier than previously assumed, impacting the brain’s cognitive hierarchy already at the level of perception. Importantly, as no correlation between meditation experience and the DAI was found, it is unclear whether meditation n.s. is the mechanism-of-action. It is possible that there is a ceiling effect, where hundreds rather than thousands of practice hours are sufficient for shifting the brain from denial to acceptance. Alternatively, it could be the case that the embodied inculcation of certain contemplative notions at the heart of the mindfulness framework (Anālayo et al. 2022), and particularly that of impermanence (Hirschberger and Shaham 2012, Gokhale and Lawrence 2021), could be the mechanisms-of-action rather than, or in conjunction with, meditation practice. Such a view would be compatible with a few studies showing that inducing an embodied experience of our finitude, rather than a conceptual one, can lead to mortality-induced growth (Cozzolino 2006, Cozzolino et al. 2014). These questions should be further investigated using longitudinal studies ideally with different styles of meditations (e.g. analytical vs. experiential) including measures of constructs such as impermanence (Fernández-Campos et al. 2021). Nevertheless, the present findings do indicate that the neural markers of death acceptance cannot be explained by heightened afterlife beliefs, arguably a reflective form of mortality denial (Holbrook et al. 2020, Dor-Ziderman and Keenan 2024, Metzinger 2024), but rather by the phenomenology of embodied self-dissolution. This observation highlights the critical role of experiential insights in mediating meditation-related changes.

A remarkable neuro-phenomenological finding of the present study was that neural markers of death acceptance strongly predicted positively-valenced and less anxious meditation-induced dissolution experiences. This finding is not only novel, but also important, as it allows for putting forth the first empirically-grounded hypothesis as to why some experience self-dissolution as blissful while others as terrible. Self-dissolution experiences readily occur in experienced meditators. Lindahl et al. (2017) reported that from a sample of 60 advanced meditation practitioners from different Buddhist traditions, 75% reported alterations in embodied-self features, the most common of which were various alterations to the sense of the self-world boundary. Furthermore, a range of affective responses were associated with this change from neutral curiosity, to bliss and joy, to fear and terror. Recent reports highlight that meditation-induced experiences may manifest in unwholesome ways with severe and prolonged debilitating effects on well-being and social function (Lindahl et al. 2017, Compson 2018, Van et al. 2018, Baer et al. 2019, Schlosser et al. 2019).

Importantly, radical disruptions to self experience and their associated dangers are not limited to meditation. Ego dissolution is a central and common (Lebedev et al. 2015, Letheby and Gerrans 2017, Stoliker et al. 2022) experience occurring under acute psychedelic intake—a mode of intervention fast becoming mainstream for treating a number of psychopathologies (Carhart-Harris and Goodwin 2017). Ego dissolution has been shown, on the one hand, to be an important mechanism-of-action of therapeutic effects (Roseman et al. 2018, Kałużna et al. 2022); yet, on the other hand, is often accompanied by intense fear (Lebedev et al. 2015, Deane 2020) and may result in further psychopathology (Evans et al. 2023, Sielaff et al. 2023). Thus, the present study contributes to the effort of mapping what constitutes a safe and wholesome manifestation of selfless insight in meditators (Van et al. 2018, Baer et al. 2019, Schlosser et al. 2019), thus providing potentially important information for psychedelic psychotherapies involving the induction of such states. Facing one’s fear, denial and anxiety of death prior to engaging in practices or therapeutics which may induce states of self/ego-death might reduce the risks involved and yield better clinical outcomes.

Study limitations include the lack of a proper control group. However, it is important to emphasize that the meditators’ results were not different in degree from the previously published results, but rather showed qualitatively different patterns of activity. Additionally, a formal exploratory analysis reinforced the group differences and pinpointed the changes to the condition coupling death and self, as hypothesized. Future experimental considerations include recruiting better balanced populations in terms of ontological belief, as well as employing longitudinal designs which allow inferring causality. Furthermore, such studies could employ more sophisticated neurocomputational modeling for delineating the predictive processing path to meditation-induced death acceptance. Another possible limitation of the study is that participants’ descriptions of dissolving self-world boundaries were affected by demand characteristics. While this interpretation is in principle possible, we believe it is unlikely due to the validity and reliability safeguards of the micro-phenomenological interview and analysis methodology (Petitmengin 2006, Valenzuela-Moguillansky and Vásquez-Rosati 2019), which were implemented here in the fine-grained phenomenological descriptions (as reported in Nave et al. 2021).

In sum, we have shown that prediction-based self-specific brain defenses against mortality are expressed differently in experienced meditators compared to novices. These data provide empirical evidence supporting the hypothesis that meditation can alter the brain’s early response to death from denial to acceptance, thus demonstrating that the neural mechanisms engaging in the denial of death are amenable to change through mental training. We also showed that neurophysiological markers of death acceptance were associated with trait well-being, as well as more positively-valenced advanced meditative self-deconstruction states. We thus made a case for implementing mindfulness-based training in death acceptance as a potentially important clinical factor when considering psycho/pharmacological interventions with the potential of radically disrupting self-experience. It is our hope that this study will spark public, scientific, and clinical interest in understanding how we deal with our mortality, and the transformative potential that facing up to it may hold.

Acknowledgements

We thank Thomas Metzinger for reading an earlier version of the manuscript and offering useful clarifications and suggestions.

Author contributions

Yair Dor-Ziderman (Conceptualization, Methodology, Formal analysis, Investigation, Writing—original draft, Visualization, Funding acquisition), Yoav Schweizer (Investigation, Data curation, Writing—review & editing), Ohad Nave (Methodology, Resources, Writing—review & editing), Fynn-Mathis Trautwein (Conceptualization, Methodology, Resources, Writing—review & editing), Stephen Fulder (Conceptualization, Writing—review & editing), Antoine Lutz (Conceptualization, Methodology, Writing—review & editing, Supervision), Abraham Goldstein (Conceptualization, Resources, Writing—review & editing, Supervision), Aviva Berkovich-Ohana (Conceptualization, Resources, Writing—review & editing, Supervision, Project administration, Funding acquisition).

Supplementary data

Supplementary data is available at Neuroscience of Consciousness online.

Conflict of interest

The authors declare no conflicting interests.

Funding

The research was supported by the Foundational Questions Institute (FQxI) grant 2020–223973, the Bial Foundation grant 191/20, Israel Scientific Foundation (ISF) grant 677/21, and The Tiny Blue Dot (TBD) foundation grant 43777846.

Data and code availability

Data and code will be shared on request with the corresponding author after signing a data-sharing agreement, and contingent on approval from the requesting researcher’s local ethics committee.

References