The relationship between BOLD signal and
autonomic nervous system functions: Implications
for processing of “physiological noise”
Vittorio Iacovella1,* and Uri Hasson1,2
1Center for Mind/Brain Sciences (CIMeC), The University of Trento, Italy
2Faculty of Cognitive Science, The University of Trento, Italy
Functional magnetic resonance imaging (fMRI) research has revealed not only important aspects of the neural basis of cognitive
and perceptual functions, but also important information on the relation between high-level brain functions and physiology. One
of the central outstanding questions, given the features of the Blood Oxygenation Level-Dependent (BOLD) signal, is whether
and how autonomic nervous system (ANS) functions are related to changes in brain states as measured in the human brain. A
straightforward way to address this question has been to acquire external measurements of ANS activity such as cardiac and
respiratory data, and examine their relation to the BOLD signal. In this work we describe two conceptual approaches to the
treatment of ANS measures in the context of BOLD fMRI analysis. On the one hand, several research lines have treated ANS
activity measures as noise, considering them as nothing but a confounding factor that reduces the power of fMRI analysis
or its validity. Work in this line has developed powerful methods to remove ANS effects from the BOLD signal. On the other
hand, a different line of work has made important progress in showing that ANS functions such as cardiac pulsation, heart
rate variability and breathing rate, could be considered as a theoretically meaningful component of the signal that is useful for
understanding brain function. Work within this latter framework suggests that caution should be exercised when employing
procedures to remove correlations between BOLD data and physiological measures. We discuss these two positions and the
reasoning underlying them. Following, we draw on the reviewed literature in presenting practical guidelines for treatment of
ANS data, which are based on the premise that ANS data should be considered as theoretically meaningful information. This
holds particularly when studying cortical systems involved in regulation, monitoring and/or generation of ANS activity, such as
those involved in decision making, conﬂict resolution and the experience of emotion.
This is a pre-publication version of a manuscript published in Magnetic Resonance Imaging Volume 29, Issue 10 (2011),
https://doi.org/10.1016/j.mri.2011.03.006. It contains minor stylistic deviations from the version of record, and anticipates our
later empirical work on these issues, https://doi.org/10.1016/j.neuropsychologia.2018.03.005
BOLD functional magnetic resonance imaging (fMRI) is one the most powerful and popular methods for non-invasive
examination of whole-brain and regional neural activity patterns. However, the relationship between BOLD ﬂuctuation and pre-
or post-synaptic activity at the neural level is non-trivial (Logothetis 2003). BOLD ﬂuctuations in a given area are induced not
only by changes in metabolic demand related to neural activity, but also by a wide range of factors that are considered artifacts
such as system noise, hardware limitations (Renvall and Hari 2009) and confounding effects such as participant’s movements
inside the scanner. In particular, physiological factors mediating the transfer function between neural and BOLD ﬂuctuations
(Logothetis 2008) play an important and potentially confounding factor in the interpretation of BOLD data. Determining the
impact of these various noise sources on the BOLD signal is an active domain of research (e.g., Bianciardi, Fukunaga et al.
One of the main external factors known to co-vary with BOLD measurements is a set of physiological parameters reﬂecting
activity in the autonomic nervous system (ANS). The ANS is a part of the nervous system that controls functions such as
perspiration, respiration, heart rate and blood pressure. It is well documented that ﬂuctuations in these functions are correlated
with the BOLD signal (Lund, Madsen et al. 2006) and that the impact of such factors may increase in higher ﬁeld strengths,
rendering measurements performed at higher strength marginally advantageous unless this factor is adequately controlled for
(Triantafyllou, Hoge et al. 2005).
Given the sensitivity of BOLD to ANS functions, one of the important technical and theoretical questions facing scientists
relying on BOLD measures is to understand how ANS functions are related to changes in the BOLD signal. To date, this issue
has been largely addressed by acquiring external measurements of ANS states and examining their relation to the BOLD signal.
With progress in the ﬁeld, it appears that two implicit approaches to the treatment of ANS data have been developed. On the one
hand, studies employing an “ANS as noise” approach have mainly treated physiological acquisitions as artifacts, considering
them, from a cognitive or theoretical point of view, as nothing but a confounding factor that reduces the power of an analysis or
its validity. Work in this line has developed powerful methods to remove ANS effects from the BOLD signal. On the other hand,
other work within various neuroscience domains has made important progress in showing that ANS functions such as cardiac
pulsation and breathing rate could be considered as cognitively relevant and interesting components of the BOLD signal. Our
aim in this review is to track the rationale for these approaches and derive pragmatic conclusions based on these considerations.
BOLD: a product of a complex transfer function
The BOLD signal is intrinsically a metabolic effect since it is not directly related to electrophysiological changes induced by
neuronal populations. Evidence for its biophysical nature was originally shown in PET studies documenting an uncoupling
between cerebral blood ﬂow (CBF) and a metabolic quantity of the oxygen consumption (CMRO2; Fox and Raichle 1986).
Quantitative models characterizing aspects of the BOLD effect were developed on the basis of metabolic variables (e.g. the
Balloon Model, Buxton, Wong et al. 1998). The transfer function thought to mediate between neural activity and BOLD has
been termed “hemodynamic response” function (HRF), to underline both its mechanical and hydrodynamic aspects. Numerous
following studies explored factors affecting vascular aspects of the BOLD signal, using methods such as the examination
of blood ﬂow and volume variations induced by natural or carbon dioxide calibrated respiration. For example, controlled
breathing studies have shown the unique impact of vasodilation on the BOLD response (Davis, Kwong et al. 1998; Wise, Ide et
al. 2004). Other work (Lu, Zhao et al. 2008) has shown that factors such as baseline CO2 levels determine the dynamic range
of BOLD and CBF responses to stimulations. One of the interesting points raised by such work is the fact that the asymmetric
coupling between CBF and CBV is spatially heterogeneous: this leads to different rates of change in BOLD ﬂuctuations across
brain regions. Moreover, inter-individual differences could be well accounted for using such parameters given that individuals’
BOLD ﬂuctuations could be related to absolute changes in CO2 partial pressure, and the ﬁnal results could be more accurate
and speciﬁc (Murphy, Harris et al. 2010). In all, these studies have served useful in highlighting how breathing induces changes
that affect the BOLD signal. In tandem with studies that have used breathing challenges as a method for understanding the
relation between BOLD and physiological factors, other work had demonstrated that although BOLD changes are related to
ﬂuctuations in neural activity (Logothetis, Pauls et al. 2001; Shmuel, Augath et al. 2006), neural activity is not the only factor
that causes BOLD ﬂuctuations. In particular, it was recognized that physiology may be one factor affecting ongoing changes
in regional metabolism, given that continuous ANS functions like respiration and cardiac rate trends are tightly linked to the
process of oxygenation.
Physiology as noise
There are a number of ways by which physiological processes can impact BOLD measures, and some of these induce ﬂuctuations
that should be treated as noise. For instance, respiration processes may induce systematic motion patterns that reduce the
accuracy of BOLD analyses and necessitate correction (Noll and Schneider 1994). One type of movement effect is driven
by changes in blood pressure through major vessels, which induces movement in nearby tissue (Dagli 1999). Respiration
induced effects can translate into head motion (Glover, Li et al. 2000) but also result in variations in oxygen concentration and
susceptibility effects induced due to the breathing process (Van de Moortele, Pfeuffer et al. 2002). Given such impact, work
treating ANS correlates as noise has developed analysis methods that treat ANS measurements similarly to how motion data are
treated during fMRI preprocessing. A number of approaches have been examined, including the removal of ANS covariates in
K-space or during retrospective processing in the image domain (Glover, Li et al. 2000), or the joint removal of physiological
effects in tandem with head motion (Jones, Bandettini et al. 2008). Other work has examined related covariates, including
heart-rate variance and respiration variance (Chang and Glover 2009), as well as the effectiveness of various models combining
these factors in the context of correction (Jo, Saad et al. 2010).
In practice, a common general method to link physiological time-courses with fMRI data is the one in which the phase
of the physiological time-course is used to produce Fourier expansions of low order. The relation of these expansions with
the BOLD data is evaluated using linear ﬁt against the fMRI time-series. When considering physiological effects as a source
of noise, the image-based method for retrospective correction of physiological motion effects (RETROICOR, Glover, Li et
al. 2000) is commonly utilized to assess the amount of variance in the fMRI data explained by ANS measures. Given that
ANS recordings have a higher temporal resolution than BOLD (e.g., 50Hz) some researchers have opted to undersample the
measures to the timing of the acquisition TR (van Buuren, Gladwin et al. 2009), whereas others account for ANS effects at the
single-slice level, which is acquired at a rate of [TR / N(slices)] Hz.
In absence of physiological recording, several researchers have examined the use of proxy measures derived from the BOLD
signal itself. The particular issue of whether physiological effects are aptly summarized in the “global mean” of the BOLD
signal and whether the global mean can be used as a proxy for physiological parameters has been addressed in several studies
(Macey, Macey et al. 2004; Fox, Zhang et al. 2009; Sch
olvinck, Maier et al. 2010). Furthermore a direct comparison, by means
of correlation coefﬁcients, between physiological acquisitions and global mean signal showed that for certain participants
it was possible to ﬁnd a reliable correlation between the global mean of the BOLD response and physiological responses
(Chang and Glover 2009). However, we note that interpreting the reliability of correlations in this context is complicated by
the fact that both the global mean time series and the physiological data themselves demonstrate strong serial autocorrelation,
which makes it difﬁcult to assign reliability values to correlations between the two variables, and this is a topic that should
be further addressed. The issue of global mean removal has also been examined by several studies, which suggest it is best
to not partial out this property for various analytic reasons (Murphy, Birn et al. 2009), and that at minimum, global mean
removal should be employed only in those cases where it does not create interpretive confounds. From a theoretical perspective,
recent work (Hyder and Rothman 2010; Sch
olvinck, Maier et al. 2010), has examined the neuronal correlates of global BOLD
ﬂuctuations, by combining local ﬁeld potentials derived from single-cell recordings and fMRI acquisitions. This work has
shown that correlations between gamma frequency (40-80 Hz) of the LFP spectrum and the fMRI signal are observed over
large parts of the cortex. This correlation was found to be signiﬁcant for neural events recorded in diverse brain areas. The
fact that about 10% of the fMRI variance could be related to these transient-induced LFP changes suggests that global signal
contains cognitively relevant information, and as such should not be removed from the signal. One avenue for future work on
removal of physiological effects from the BOLD signal is the development of data driven methods such as ICA to identify
components in the BOLD signal whose properties match that of physiological time series (Soldati, Robinson et al. 2009).
The importance of accounting for potential physiological effects on BOLD data has increased with the theoretical focus
on understanding resting state processes in the human and primate brain and their relation to cognition. Since many of the
fMRI methods for studying the resting state depend on quantifying patterns of spatially synchronized ﬂuctuations in the
BOLD response, any factor that may drive correlated BOLD ﬂuctuations should be corrected for in order to determine that the
patterns of spatial correlation identiﬁed are indeed driven by synchronized neural sources. This is particularly a concern since
physiological measures induce correlation between different brain regions (Chang and Glover 2009), and given that motion
may induce similar patterns of activity that affect the results of clustering solutions (Mezer, Yovel et al. 2009) in the resting
state (Jones, Bandettini et al. 2008). That is, dissociating between “functional connectivity” and “non-functional connectivity”
is becoming an important domain of inquiry. A number of studies have examined the relation between physiological processes,
cortical activity, and connectivity amongst various cortical networks.
The relation between respiration effects and the BOLD signal has been evaluated in different ways: modeled as lagged
boxcar functions (Kastrup, Li et al. 1999), or considered as factors to be convolved with an HRF in a block design context
(Thomason, Burrows et al. 2005). Work by Birn and colleagues examined several interesting aspects of respiration induced
effects on the BOLD signal: they started from the observation that variations in respiration rate occur at frequencies (about
0.03 Hz) similar to those related to resting state connectivity (Biswal, Zerrin Yetkin et al. 1995). They showed that the
lagged envelope of the respiration time-courses, termed the Respiration Volume per Time (RVT), is an important factor in
explaining BOLD ﬂuctuations, both at rest and during a cognitive task (Birn, Diamond et al. 2006). Furthermore, the removal of
physiological effects reduced the standard deviation of the signal, suggesting an improvement in data quality. The interrelation
between RVT and brain function was examined by probing for correlations between BOLD time-series and RVT lagged
regressors obtaining signiﬁcant values at hemodynamically relevant time lags (about 8 seconds). An extension of this framework
lead to the development of a respiration response function (RRF) (Birn, Smith et al. 2008), which was shown to be a valid basis
function for respiration induced ﬂuctuations across the brain.
Cardiac effects on BOLD have also been examined. Heart rate and heart rate variability over time have been incorporated
into the RETROICOR framework to take in account non-cyclic effects explained by cardio-respiratory phases, and the impact
of these factors was evaluated using connectivity studies (van Buuren, Gladwin et al. 2009). Regressors for non-cyclic effects
were constructed lagging physiological acquisitions and choosing the shifts that maximized the correlation with BOLD signal.
Additional variance (about 5% more than explained by the initial RETROICOR procedure) was found to be explained by
non-cyclic effects, including RVT. Importantly, functional connectivity during rest was still reliable after the removal of these
effects suggesting that even after physiologically-induced correlations are removed, some correlation is indeed related to neural
activity per se.
Chang and colleagues (Chang, Cunningham et al. 2009) aimed to derive a cardiac response function through a deconvolution
of the fMRI data. They showed that their model accounted for more variance (taking into account the additional parameter)
in about 1/3 of the brain voxels at rest. This improvement was also spatially heterogeneous; it produced an increase of the
connectivity within nodes of the default mode network, and a decrease outside it.
The speciﬁc issue of the spatial heterogeneity of physiological effects was also examined in several other studies. Saad et al.
(Saad, Glen et al. 2009) demonstrated that due to the spatial heterogeneity of physiological effects, masks of particular brain
areas including white matter and cerebrospinal ﬂuid may be constructed from structural images, so that the mean time-series
from those masks can be used as additional explanatory variables when accounting for physiological noise. A related study
demonstrated the use of a completely image-based strategies to remove variance related to global effects (Giove, Gili et al.
2009), and this was shown to improve the data quality of the signal. This particular attention to tissue features suggests that
correction of ANS effects could be better performed when considering the particular anatomical context of various brain
regions. The studies we described explored the relationship between autonomic indicants and neural activity by examining
correlations between ANS measures and the BOLD response, and their relation to connectivity patterns or BOLD data quality.
These relations have also been examined outside the domain of BOLD response. For example, optical topography has been
used as non-invasive imaging technique to directly evaluation hemoglobin concentrations (HbCC). Katura and colleagues
acquired optical topography data as well as two autonomic indicants (heart rate and mean arterial blood pressure) at rest
to examine the relationship between cardiovascular dynamics and low-frequency HbCC ﬂuctuations in the cortex (Katura,
Tanaka et al. 2006). They evaluated whether there exists a common low-frequency band where oscillations occur both for the
hemodynamics and autonomic functions. They identiﬁed a frequency band at around 0.1 Hz that could be distinguished within
hemodynamic behavior from other ﬂuctuation components. Subsequent work showed that this relation could be modulated
pharmacologically (Obrig, Neufang et al. 2000). Using transfer entropy as a measure of interrelation between HbCC, heart rate
and mean arterial pressure, Katura et al. also showed that the low-frequency ﬂuctuations in cerebral oxy- and deoxy-hemoglobin
could be associated, in a signiﬁcant fraction (up to 35%), with ANS indicants. This suggests that when analyzing BOLD
signal ﬂuctuations in this frequency band, ANS inﬂuences could be considered as a signiﬁcant source of ﬂuctuations. Tong and
Frederick (2010) extended the investigation of low-frequency ﬂuctuations using near infrared spectroscopy acquisitions (NIRS)
combined with fMRI. NIRS extends the information obtainable from fMRI since the former has high temporal resolution (12.5
Hz). The authors observed a correlation between low-frequency oscillations of the NIRS signal and the BOLD responses widely
distributed throughout the brain. This correlation, performed between BOLD and down-sampled, multi-lagged versions of the
spectroscopy time-series, showed a meaningful temporal pattern as well. According to the distributions of correlation in time
(about 6s to travel across the whole brain) and in the space (their arrangement resembled main blood vessels maps), the authors
were able to identify the presence of heterogeneous sources for low-frequency ﬂuctuations, including non-neural nature.
Work using fMRI has attempted addressing this issue as well. The BOLD signal itself is related to the ﬂuctuations in HbCC,
and the blood volume itself could be related to blood pressure effects. Shmueli and colleagues (Shmueli, van Gelderen et
al. 2007) studied the relationship between cardiac rate variations time-courses and BOLD signal at rest, using correlations
and regressions. The ANS indicant they examined was heart rate variability (HRV), constructed from the cardiac pulsation
time-courses. The spectrum of this parameter was shown to be higher in the relevant (about 0.1 Hz) range of frequency. Adding
cardiac rate regressors to a RETROICOR framework lead increased variance explained. Furthermore, the spatial distribution
of the correlations with the cardiac rate variations was not localized around large vessels (as commonly occurs for cardiac
pulsations and respiration time-courses). These results may suggest that cardiac rate variations induce their effects in ways that
are not mediated by motion or susceptibility-change effects, and should be considered when one wants to signiﬁcantly take in
account for autonomic nervous system effects on the ﬂuctuations of the BOLD signal.
The role of the autonomic nervous system indicants in cortical activity
While the above-reviewed studies aim to characterize the variance explained by various physiological quantities and remove
their effect from the BOLD data, another body of work has approached the analysis of ANS indices from a completely
different approach, treating these as meaningful signals. This line of work has focused on the neural and behavioral correlates
of autonomic indicants in order to understand which brain regions are involved in triggering, modulating or monitoring
ANS functions. From the perspective of such studies, the BOLD variance explained by physiological signal is a potentially
meaningful component whose properties should be examined.
The idea that bodily states are monitored by the brain and processed at different levels of conscious experience has an
established history in the study of psychology and cognitive neuroscience. Both cortical and subcortical structures are known
to monitor physiological states such as breathing rate and cardiac responses (Evans 2010). These monitoring and control
operations occur largely on an unconscious level. However, different lines of theoretical and experimental work have also
developed the notion that ongoing experience is partly determined by a person’s awareness or monitoring of his or her bodily
states. One of the earliest and stronger forms of this approach postulated that the emotional aspect of human experience is
actually generated as a result of monitoring visceral responses (see for example Lang 1994). This strong form of the hypothesis
was later strongly criticized on various grounds (Cannon 1987), including the fact that severing connections from internal organs
(viscera) to the nervous system still allows the experience of emotions, and that visceral responses may be triggered after an
emotion is experienced. Recent lesion-based dissociations support this line of criticism, suggesting that the emotion-experience
and autonomic responses evoked by emotional stimulus are associated with different systems and are independent of each other
(Johnsen, Tranel et al. 2009).
Nonetheless, the notion that bodily states, including those triggered by the ANS, play a functional role in human experience
at different levels has been continuously inﬂuential in the cognitive neurosciences. The somatic marker hypothesis developed
by Bechara and colleagues (Bechara, Damasio et al. 2000) is a theory of human decision making that holds that decisions
are “inﬂuenced by marker signals that arise in bioregulatory processes, including those that express themselves in emotions
and feelings”. This approach conjointly holds that bioregulatory processes affect or “colour” ongoing experience, as well as
decision making processes. This hypothesis is based on several types of empirical studies, most generally those suggesting that
that risky decision making has not only cognitive components but also emotional ones. For instance, whereas non-clinical
participants generate skin conductance responses before making an uncertain choice, patients with ventromedial damage do
not generate this response. This line of work does not directly address the issue of directionality – i.e., is the relation between
ventromedial activity and the ANS response due to an afferent track by which the ventromedial cortex drives ANS activity, or
to an efferent track by which the region monitors the ANS state. In addition, given the heterogeneity of this frontal region, it is
possible that it is involved in both generation and monitoring of ANS responses.
Subsequent work has documented that instances of conﬂict resolution, such as those evoked by conﬂict states (e.g., a
Stroop task) similarly induce changes in skin conductance, suggesting that cognitive conﬂict is associated with ANS changes
(Kobayashi, Yoshino et al. 2007). This has also been documented for cognitive conﬂict occurring during high level mental
processes such as reasoning, which was shown to induce changes in ANS activity (De Neys, Moyens et al. 2010). These
ﬁndings are particularly important given the well established association between one particular ventromedial region, the ACC,
and cognitive conﬂict. To the extent that the ACC is involved in monitoring of somatic states or their generation, such ﬁndings
could pose an interesting explanation for some ACC activation patterns during conﬂict.
Consistent with this body of theoretical and experimental work, a body of neuroimaging work to date has attempted to
identify neural structures that drive or monitor ANS responses, with particular interest in cortical structures. An important body
of work in the area was pioneered by Critchley and colleagues whose work examines the links between ANS measures and
neural function. This work merges classical emotional theories (see for example, Harrison, Gray et al. 2010) with physiological
and neuroanatomical concepts (see, e.g., Craig 2002). In a recent work (Harrison, Gray et al. 2010), cortical correlates of
emotional feelings were studied, using fMRI and gastric electrogastrogram acquisitions. Participants were instructed to watch
and judge videos where several kinds of feelings of disgust were elicited. Changes in ANS indicants were reﬂected in the
activity of the insula: this area showed correlated activity with increased high frequency heart power. This suggests that the
insula could be involved in the representation of afferent information from the ANS towards the cortex (see also, Critchley
2005). Brain activity increases in the insula and in the inferior parietal lobule were shown to be related to the accuracy in
counting heart beats (Pollatos, Schandry et al. 2007), suggesting the area mediates attention to one’s own arousal.
Other support for high-level modulation of the relationship between brain function and ANS was examined in a study by
Metz-Lutz and colleagues. In that study fMRI and ECG were acquired to explore human adhesion during movie-watching (the
belief that what is shown is real). They studied changes in HRV using a lagged-autocorrelation index (correlation coefﬁcient in
e plots) to explore at what points do heart rate changes occur. They found that decreases in heart-rate variability occurred
in the same time as the activation of several brain regions, including Brodmann areas 21/22 and 37/39. From a physiological
perspective, this exempliﬁes how naturally occurring emotions during movie viewing are associated with ANS indicants such
as heart rate variability. From a technical viewpoint, such a ﬁnding highlights the important of considering this relation during
data analysis, since the link between this ANS measure and cognitive activity implies that procedures that remove the effect of
hear rate variability from BOLD signal could well remove meaningful components of the BOLD signal as well.
The relationship between ANS indicants and BOLD has also been examined outside the domain of high-level cognitive
function. In an early study (Critchley, Corﬁeld et al. 2000) PET data were collected while individuals’ blood pressure (mean
arterial pressure) and heart rate were monitored. Participants in that study were instructed to perform either effortless or
effortful variants of two tasks: an isometric (squeezing a bulb) exercise or a mental arithmetic (serial subtraction) stressor task.
Conjunction analyses were performed to explore areas common to both the tasks, revealing activity in subcortical structures
including the cerebellar vermis and brainstem, but also in the right anterior cingulate (ACC). These same regions and the right
insula were shown to covary on a group level with mean arterial pressure. Finally, activity in the pons, cerebellum, and right
insula was associated with heart rate in the same way.
Follow up work used fMRI and electrocardiographic acquisitions (Critchley, Mathias et al. 2003). In this study, cardiac time-
courses were processed to separate low (0.05-0.15 Hz) and high (0.15-0.50 Hz) frequency components that refer, respectively,
to sympathetic and para-sympathetic neural inﬂuences on heart rate (Montano, Porta et al. 2001). The tasks utilized were
similar to those used by Critchley et al. (2000). The study found a signiﬁcant correlation between the low frequency cardiac
component and anterior cingulate activity. The authors interpreted this ﬁnding as showing an involvement of the anterior
cingulate cortex in modulation of cardiac function. Importantly, the same study demonstrated the patients with a lesion in
the ACC performed well on both tasks but did not show the same sort of ANS modulation. Instead, they showed relatively
reduced cardiovascular responses to effortful cognitive tasks. This was taken to suggest that that the anterior cingulate plays an
important role in regulation of bodily states of arousal, to meet concurrent behavioral demands.
The study of the relationship between brain activity and ANS indicants could be performed by modeling their relation to
brain activity as mediated by high-level cognitive functions. Wager and colleagues (Wager, Waugh et al. 2009), demonstrated
that social evaluative threat was correlated with cardiovascular responses. In that study, they combined fMRI and physiological
monitoring and presented participants stimuli eliciting this kind of threat in a block-design study. A conjunction analysis
identifying regions correlating with both the stimulus and heart rate, identiﬁed a set of brain regions including the pregenual
ACC, orbitofrontal cortices, and the putamen. An analysis of the information pathways between these three areas using
mediation analysis showed that they play independent roles in regulating the relationship between activity patterns related to a
task’s cognitive demand and those activity patterns related to ANS indicants such as heart rate. The non-artifactual nature of
these ﬁndings was supported by the strong spatial localization of the effect, which stands in contrast to the more wide-spread
correlates of physiological ﬂuctuations. Based on the same data, other work (Wager, van Ast et al. 2009) underlined the
importance of prefrontal and subcortical systems as well: ventromedial prefrontal cortex and rostral ACC were shown to be
involved in the relationship between brain function and heart rate, through the mediation of periacqueductal gray.
Another study (Gray, Rylander et al. 2009) explored the interrelations between ANS and neural activity using sensorial
mechanisms. In that fMRI study, several autonomic measures were acquired during the scan: electrocardiogram (ECG), heart
rate and mean arterial pressure data. Participants were instructed to wait for slight un-painful electrical shocks. These shocks
were given either synchronously with or delayed with respect to the cardiac peaks (R-wave, dominant peak in ECG cycle).
This particular experimental setup was designed to examine whether stimulus processing would depend on a person’s ongoing
visceral state. Analyzing ANS indicants, they found that mean arterial pressure increases when shocks were given synchronously
with the ECG R-wave. BOLD signal was analyzing using regressors constructed by convolving the timing of synchronized and
unsynchronized shocks with a hemodynamic response function. It was shown that the anterior insula, amygdala and brainstem
showed different responses to synchronous and delayed shocks. Speciﬁcally, the left (and, statistically less signiﬁcant, the
right) anterior insula and the mid pons showed greater activity for synchronous shocks, and the right amygdale showed weaker
activity. This suggests that there exist a set of brain regions that support the integration of somatosensory information with
cardiovascular feedback to control autonomic arousal.
The spatial localization of the effects we previously described could be referred to the cortical back-ends of the interoceptive
information pathways on the cortex (see, Critchley 2005, for a useful schema). The afferent information from the peripheral
nervous system towards the central nervous system (Craig 2002) passes through thalamic nuclei, and then towards several
cortical areas: the insula, the anterior cingulate cortex, the orbitofrontal cortices, and terminates in the right anterior insula
(Critchley 2005; Craig 2009). The relationship between ANS and brain function emerge as a crucial factor in the BOLD
ﬂuctuations in these areas.
In this work we have presented two views of the functional role of ANS indices in the context of the fMRI BOLD analysis.
On the one hand, these indicants are sometimes treated purely as physiological noise. When treated in this way, the aim
of researchers is to remove the effect of these indices from the BOLD signal. As reviewed, a number of co-variates have
been explored in this context, including the original cardiac and respiratory recordings, their harmonic expansions, under-
sampled derivatives matched to the temporal resolution of the TR, and more complex derivatives capturing heart rate variance
and respiration variance over time. Given the relation of such measures to purely nuisance variables such as head motion,
cardiac-induced tissue motion and changes in CO2 concentration, this approach is clearly justiﬁed.
However, the importance of work showing the interesting functional relation between ANS indices and activity in both
cortical and subcortical regions should be considered as well. From this perspective, brain regions that play a meaningful
functional role in driving or monitoring ANS activity will show activation patterns that correlate with the BOLD signal. From
this perspective, technical procedures that aim to remove the effect of ANS indices from the BOLD signal will also remove a
meaningful variance component not only when studying processes associated with these regions, but when studying any process
in which there might be a confound between cognitive processing and ANS states. The impact of “physiological cleaning” can
therefore be particularly detrimental in various experimental contexts. These may include e.g., (a) manipulations of cognitive
conﬂict that are associated with ANS activity, (b) studies that explicitly use emotional stimuli, (c) studies of decision making
under uncertainty, (d) naturalistic viewing of engaging ecological stimuli such as movies that induce ﬂuctuations in ANS
activity over time, and (e) studies evoking startle or surprise responses. Furthermore, as we have outlined, the study of the
relation between the ANS and cortical function is an important topic being currently developed and describing the frequencies
driving such connectivity, the pathways of information, and the way in which the link between ANS activity and cortical activity
can be manipulated are central questions in this domain. Clearly, none of these questions can be answered if ANS indicants
were removed from the signal.
Addressing the role of the Anterior Cingulate Cortex (ACC) and its function during rest is related to both these viewpoints.
On the one hand, the ACC is a central node in the “Default Mode Network”, which is considered to be one of the dominant
resting state networks (Buckner, Andrews-Hanna et al. 2008). Functional connectivity studies examining this region have
treated ANS indicants as external factors that represent physiologically-confounding data. On the other hand, the literature
we reviewed suggests that the ACC is one of the most important brain mediators between sympathetic information and brain
activity, as was shown from studies of clinical and non-clinical populations. For this reason, it could be that the ACC mediates
unique functions not shared with the rest of the default mode network. For this reason, removing variance attributed by
physiological measurements from this region could result in removing a unique component of variance, thus leading to an
incomplete theoretical description of the ACC’s role. For instance, it could be that the ACC is partly related to other regions
mediating ANS functions, and this relation will not be found after the removal of ANS correlates.
On the basis of these considerations we suggest that researchers carefully evaluate whether their particular research question
unambiguously justiﬁes the removal of ANS indicants from the signal. While it could be that in the near future methods will
be available that will allow limiting the correction for ANS ﬂuctuations to particular brain regions, to our knowledge such
procedures have not been validated as of yet. For this reason, if the theoretical question being examined may have a functional
link to ANS activity, it may be pertinent to analyze the data with or without the removal of ANS effects. In our opinion,
increased synergy between research examining cortical systems mediating ANS function, and technical work aiming to identify
ANS-BOLD correlates holds the potential to advance research in both domains.
Bechara, A., H. Damasio, et al. (2000). ”Emotion, Decision Making and the Orbitofrontal Cortex.” Cerebral Cortex 10(3):
Bianciardi, M., M. Fukunaga, et al. (2009). ”Sources of functional magnetic resonance imaging signal ﬂuctuations in the
human brain at rest: a 7 T study.” Magnetic resonance imaging 27(8): 1019-1029.
Birn, R., J. Diamond, et al. (2006). ”Separating respiratory-variation-related ﬂuctuations from neuronal-activity-related
ﬂuctuations in fMRI.” NeuroImage 31(4): 1536-1548.
Birn, R., M. Smith, et al. (2008). ”The respiration response function: The temporal dynamics of fMRI signal ﬂuctuations
related to changes in respiration.” NeuroImage 40(2): 644-654.
Biswal, B., F. Zerrin Yetkin, et al. (1995). ”Functional connectivity in the motor cortex of resting human brain using
echo-planar mri.” Magn. Reson. Med. 34(4): 537-541.
Buckner, R. L., J. R. Andrews-Hanna, et al. (2008). ”The Brain’s Default Network.” Annals of the New York Academy of
Sciences 1124(1): 1-38.
Buxton, R. B., E. C. Wong, et al. (1998). ”Dynamics of blood ﬂow and oxygenation changes during brain activation: The
balloon model.” Magn. Reson. Med. 39(6): 855-864.
Cannon, W. B. (1987). ”The James-Lange theory of emotions: a critical examination and an alternative theory. By Walter B.
Cannon, 1927.” The American journal of psychology 100(3-4): 567-586.
Chang, C., J. Cunningham, et al. (2009). ”Inﬂuence of heart rate on the BOLD signal: The cardiac response function.”
NeuroImage 44(3): 857-869.
Chang, C. and G. Glover (2009). ”Effects of model-based physiological noise correction on default mode network
anti-correlations and correlations.” NeuroImage 47(4): 1448-1459.
Craig, A. D. (2002). ”How do you feel? Interoception: the sense of the physiological condition of the body.” Nature
Reviews Neuroscience 3(8): 655-666.
Craig, B. (2009). ”How do you feel–now? The anterior insula and human awareness.” Nature reviews. Neuroscience 10(1):
Critchley, H., C. Mathias, et al. (2003). ”Human cingulate cortex and autonomic control: converging neuroimaging and
clinical evidence.” Brain 126(10): 2139-2152. Critchley, H. D. (2005). ”Neural mechanisms of autonomic, affective, and
cognitive integration.” J. Comp. Neurol. 493(1): 154-166.
Critchley, H. D., D. R. Corﬁeld, et al. (2000). ”Cerebral correlates of autonomic cardiovascular arousal: a functional
neuroimaging investigation in humans.” The Journal of physiology 523 Pt 1(1): 259-270.
Dagli, M. (1999). ”Localization of Cardiac-Induced Signal Change in fMRI.” NeuroImage 9(4): 407-415.
Davis, T., K. Kwong, et al. (1998). ”Calibrated functional MRI: Mapping the dynamics of oxidative metabolism.”
Proceedings of the National Academy of Sciences of the United States of America 95(4): 1834-1839.
De Neys, W., E. Moyens, et al. (2010). ”Feeling we’re biased: Autonomic arousal and reasoning conﬂict.” Cognitive,
Affective, & Behavioral Neuroscience 10(2): 208-216.
Evans, K. (2010). ”Cortico-limbic circuitry and the airways: Insights from functional neuroimaging of respiratory afferents
and efferents.” Biological psychology.
Fox, M., D. Zhang, et al. (2009). ”The Global Signal and Observed Anticorrelated Resting State Brain Networks.” J
Neurophysiol 101(6): 3270-3283.
Fox, P. T. and M. E. Raichle (1986). ”Focal physiological uncoupling of cerebral blood ﬂow and oxidative metabolism
during somatosensory stimulation in human subjects.” Proceedings of the National Academy of Sciences of the United States
of America 83(4): 1140-1144.
Giove, F., T. Gili, et al. (2009). ”Images-based suppression of unwanted global signals in resting-state functional connectivity
studies.” Magnetic resonance imaging 27(8): 1058-1064.
Glover, G. H., T. Q. Li, et al. (2000). ”Image-based method for retrospective correction of physiological motion effects in
fMRI: RETROICOR.” Magn. Reson. Med. 44(1): 162-167.
Gray, M., K. Rylander, et al. (2009). ”Following one’s heart: cardiac rhythms gate central initiation of sympathetic reﬂexes.”
The Journal of neuroscience : the ofﬁcial journal of the Society for Neuroscience 29(6): 1817-1825.
Harrison, N., M. Gray, et al. ”The Embodiment of Emotional Feelings in the Brain.” J. Neurosci. 30(38): 12878-12884.
Hyder, F. and D. Rothman (2010). ”Neuronal correlate of BOLD signal ﬂuctuations at rest: err on the side of the baseline.”
Proceedings of the National Academy of Sciences of the United States of America 107(24): 10773-10774.
Jo, H. J., Z. Saad, et al. (2010). ”Mapping sources of correlation in resting state FMRI, with artifact detection and removal.”
NeuroImage 52(2): 571-582.
Johnsen, E., D. Tranel, et al. (2009). ”A neuroanatomical dissociation for emotion induced by music.” International journal
of psychophysiology 72(1): 24-33.
Jones, T., P. Bandettini, et al. (2008). ”Integration of motion correction and physiological noise regression in fMRI.”
NeuroImage 42(2): 582-590.
Kastrup, A., T.-Q. Li, et al. (1999). ”Cerebral Blood Flow-Related SignalChanges during Breath-Holding.” AJNR Am J
Neuroradiol 20(7): 1233-1238.
Katura, T., N. Tanaka, et al. (2006). ”Quantitative evaluation of interrelations between spontaneous low-frequency
oscillations in cerebral hemodynamics and systemic cardiovascular dynamics.” NeuroImage 31(4): 1592-1600.
Kobayashi, N., A. Yoshino, et al. (2007). ”Autonomic arousal in cognitive conﬂict resolution.” Autonomic Neuroscience
Lang, P. J. (1994). ”The Varieties of Emotional Experience: A Meditation on James-Lange Theory.” Psychological Review
Logothetis, N. (2003). ”The Underpinnings of the BOLD Functional Magnetic Resonance Imaging Signal.” J. Neurosci.
Logothetis, N. (2008). ”What we can do and what we cannot do with fMRI.” Nature 453(7197): 869-878.
Logothetis, N., J. Pauls, et al. (2001). ”Neurophysiological investigation of the basis of the fMRI signal.” Nature 412(6843):
Lu, H., C. Zhao, et al. (2008). ”Baseline blood oxygenation modulates response amplitude: Physiologic basis for
intersubject variations in functional MRI signals.” Magnetic resonance in medicine 60(2): 364-372.
Lund, T., K. Madsen, et al. (2006). ”Non-white noise in fMRI: does modelling have an impact?” NeuroImage 29(1): 54-66.
Macey, P., K. Macey, et al. (2004). ”A method for removal of global effects from fMRI time series.” NeuroImage 22(1):
Mezer, A., Y. Yovel, et al. (2009). ”Cluster analysis of resting-state fMRI time series.” NeuroImage 45(4): 1117-1125.
Montano, N., A. Porta, et al. (2001). ”Evidence for Central Organization of Cardiovascular Rhythms.” Annals of the New
York Academy of Sciences 940(1): 299-306. Murphy, K., R. Birn, et al. (2009). ”The impact of global signal regression on
resting state correlations: Are anti-correlated networks introduced?” NeuroImage 44(3): 893-905.
Murphy, K., A. Harris, et al. (2010). ”Robustly measuring vascular reactivity differences with breath-hold: Normalising
stimulus-evoked and resting state BOLD fMRI data.” NeuroImage.
Noll, D. and W. Schneider (1994). Theory, Simulation and compensation of physiological motion artifacts in functional
MRI. Proceedings of IEEE International Conference on Image Processing, Austin, Texas.
Obrig, H., M. Neufang, et al. (2000). ”Spontaneous low frequency oscillations of cerebral hemodynamics and metabolism
in human adults.” NeuroImage 12(6): 623-639.
Pollatos, O., R. Schandry, et al. (2007). ”Brain structures mediating cardiovascular arousal and interoceptive awareness.”
Brain Research 1141: 178-187.
Renvall, V. and R. Hari (2009). ”Transients may occur in functional magnetic resonance imaging without physiological
basis.” Proceedings of the National Academy of Sciences of the United States of America 106(48): 20510-20514.
Saad, Z., D. Glen, et al. (2009). ”A new method for improving functional-to-structural MRI alignment using local Pearson
correlation.” NeuroImage 44(3): 839-848.
olvinck, M., A. Maier, et al. (2010). ”Neural basis of global resting-state fMRI activity.” Proceedings of the National
Academy of Sciences 107(22): 10238-10243.
Shmuel, A., M. Augath, et al. (2006). ”Negative functional MRI response correlates with decreases in neuronal activity in
monkey visual area V1.” Nature Neuroscience 9(4): 569-577.
Shmueli, K., P. van Gelderen, et al. (2007). ”Low-frequency ﬂuctuations in the cardiac rate as a source of variance in the
resting-state fMRI BOLD signal.” NeuroImage 38(2): 306-320.
Soldati, N., S. Robinson, et al. (2009). ”Automatic classiﬁcation of brain resting states using fMRI temporal signals.”
Electronics Letters 45(1): 19.
Thomason, M., B. Burrows, et al. (2005). ”Breath holding reveals differences in fMRI BOLD signal in children and adults.”
NeuroImage 25(3): 824-837.
Tong, Y. and B. d. Frederick (2010). ”Time lag dependent multimodal processing of concurrent fMRI and near-infrared
spectroscopy (NIRS) data suggests a global circulatory origin for low-frequency oscillation signals in human brain.” NeuroImage
Triantafyllou, C., R. D. Hoge, et al. (2005). ”Comparison of physiological noise at 1.5 T, 3 T and 7 T and optimization of
fMRI acquisition parameters.” NeuroImage 26(1): 243-250.
van Buuren, M. t., T. Gladwin, et al. (2009). ”Cardiorespiratory effects on default-mode network activity as measured with
fMRI.” Human brain mapping 30(9): 3031-3042.
Van de Moortele, P. F., J. Pfeuffer, et al. (2002). ”Respiration-induced B0 ﬂuctuations and their spatial distribution in the
human brain at 7 Tesla.” Magn. Reson. Med. 47(5): 888-895.
Wager, T., V. van Ast, et al. (2009). ”Brain mediators of cardiovascular responses to social threat, part II: Prefrontal-
subcortical pathways and relationship with anxiety.” NeuroImage 47(3): 836-851.
Wager, T., C. Waugh, et al. (2009). ”Brain mediators of cardiovascular responses to social threat: part I: Reciprocal dorsal
and ventral sub-regions of the medial prefrontal cortex and heart-rate reactivity.” NeuroImage 47(3): 821-835.
Wise, R., K. Ide, et al. (2004). ”Resting ﬂuctuations in arterial carbon dioxide induce signiﬁcant low frequency variations
in BOLD signal.” NeuroImage 21(4): 1652-1664.