Acupuncture is a multi-dimensional intervention. Needle stimulation, the essential component of the acupuncture procedure, is known to induce different types of somatosensory afferent activation and consequent modulatory effects on the brain (Chae et al., 2013; Maeda et al., 2017; Kim et al., 2020), as well as responses in other systems, including the immune and endocrine systems (Wei and Hsieh, 2020; Liu et al., 2021; Oh and Kim, 2021). Concurrently, in clinical settings, the needling procedure is necessarily accompanied by additional and supplementary procedures (e.g., preparations of needle insertion), which may potentially lead to various therapeutic contexts. The manipulation of therapeutic conditions in acupuncture (Cao et al., 2018) suggests that not only somatosensory but also other sensory modalities and related cognition are involved in acupuncture in the brain.

Indeed, existing findings in neuroscience consolidated cognitive modulation to somatosensory process in brain. Primary somatosensory cortex (SI) was modulated by attentional process (Steinmetz et al., 2000; Bardouille et al., 2010; Dockstader et al., 2010), cognitive imagery (Jafari et al., 2020), and action understanding (Valchev et al., 2017). Moreover, an animal study showed that decision and reward history affected the neuronal activity in the primary sensory thalamus in the ascending sensory pathway (Waiblinger et al., 2018). These brain mechanism suggests that somatosensory afference and contextual manipulation of acupuncture, a compound stimulus, can be interacted in brain, which differs from placebo mechanism in brain (Wager and Atlas, 2015).

In our previous studies, we implemented an experimental condition of sham acupuncture (phantom acupuncture, PHNT), which dissociated somatosensory afferents of needling stimulation from the total effect of acupuncture treatment (Lee et al., 2014, 2019; Makary et al., 2018). PHNT induced needling credibility using only contextual manipulation of visual information in the stimulation session with verbal instruction in the preparatory session, and showed activation of multiple brain regions (e.g., dorsolateral prefrontal cortex and ventrolateral prefrontal cortex). These brain regions could be associated with the cognitive/affective process (Makary et al., 2018) and potentially lead to an expectation-related placebo analgesic effect (Lee et al., 2019). Moreover, actual needling stimulation in addition to the same protocol of real acupuncture (REAL) showed greater activation in brain regions associated with somatosensory afferents (e.g., the posterior insular cortex and somatosensory cortex) than PHNT. Nevertheless, the model-based voxel-wise assessment of functional brain data in previous studies had limitations for taking account of underlying brain intrinsic processes.

In fact, REAL and PHNT induced not only therapeutic components of somatosensory afferent or cognitive/affective but also non-therapeutic components, which were implicitly engaged with the protocols. Moreover, potential therapeutic components could exist in both REAL and PHNT and may interact with other components. For instance, while both REAL and PHNT induced cognitive/affective brain regions as expected, their associations with expectations for acupuncture treatment differed between REAL and PHNT. Furthermore, repeated needling stimulations in a session showed heterogeneous brain response patterns (Napadow et al., 2013), indicating intrinsic interactions around somatosensory afferent process in the brain. These phenomena can be considered in the typical setting of acupuncture and its related therapeutic outcomes.

Here, we conducted another REAL/PHNT experiment with twenty-five healthy participants for fMRI scan, and independent component (IC)-wise assessment with the concatenated fMRI data to investigate underlying brain intrinsic processes induced by acupuncture. Independent component analysis (ICA) separated brain responses to contrasted experimental paradigms into ICs (i.e., data-driven approach) and generalized linear model (GLM) assesses their significances for experimental protocols (i.e., model-based approach). By combining experimental dissociation (i.e., REAL and PHNT) and analytic dissociation (i.e., IC-wise GLM), the intrinsic brain components were identified. Moreover, we assessed the psychological (i.e., somatosensory sensation) and physiological [i.e., heart rate (HR)] measures of REAL and PHNT to elucidate the functional characteristics of the brain components.

Acupuncture is a clinical intervention that consists of multiple therapeutic components of specific effects from needling insertion (i.e., somatosensory afferent) and non-specific effect from patient’s belief and knowledge in acupuncture treatment (i.e., cognitive/affective) (Ho et al., 2021). Despite accumulating clinical evidence of acupuncture stimulation/treatment in patients with various diseases and even in healthy subjects, the underlying mechanism that explains the specific and non-specific effects of acupuncture has not yet been fully investigated. In this study, we attempted to identify specific underlying brain mechanisms for these two therapeutic components induced by compound stimulus of somatosensory stimulation and therapeutic contextual manipulation, which is known to be essential in a typical clinical setting. Using our team’s novel experimental setting, we were able to separate two therapeutic components during the fMRI scan.

However, each participant’s brain activation was not simply manipulated by the experimental protocols as we intended. Not only the manipulated therapeutic components (i.e., somatosensory afferent and cognitive/affective) but also non-therapeutic components, which were implicitly engaged with the protocols, and potential therapeutic components, which could exist in both REAL and PHNT, might interact with each other. Thus, model-based analysis (e.g., voxel-wise GLM) had difficulties assessing these underlying brain processes.

To address the limitation of voxel-wise GLM, we performed data-driven source separation for brain responses by REAL and PHNT. Concatenated BOLD data were decomposed into statistically independent components (ICs), and the dependencies of these ICs on the protocols were assessed (IC-wise GLM). Thus, the combination of experimental dissociation (i.e., REAL and PHNT) and analytic dissociation (i.e., IC-wise GLM) could identify brain response patterns in aspects of somatosensory afferent and cognitive/affective process components.

The temporal concatenated ICA for IC-wise GLM is based on the heterogeneity of brain response patterns across repeated stimulations. REAL and PHNT induced activation in multiple brain regions, which was demonstrated in group average maps from voxel-wise GLM. However, brain activation also had intrinsic variabilities across repeated stimulations during scans, which were variances or standard errors of PEs in the voxel-wise GLM. For instance, a previous fMRI study showed that repeated needling stimulation induced HR decreases as a group average; however, responses of HR increases (Napadow et al., 2013) was also shown in minority in stimulations. This increased HR was accompanied by a different pattern of brain response from that of decreased HR stimulations (Napadow et al., 2013). In addition, habituation from repeated stimulations could decrease voxel activation differentially according to brain regions (Klingner et al., 2011), which causes altered brain response patterns for the same stimulation. Moreover, brain response patterns for somatosensory process could differ across individuals due to behavioral characteristics (Chae et al., 2014). We concatenated REAL and PHNT for IC-wise GLM, which could enhance variability for somatosensory afferent and cognitive/affective process in addition to the heterogeneity in brain responses to repeated stimulations.

Then, we performed GLM analysis for the time series of each IC, which was comparable with voxel-wise GLM that used time series of voxels as independent variables. Instead of FWE correction for whole brain voxels in voxel-wise GLM, our IC-wise GLM performed a Bonferroni correction for multiple comparisons of the resulting 39 ICs, which carried out REAL and PHNT-associated ICs (i.e., SA1, SA2, CA1, and CA2 for REAL; and CA1 and CA2 for PHNT). Moreover, paired comparisons between REAL and PHNT showed that REAL had statistically stronger SA1 and SA2 associations than PHNT. Thus, REAL-specific SA1 and SA2 were identified as somatosensory afferent components according to the experimentally designed dissociation. CA1 and CA2 were identified as cognitive/affective process components because they are common to both REAL and PHNT. Furthermore, the component separations within somatosensory or afferent cognitive/affective process indicate the significance of intrinsic variabilities in brain response patterns.

The thresholded spatial maps of SA1 include SMA, bilateral aIns, and PAG, which are involved in interoceptive attention and motor reaction processes. aIns plays a crucial role in the salience network in the detection of external salient stimuli (Menon and Uddin, 2010) and interoception (Craig, 2009). In particular, activation of the left aIns was found for positive and affiliative emotional feelings from parasympathetic afferents, whereas activation in the right aIns was shown for arousing to the body from sympathetic afferents (Craig, 2002, 2009). PAG is involved in descending control circuitries for somatosensory (Goksan et al., 2018) and motor (Koutsikou et al., 2017) processes to react against external stimuli. SMA is known to be crucial for motor behavior process for action preparation and initiation (Nachev et al., 2008), which also shows activation by multiple sensory modalities (Korvenoja et al., 1999; Lanzilotto et al., 2016; Lima et al., 2016).

While, SA2 includes brain areas of left (ipsilateral) SII and pIns, which are associated with the somatosensory representation process. Indeed, SA2 response was positively correlated with somatosensory sensations in REAL. Studies with somatosensory stimulation have shown ipsilateral activation of SII for tactile representation and perception process (Yu et al., 2017; Lamp et al., 2018; Pala and Stanley, 2022). Moreover, ipsilateral SII activation was accompanied by distinct relationships from contralateral SII regarding functional connectivity (Yu et al., 2017) and descending connectivity modulation (Fardo et al., 2017), supporting the lateralized SII contribution to SA2. pIns is also involved in somatosensory processes (Yeo et al., 2011; Rachidi et al., 2021). Studies have shown nociceptive-specific pIns activation by somatosensory afferent (Craig et al., 2000; Mazzola et al., 2012), suggesting the role of pIns in somatosensory perception (Oertel et al., 2012). Moreover, our previous study demonstrated that REAL inhibited pIns functional connectivity with default mode network and reduced pain level of low back pain patients (Lee et al., 2019).

Moreover, we found that the cardiac response was associated with the decoupling of somatosensory afferent process. Strong coupling between SA1 and SA2 was associated with a greater decrease in HR. Studies with acupuncture stimulation showed parasympathetic cardiac response during stimulations (Nishijo et al., 1997; Uchida et al., 2018; Nakahara et al., 2021), which could be induced by neural circuitry under the brain stem (Guyenet, 2006; Liu et al., 2020). Neuroimaging studies have reported cardiac associations with brain regions of SA1 and SA2, including aIns (Guo et al., 2016), PAG, and pIns (Napadow et al., 2008; Beissner et al., 2013), which suggests that somatosensory perception and reaction are related to the parasympathetic cardiac response. However, causal relationship between SA1/SA2 and the cardiac response was not clearly elucidated in this study.

Again, the thresholded spatial map of CA2 includes V5/MT+, SI, SII, PL, and PCC for bilateral hemispheres, which are involved in dorsal attention network and somatosensory associating processes. V5/MT+ is essential for visual motion perception (Newsome and Paré, 1988; Rees et al., 2000). SI and SII play roles in somatosensory representations for coding and perception (Lamp et al., 2018; Romo and Rossi-Pool, 2020). Studies have shown activation in SI and SII in visual-induced vicarious sensation (Keysers et al., 2010; Schaefer et al., 2012; Bolognini et al., 2014; Makary et al., 2018), which is accompanied by body ownership (Botvinick and Cohen, 1998) or social mirror mechanisms (Keysers et al., 2010). Moreover, brain spatial parcellation with 7 network estimations showed that CA2 included dorsal attention network for PL and PCC, supported CA2 association with vicarious sensation (Vossel et al., 2014; Machner et al., 2021).

CA1 includes dlPFC, IFG, MCC, SII, insular cortex, and supramarginal gyrus, which involves multiple domains of brain function with fronto-pariatal network. The spatial parcellation map using the 7-network demonstrated that IFG, supramarginal gyrus, and insular cortex corresponded to ventral attention network, which is implicated in stimulus-driven attentional control (e.g., reorienting and filtering) (Vossel et al., 2014). MCC corresponds to dorsal attention network, which is involved in top-down orienting of attention (Vossel et al., 2014). Moreover, MCC showed activation with pain for observing (Zhao et al., 2020; Ardizzi et al., 2021) and perception (Schwedt et al., 2014; Bi et al., 2021), and coupling with somatosensory regions (Oane et al., 2020). dlPFC is crucial for behavioral execution process in fronto-pariatal network (Yamagata et al., 2012; Cieslik et al., 2013; Rozzi and Fogassi, 2017). For instance, placebo mechanism is implicated in the top-down execution of dlPFC (Wager and Atlas, 2015; Schenk and Colloca, 2019). Furthermore, CA1 response to PHNT was positively correlated with the expectation for treatment, which is essential for initiating and maintaining placebo effect (Price et al., 2008). Thus, the spatial map of CA1 indicated that the co-working of multiple domains of brain function led top-down execution.

The coupling of CA1 and CA2 was positively correlated with HR decrease after the 4-s stimulation periods for both REAL and PHNT. Our previous studies with PHNT also showed the delayed HR decrease for 6 s (Lee et al., 2014; Makary et al., 2018), suggesting an effect of the top-down execution process on the delayed HR decrease. Moreover, the coupling between CA1 and CA2 for PHNT showed positive correlation with somatosensory sensation (i.e., vicarious sensation) (Figure 5C). Thus, interaction between fronto-pariatal and dorsal systems may lead to the establishment of a vicarious sensation in PHNT (Corbetta et al., 2008). Furthermore, CA1 showed positive correlation with the expectation (general belief for acupuncture treatment) for PHNT, indicating the establishments of the needling credibility and the vicarious sensation were associated with the expectation for acupuncture.

REAL induced HR decrease during the stimulation and the delayed period. While, PHNT induced HR decreases only for the delayed period, indicating that the contextual manipulation could lead to the delayed HR decreases. Moreover, the delayed HR decreases in REAL and PHNT were associated with CA1, suggesting effects of the top-down execution process on the delayed HR decrease. However, the HR decreases during stimulation in REAL were associated with coupling between somatosensory representative and interoceptive attentional processes. Thus, two different brain mechanisms could affect the heart rate decreases in acupuncture treatment.

Again, we found different temporal characteristics among the components. The peak timing of the optimized HRF indicated that the activation of SA2 and CA1 was more sustained or delayed than that of SA1. Studies have shown peak timing differences of the optimized HRF according to cortical and subcortical regions (Steffener et al., 2010; Lewis et al., 2018; Sclocco et al., 2018). Moreover, the peak latency of BOLD response showed differences between brain regions for somatosensory afferent and affective processes regarding tactile pattern discrimination processes (Hegner et al., 2017), supporting the peak timing delay of CA1 compared with SA1. Furthermore, somatosensory stimulation induced stronger aIns and SMA activation during onset timing than sustained phases, whereas activations in SII and pIns remained during sustained phases (Hu et al., 2015). Thus, BOLD responses for SII and pIns were longer than those for aIns and SMA, supporting the late peak timing of SA2 compared to SA1.

Finally, partial correlation analysis revealed intrinsic interactions of SA and CA components in REAL. The partial correlation network of whole pairs of components demonstrated that SA1 and SA2 were not directly related, however, CA1 linked between SA1 and SA2 (Figure 6). These relationships suggested a bi-directional interaction between cognitive/affective and somatosensory process. While, CA2 for goal-directed attention showed no relationship with other components, indicating that the cognitive-somatosensory interaction in REAL differed from goal-directed vicarious sensation mechanism in PHNT. Moreover, we found direct expectation effects (general belief for acupuncture treatment) on CA1 and SA1 using partial correlation analysis (Figure 6 and Table 3). Higher expectation induced weak CA1 responses but strong SA1 responses, which was distinct from the positive expectation effect on CA1 in PHNT. This specific phenomenon may be associated with a characteristic of acupuncture, which induces interoceptive attention for somatosensory input (Craig, 2009; van Ede et al., 2014; Kuehn et al., 2016).

This study has some limitations. Regarding the experimental aspect, a protocol for cognitive/affective dissociation from acupuncture treatment was not incorporated in this study. Our results with IC-wise GLM showed that the four separated ICs, including CA1 and CA2, for cognitive/affective process; however, we could not assess somatosensory afferent process without cognitive/affective interaction.

In conclusion, we identified the four separated ICs induced by compound stimulus of acupuncture. Brain correlates for somatosensory afference were separated into two ICs for interoceptive attention/motor reaction (SA1) and somatosensory representation (SA2) process. Coupling between SA1 and SA2 was associated with strong HR decrease during stimulation. Brain correlates for cognitive/affective modulation were separated into two ICs for executive control (CA1) and goal-directed sensory process (CA2). CA1 is associated with acupuncture sensation and delayed HR decrease. Moreover, partial correlation network with these components suggested a bi-directional interaction between cognitive/affective and somatosensory afferent process, which differed from vicarious sensation mechanism in PHNT. Furthermore, the expectation for the treatment negatively affected CA1 but positively affected SA1 in REAL, indicating a specific bottom-up process associating interoceptive attention for acupuncture treatment. Our findings on brain interactions in acupuncture treatment elucidated the underlying mechanisms of interaction between somatosensory afference and therapeutic contextual manipulation, which might be a specific response for acupuncture.

The data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by KHNMC-OH-IRB 2009-006. The participants provided their written informed consent to participate in this study.

KP conceptualized the study design. CJ, JK, and KP collected the data. CJ and KP analyzed and interpreted the data. CJ wrote the first draft of the manuscript. All authors contributed to critically revising the manuscript and approved the final manuscript.