3978 • The Journal of Neuroscience, April 6, 2016 • 36(14):3978 –3987

Behavioral/Cognitive

Brain Signal Variability Differentially Affects Cognitive

Flexibility and Cognitive Stability
Diana J. N. Armbruster-Genç,1,2,3 Kai Ueltzhöffer,1,2,3 and Christian J. Fiebach1,2,3,4,5
1Department of Psychology, Goethe University Frankfurt, D-60323 Frankfurt am Main, Germany, 2Bernstein Center for Computational Neuroscience,

Heidelberg University, D-68161 Mannheim, Germany, 3Department of Neuroradiology, Heidelberg University, D-69120 Heidelberg, Germany, 4Donders
Institute for Brain, Cognition, and Behaviour, Radboud University, 6525 EN Nijmegen, The Netherlands, and 5IDeA Center for Individual Development and
Adaptive Education, D-60486 Frankfurt am Main, Germany

Recent research yielded the intriguing conclusion that, in healthy adults, higher levels of variability in neuronal processes are beneficial
for cognitive functioning. Beneficial effects of variability in neuronal processing can also be inferred from neurocomputational theories
of working memory, albeit this holds only for tasks requiring cognitive flexibility. However, cognitive stability, i.e., the ability to maintain
a task goal in the face of irrelevant distractors, should suffer under high levels of brain signal variability. To directly test this prediction,
we studied both behavioral and brain signal variability during cognitive flexibility (i.e., task switching) and cognitive stability (i.e.,
distractor inhibition) in a sample of healthy human subjects and developed an efficient and easy-to-implement analysis approach to
assess BOLD-signal variability in event-related fMRI task paradigms. Results show a general positive effect of neural variability on task
performance as assessed by accuracy measures. However, higher levels of BOLD-signal variability in the left inferior frontal junction area
result in reduced error rate costs during task switching and thus facilitate cognitive flexibility. In contrast, variability in the same area has
a detrimental effect on cognitive stability, as shown in a negative effect of variability on response time costs during distractor inhibition.
This pattern was mirrored at the behavioral level, with higher behavioral variability predicting better task switching but worse distractor
inhibition performance. Our data extend previous results on brain signal variability by showing a differential effect of brain signal
variability that depends on task context, in line with predictions from computational theories.
Key words: behavioral variability; BOLD-signal variability; cognitive flexibility; cognitive stability; fMRI

Significance Statement
Recent neuroscientific research showed that the human brain signal is intrinsically variable and suggested that this variability
improves performance. Computational models of prefrontal neural networks predict differential effects of variability for different
behavioral situations requiring either cognitive flexibility or stability. However, this hypothesis has so far not been put to an
empirical test. In this study, we assessed cognitive flexibility and cognitive stability, and, besides a generally positive effect of
neural variability on accuracy measures, we show that neural variability in a prefrontal brain area at the inferior frontal junction
is differentially associated with performance: higher levels of variability are beneficial for the effectiveness of task switching
(cognitive flexibility) but detrimental for the efficiency of distractor inhibition (cognitive stability).

Introduction al., 2008; Deco et al., 2009; Garrett et al., 2013b). Although one
Studies from various fields of neuroscience have begun to show might intuitively assume that variability in neural processing is
that the brain is inherently variable in its processing and that this associated with unconstrained states and that it results in im-
variability might in fact be functional (Stein et al., 2005; Faisal et paired performance, it was shown that brain signal variability
increases from rest to task (Garrett et al., 2013a) and is linked to
task difficulty (Garrett et al., 2014). Also, neural variability is
Received June 19, 2014; revised Feb. 3, 2016; accepted Feb. 11, 2016. higher in healthy adults (showing better performance in cogni-
Author contributions: D.J.N.A.-G., K.U., and C.J.F. designed research; D.J.N.A.-G. and K.U. performed research;
D.J.N.A.-G. analyzed data; D.J.N.A.-G., K.U., and C.J.F. wrote the paper.
This research was funded by German Ministry of Education and Research Grant 01GQ1003A. C.J.F. was also Correspondence should be addressed to Diana J. N. Armbruster-Genç, Goethe University Frankfurt am Main,
supported by the State Offensive for the Development of Scientific and Economic Excellence initiative of the State of Department of Psychology, Theodor-W.-Adorno-Platz 6 (PEG), 60323 Frankfurt am Main, Germany. E-mail:
Hessen, the Emmy Noether Program of the German Research Foundation (Grant FI 848/3-1), and VIDI Grant [email protected].
45209006 from the Netherlands Organization for Scientific Research. DOI:10.1523/JNEUROSCI.2517-14.2016
The authors declare no competing financial interests. Copyright © 2016 the authors 0270-6474/16/363978-10$15.00/0
Armbruster-Genç et al. • Differential Effects of Brain Signal Variability J. Neurosci., April 6, 2016 • 36(14):3978 –3987 • 3979

tive tasks) compared with older individuals who show worse cog- Experimental procedure and behavioral analyses
nitive performance (Garrett et al., 2011). Thus, variability in The task is a highly time-efficient event-related fMRI paradigm that has
neural processing may in fact be beneficial. Importantly, BOLD been designed specifically to probe both cognitive flexibility and stability
mean activity and BOLD variability are essentially independent (Fig. 1A; cf. Armbruster et al., 2012): in 80% of the trials, subjects have to
indicate by button press whether a digit between 1 and 9 (excluding 5)
(Garrett et al., 2011), which underlines the importance of consid-
appearing above the central fixation cross is odd or even ( parity
ering variability of neural processing as a relevant indicator of
judgment; ongoing task; total of 240 trials). In 20% of the trials, a second
brain function. These conclusions are further supported by de- digit appears below the fixation cross, and subjects are instructed to
velopmental studies that show an increase of variability from answer to the brighter of the two digits. If this is the upper digit, they have
childhood to adulthood (McIntosh et al., 2008, 2010). to ignore the lower digit and continue responding to the upper digit
The importance of neural variability is also highlighted in according to the odd/even rule (distractor inhibition condition, assessing
neurocomputational models of working memory processes in cognitive stability; total of 20 trials). However, if the lower digit is
prefrontal circuits: Durstewitz and Seamans (2008) suggest that brighter, subjects are instructed to switch to this lower digit and to per-
the signal-to-noise ratio (SNR) of cortical networks varies be- form a magnitude judgment, i.e., decide whether it is greater or smaller
tween network states, i.e., between states supporting stable rep- than 5 (task switch condition, assessing cognitive flexibility; total of 20
resentations within prefrontal working memory circuits (i.e., trials). From the next trial on, participants continue to apply the ongoing
task to the upper digit again. Trial length is fixed to 2 s. Note that, in this
characterized by low variability and high SNR) and rather unsta-
task design, task switching involves also a component of inhibitory pro-
ble network states characterized by higher levels of neural vari- cesses, i.e., suppressing the response to the upper stimulus and suppress-
ability that make the system vulnerable to random fluctuations ing the task set of the ongoing condition. At least the latter aspect is
but also allow it to flexibly switch between network states if nec- shared with other variants of task-switching paradigms. Also, experi-
essary. From this work, we can infer that variability of neural mental condition (i.e., ongoing and distractor inhibition versus task
processing supports the dynamic adaptation to changing envi- switching) is to a certain degree confounded with task (i.e., parity versus
ronmental demands, a notion that can also be found in work by magnitude judgment). Both of these characteristics stem from the fact
Deco et al. (2009) and Rolls and Deco (2010), who argue that that we strived to design the task as representative as possible for the
stochastic dynamics of neural systems promote probabilistic de- computational model on which our work is based, i.e., the “dual-state
cision making, signal detection, or creative thought. theory” by Durstewitz and Seamans (2008). This, in our view, increases
the theoretical relevance of our results (for a more detailed discussion, see
Given this background of empirical and theoretical work, we
Armbruster et al., 2012). These potential limitations are ameliorated by
hypothesized that individuals with higher variability in neural pretesting involving an fMRI experiment in an independent sample of
processing benefit on tasks requiring cognitive flexibility but suf- n ⫽ 8 participants. In that study, participants performed magnitude and
fer when the situational demands require a higher degree of parity judgments with equal frequencies in a block design (three blocks
cognitive stability and vice versa. We especially aimed at investi- per condition, 150 trials per block; in all other aspects comparable with
gating a potential role of variability in the inferior frontal junc- the present study). Results showed no differences between the two tasks
tion (IFJ) region, because work from our laboratory and from in accuracy (t ⫽ ⫺0.73, p ⫽ 0.49) but faster response times (RTs) for the
others has supposed this region to represent task rules and medi- magnitude judgment (t ⫽ 4.5, p ⫽ 0.003). This indicates a higher degree
ate task switches (Derrfuss et al., 2005; Armbruster et al., 2012; of automaticity for the magnitude judgment, which is line with other
Ueltzhöffer et al., 2015). To test this hypothesis, we investigated research showing, for example, that understanding of parity evolves later
brain signal variability under different task demands—i.e., cog- during development (Berch et al., 1999). However, as participants
switched from parity judgments (ongoing condition) to magnitude judg-
nitive flexibility versus cognitive stability—and its effect on be-
ments (switch condition), this difference, if at all, works against the di-
havioral performance, which is in our view crucial for gaining a rection of the expected switch cost effects in behavior, which should
deeper understanding of the function of brain signal variability result in prolonged RTs for switching/magnitude judgments. Brain acti-
for cognition. Moreover, we investigated the relationship be- vation did not differ between magnitude and parity judgment, neither
tween behavioral variability and performance in cognitive flexi- across the whole brain nor in brain regions that show central results in
bility and cognitive stability, because the reasoning derived from the present study. Specifically, an ROI analyses in the left IFJ (mask
computational theories might also hold for variability at the be- derived from the results in the present study; see below; see Fig. 3B)
havioral level. Finally, we also explored directly the relationship applying a lower threshold did not show any differences between the
between brain signal variability and behavioral variability. Brain tasks (t ⫽ 2.34, p ⫽ 0.2). These results make it highly unlikely that the
signal variability was assessed using the difference of residuals reported results stem from a confound between task condition and
the nature of the task itself.
(DoR), a new and easy-to-implement analysis approach that re-
For analyses of the behavioral data, first all trials with an RT ⱕ 150 ms
lies on model comparison based on the residual maps produced were eliminated as suggested by previous research (MacDonald et al.,
during model estimation by fMRI analysis packages such as SPM 2006b; Dixon et al., 2007; Garrett et al., 2011). An upper boundary for
(for Statistical Parametric Mapping) or FSL (for Functional MRI legitimate RTs was not applied because trials lasted only 2 s and thus
of the Brain Software Library). slower responses automatically counted as misses. Additionally, for the
baseline condition, trials directly after the performance of a task-
Materials and Methods switching or distractor inhibition trial were excluded from analyses be-
Participants cause they might be “contaminated” with other cognitive processes, such
Ninety-five participants were recruited by sending advertisement letters as switching back to the upper digit. Costs in behavioral performance
to 3200 randomly selected citizens from the city of Heidelberg, Germany. from task switching or distractor inhibition were calculated by subtract-
A total of 19 subjects had to be excluded from data analysis because of ing the mean RT or mean error rate in the baseline condition from the
reporting strategic response behavior, an error rate ⬎30%, or neurolog- mean RT or mean error rate in the respective task condition, resulting in
ical abnormalities. The final sample consisted of 76 subjects (42 males; switch costs in RT and error rate, as well as distractor costs in RT and
aged 20 –51 years; mean age, 31.4 years). All had normal or corrected-to- error rate (Monsell, 2003). For assessment of behavioral variability, the
normal vision, reported no history of any neurological or psychiatric intra-individual coefficient of variation (ICV) was calculated for each
diseases, were right-handed, and gave written informed consent accord- participant by dividing the individual mean SD by the individual mean of
ing to a protocol approved by the ethics committee of the University of RTs (Guilford, 1956; Schmiedek et al., 2009). This was done for the
Heidelberg. ongoing condition for which a total of 240 trials was available, thus
3980 • J. Neurosci., April 6, 2016 • 36(14):3978 –3987 Armbruster-Genç et al. • Differential Effects of Brain Signal Variability

allowing for a reliable estimation of behavioral

variability. Behavioral variability was also as-
sessed in a condition-specific manner, i.e., the
ICV was also calculated for RTs of task switch-
ing and distractor inhibition. A residualization
of age from the behavioral data was not per-
formed because age did not correlate with any
of the behavioral measures. The behavioral
measures were subjected to correlational anal-
yses with each other and were also screened for
possible speed–accuracy tradeoff effects as in-
dicated by highly negative correlations be-
tween RTs and error rates of one condition
(Heitz, 2014). Finally, correlations between be-
havioral data and BOLD-signal variability were
examined.

FMRI procedure
All images were acquired on a 3 Tesla Siemens
Trio MRI scanner equipped with a fast gradient
system for EPI and a 32-channel head coil. A
total of 350 functional volumes were acquired
in two runs, each lasting ⬃5 min, using a T2*-
weighted BOLD-sensitive gradient-echo, EPI
sequence with 32 oblique axial slices (thick-
ness, 3 mm; interslice gap, 1 mm; field of view,
192 mm; matrix size, 64 ⫻ 64; in-plane resolu-
tion, 3 ⫻ 3 mm; repetition time, 1800 ms; echo
time, 30 ms; flip angle, 80°). The first eight
volumes of each run were discarded to allow
for stable magnetization. In addition, a T1-
weighted magnetization prepared-rapid gradi-
ent echo scan was acquired (thickness, 1 mm;
field of view, 256 mm; matrix size, 256 ⫻ 256;
in-plane resolution, 1 ⫻ 1 mm; repetition time,
1570 ms; echo time, 2.63 ms; flip angle, 30°).

fMRI data analysis

Data preprocessing. fMRI data were first pre-
processed using the SPM8 software (www.fil.
ion.ucl.ac.uk/spm/software/spm8/): EPI im-
ages were corrected for slice-timing differences
and motion. Segmentation of the individual
subject’s structural image provided normaliza-
tion parameters that were used to normalize
the functional images to the Montreal Neuro-
logical Institute (MNI) template reference
brain. Finally, images were smoothed with an 8
mm FWHM Gaussian kernel. To further de-
noise the data, individual independent compo-
nent analyses (ICA) were conducted for each
subject as implemented in FSL MELODIC
(Beckmann and Smith, 2004). After visual in-
spection of the components following the pro-
cedure described by Kelly et al. (2010), the
individual noise components were filtered
from the original fMRI data by including their
time series into the general linear model
(GLM) as additional regressors. Analogously,
time series of white matter and CSF ROIs were
filtered out. These additional preprocessing
steps have been shown to reduce variance of
the signal considerably while, at the same time,
increasing predictive power of the actual Figure 1. Task paradigm and analysis pipeline for BOLD-signal variability. A, Task paradigm showing examples of ongoing task
BOLD-signal variability measure (Garrett et trials, distractor inhibition trials, and task-switch trials. B, Standard regression model including one regressor per condition (plus
al., 2010, 2014). Mean BOLD activity was ana- 1st and 2nd derivative; see also Model A). C, Trial-by-trial regression model including one regressor (plus 1st and 2nd derivative)
lyzed as described previously (Armbruster et per trial for the condition of interest (see also Model B). D, Brain map exemplifying a DoR (i.e., standard model minus trial-by-trial
al., 2012). regression model) map for one subject.
Calculation of BOLD-signal variability. We
quantified BOLD-signal variability by calculat-
Armbruster-Genç et al. • Differential Effects of Brain Signal Variability J. Neurosci., April 6, 2016 • 36(14):3978 –3987 • 3981

ing the DoR of two different regression models within SPM8 (Fig. 1B–D), Table 1. Descriptive statistics for behavioral data
using both a GLM with a canonical hemodynamic response function RTs (ms) Error rates (%)
including time and dispersion derivatives to account for variations in the
peak and width of the hemodynamic response, and a high-pass filter with Condition Mean SE Mean SE
128 s cutoff (Friston et al., 1998). Model A (Fig. 1B) is a standard regres- Ongoing task 679.4 10.3 2.98 0.6
sion model for an individual subject as it is typically used in fMRI data Distractor inhibition 903.5 14.8 9.08 0.9
analyses, including one regressor per condition, one for error trials, plus Task switching 961.9 11.5 6.45 0.9
additional regressors for motion parameters and for the individual noise Distractor costs 224.1 16.3 6.10 1.1
components from the ICA. Model B (Fig. 1C) is a condition-specific Switch costs 282.5 16.3 3.47 1.1
trial-by-trial regression model that is equivalent to the first model, i.e., SE was computed according to Cousineau (2005) and Morey (2008), i.e., the between-subject variance was removed
including one regressor per condition, with the exception that, for the as a repeated-measures ANOVA was applied for data analysis.
condition of interest (e.g., task switching in Model B), each single correct
trial is represented by its own regressor. Note also that, in this model,
freedom of the regression model. However, such a correction would bias
error trials are modeled within one regressor.
the DoR analysis, because the number of degrees of freedom is con-
Model A shows the standard regression model:
founded with the error rate in the trial-to-trial regression model (in
y 共 t 兲 ⫽ meanA ⫹ ␤errxerr共t兲 ⫹ ␤aSxaS共t兲 ⫹ ␤aNSxaNS共t兲 ⫹ ␤distrxdistr共t兲 which only correct trials were modeled with trial-specific regressors). To
avoid this bias, which would lead to a smaller DoR in subjects with higher
⫹ ␤switchxswitch共t兲 ⫹ movement ⫹ noiseICs ⫹ eA共t兲, error rates, we removed this correction for both regression models by
adjusting the respective MATLAB code that implements the GLM
and Model B shows the trial-by-trial regression model (exemplified for (SPM8: spm_spm.m, VResMS.pinfo(1) ⫽ 1/xX.trRV in line 899 deleted).
task switching): Group analyses were performed in the standard framework of SPM by
feeding the individual, condition-specific DoR maps into a group-level
y 共 t 兲 ⫽ meanB ⫹ ␤errxerr共t兲 ⫹ ␤aSxaS共t兲 ⫹ ␤aNSxaNS共t兲 ⫹ ␤distrxdistr共t兲 random-effects model. BOLD-signal variability during distractor inhibi-
tion and during task switching were analyzed separately.
⫹ ␤switch1xswitch1共t兲 ⫹ … ⫹ ␤switchN xswitchN 共t兲 ⫹ movement
Individual differences analyses. For the purpose of this study, we fo-
⫹ noiseICs ⫹ eB共t兲, cused on individual difference analyses, i.e., correlational analyses. Thus,
covariates were entered into the group DoR analyses of BOLD-signal
where mean indicates intercept, err indicates error trials, aS indicates variability. We were interested specifically in potential relationships be-
ambiguous switch trials, aNS indicates ambiguous non-switch trials, tween BOLD-signal variability and the behavioral costs of task switching
distr indicates distractor trials, switch indicates switch trials, and distractor inhibition as determined using RTs and error rates. Fur-
switch1…switchN indicates correct switch trials 1 to N (maximum 20), thermore, we examined the relationship between BOLD-signal and be-
movement indicates movement parameters from the SPM realignment havioral variability (ICV; see above). Because age has been shown to
procedure during preprocessing, six parameters in total, each modeled affect brain signal variability (Garrett et al., 2010) and because brain
with its own regressor, noiceICs indicates independent components signal variability and age were also correlated in our data (see Results), we
identified as artifacts, each IC modeled with its own regressor, and e controlled for age effects by including age as a covariate of no interest.
indicates error term. Finally, all individual differences results, i.e., bivariate correlations, were
The main rational underlying the present analysis is that the DoR of checked for multivariate outliers with the Mahalanobis distance measure
these two regression models provides an estimate of the trial-to-trial (Mahalanobis, 1936).
variability in one specific condition of the experiment by quantifying Throughout all analyses, a significance level of 5% (familywise correc-
how much additional variance can be explained when explicitly account- tion for multiple comparisons) was applied. To this end, a voxel-level
ing for trial-to-trial variations in BOLD activation strength by separately threshold of p ⬍ 0.005 (uncorrected) was combined with a non-arbitrary
modeling each individual trial from that condition. Because residuals are cluster-extent threshold (Forman et al., 1995). The cluster-extent thresh-
the not-explained variance (varres) of a regression model equaling total old (k ⫽ 133) resulting in a corrected probability of 5% for false positives
variance (vartot) minus explained variance (varex), it can be shown that was determined via a Monte Carlo simulation with 10,000 iterations
the DoR measure calculated here is equivalent to the difference of ex- using the AFNI (for Automated Functional Neuro-Imaging) routine Al-
plained variance of Model B minus the explained variance of Model A: phaSim (Ward, 2000) to model the functional image matrix, excluding
cerebellar and brainstem areas as our focus lies on cortical functions.
vartot ⫽ varex ⫹ varres 7 varres ⫽ vartot ⫺ varex,
Results
difference of the residuals of the two models:
Behavioral performance
varresa ⫺ varresb ⫽ 共vartota ⫺ varexa兲 ⫺ 共vartotb ⫺ varexb兲 7 As expected given previous results in an independent sample
(Armbruster et al., 2012), we observed greater difficulty—as in-
from vartota ⫽ vartotb follows dicated by performance costs relative to the ongoing task—for
both task switching (i.e., cognitive flexibility) and distractor in-
varresa ⫺ varresb ⫽ varexb ⫺ varexa.
hibition (i.e., cognitive stability; cf. Table 1). RT costs were sig-
By taking the DoR variance (or the difference of explained variance) of nificantly higher for task switching compared with distractor
two models that are completely equivalent apart from the fact that there inhibition (t(75) ⫽ 3.6, p ⫽ 0.001; Cohen’s d ⫽ 0.51), whereas
is one “mean” regressor for the condition of interest in the one model distractor inhibition resulted in higher error rate costs (t(75) ⫽
(Model A/standard) and one regressor for each trial of this condition in 2.3, p ⫽ 0.024; Cohen’s d ⫽ 0.37). Task switching error costs and
the other model (Model B/trial-by-trial; in this study, up to 20 separate distractor inhibition RT costs showed a trend toward a negative
regressors depending on error rate), the DoR measure is a variance-based correlation (r ⫽ ⫺0.20; p ⫽ 0.08), suggesting that those partici-
measure of neural variability (see also Results Garrett et al., 2013b). A
pants that inhibit distractors more efficiently, i.e., that need less
crucial advantage of the DoR measure presented here is that the residuals
per voxel are computed automatically by SPM during model estimation time to do so, tend to commit more errors during task switching.
and saved as residual maps (ResMS.img), so that only a difference be- Although an inverse relationship between performance accuracy
tween two images has to be calculated. This analysis method can be and speed can be an indicator of speed–accuracy tradeoffs, this
implemented very efficiently by estimating two GLMs and then calculat- result was here found across conditions, which is not typically
ing DoR maps. By default, SPM corrects residual maps by the degrees of considered evidence for a speed–accuracy tradeoff (Heitz, 2014).
3982 • J. Neurosci., April 6, 2016 • 36(14):3978 –3987 Armbruster-Genç et al. • Differential Effects of Brain Signal Variability

However, we also found a significant negative correlation be- IFJ, bilateral middle frontal gyrus (MFG), right inferior frontal
tween task switching RTs and task switching error rates (r ⫽ gyrus (IFG) and insula, the middle and anterior cingulate cortex
⫺0.24, p ⫽ 0.04), which could indeed hint toward a speed–accu- (ACC), supplementary motor area (SMA), medial prefrontal
racy tradeoff. When examining this correlation in more detail, it cortex (MPFC), occipital regions, thalamus, and right basal gan-
became evident that it was driven by two subjects exhibiting very glia (Fig. 2B, areas in red; Table 2). Thus, individuals with higher
high error rates and very short RTs at the same time; also, a brain signal variability in these areas show smaller task-switching
Mahalanobis distance analysis for multivariate outliers identified costs with respect to error rates and are thus more effective in task
these subjects as extremes, although not as outliers (MD1 ⫽ 9.8; switching. A subset of these areas also showed a negative correla-
MD2 ⫽ 12.8). When excluding these subjects, the correlation tion between BOLD-signal variability and error rate costs for
became insignificant (r ⫽ ⫺0.12, p ⫽ 0.33), whereas all other distractor inhibition (Fig. 2B, areas in blue, overlap with task
results only slightly changed numerically but were unchanged switching in magenta; Table 2).
with respect to significance level. For distractor inhibition, no In contrast, RT costs for distractor inhibition correlated pos-
evidence for a speed–accuracy tradeoff was found (r ⫽ 0.16, itively with inhibition-related BOLD-signal variability in the left
p ⫽ 0.18). IFJ (Fig. 3A; Table 2). A conjunction analysis revealed that the
negative correlation between BOLD signal variability and distrac-
Effects of behavioral variability on cognitive stability tor inhibition costs and the positive correlation between BOLD
versus flexibility signal variability and task-switching costs in fact overlapped in
Behavioral variability was determined separately for each partic- the left IFJ (Fig. 3B). Costs in RT for task switching showed no
ipant by calculating the ICV (see Materials and Methods) for RTs significant relationship with BOLD-signal variability. Thus, vari-
from the ongoing condition and for RTs in the two critical task ability of BOLD signals at the junction region of the left inferior
conditions, i.e., task switching and distractor inhibition. Behav- frontal and precentral sulci (IFJ) is antagonistically related to
ioral variability differed substantially between individuals (mean cognitive flexibility and stability (Fig. 3C).
ICV ongoing ⫾ SD, 0.2 ⫾ 0.04; mean ICV switch ⫾ SD, 0.13 ⫾
0.03; mean ICV distractor inhibition ⫾ SD, 0.2 ⫾ 0.06). Behav- Relationship between brain signal variability and
ioral variability in the ongoing condition was differentially re- behavioral variability
lated to cognitive flexibility versus stability: although the ICV for Because we were interested in exploring whether or not behav-
the ongoing task was significantly negatively correlated with error ioral variability is predictive of variability in neural processing, we
rate costs for task switching (r ⫽ ⫺0.314, p ⫽ 0.006; R 2 ⫽ 0.1; RT also examined the relationship between condition-specific
costs task switching, r ⫽ ⫺0.024, p ⫽ 0.839), it was positively BOLD-signal variability and behavioral variability as measured
related to RT costs for distractor inhibition (r ⫽ 0.40, p ⫽ 0.0004; by the ICV. For distractor inhibition-related neural variability,
R 2 ⫽ 0.16 after excluding one outlier; distractor inhibition error we did not observe any correlations with the measures of behav-
costs, r ⫽ ⫺0.027, p ⫽ 0.814). RT variability in the distractor ioral variability. However, for neural variability during task
inhibition condition did not correlate with any of the distractor switching, we observed a positive correlation with behavioral
inhibition cost measures. However, for task switching, behavioral variability in the ongoing condition, in the left IFJ, the ACC, right
variability results showed a positive correlation with task- precentral and postcentral regions, right middle temporal gyrus,
switching costs in RTs (r ⫽ 0.41, p ⫽ 0.0003; R 2 ⫽ 0.17). The as well as an occipital region in the right hemisphere and in a
three measures of behavioral variability were not correlated with subcortical cluster including the left caudate nucleus and thala-
each other (all p ⬎ 0.1), indicating that they most likely reflect mus (Table 3). For behavioral variability during task switching,
different system characteristics. Age did not correlate with any of there were only negative correlations found with neural variabil-
the behavioral performance measures (all p ⬎ 0.1). Thus, perfor- ity in this condition, comprising an extensive cluster of brain
mance data showed substantial between-individual differences in regions including precentral and postcentral gyri, bilateral SMA,
behavioral variability and suggest that behavioral variability un- inferior parietal cortex, and insula (Table 3).
der certain circumstances can be beneficial for cognitive flexibil-
ity but detrimental for cognitive stability. Comparing DoRs with ␤-estimate variance
Finally, to validate the proposed DoR method for analyzing
Mean BOLD activity for cognitive flexibility and stability BOLD-signal variability, we calculated the variance across the ␤
As demonstrated previously (Armbruster et al., 2012), task estimates for the task-switching condition as calculated in the
switching and distractor inhibition in the present task activate a trial-by-trial model for each subject, for a cluster of interest in the
widespread frontoparietal network, that mostly overlaps for both left IFJ comprising eight voxels that we identified above as show-
conditions but is at the same time more extended for task switch- ing opposing relationships between BOLD-signal variability
ing (Fig. 2A; see also Armbruster et al., 2012, their Fig. 4 A). and behavioral distractor versus switch costs (see above and
Fig. 3B). In this cluster, we found a highly significant positive
Differential effects of brain signal variability on cognitive correlation of the DoR measure with trial-by-trial variability
stability versus flexibility (r ⫽ 0.601, p ⬍ 0.0000001). This result demonstrates that the
The main purpose of this study was to explore whether or not DoR measure is highly comparable with variance-based mea-
brain signal variability is differentially related to cognitive flexi- sures of neural variability.
bility versus stability. To this end, we estimated condition-
specific BOLD-signal variability (for details, see Fig. 1 B, C and Discussion
Materials and Methods) separately for each participant and con- Although previous studies have shown that variability in neural
dition and examined the relationship between BOLD-signal processing can also be helpful (Garrett et al., 2010, 2011, 2013b,
variability and behavioral performance, as well as behavioral 2014), we show in this study that the effect of behavioral and
variability. For task switching, BOLD-signal variability correlated brain signal variability on task performance varies depending on
negatively with error rate costs in a set of areas comprising the left task demands. We implemented a new analysis pipeline for ana-
Armbruster-Genç et al. • Differential Effects of Brain Signal Variability J. Neurosci., April 6, 2016 • 36(14):3978 –3987 • 3983

Figure 2. Brain activity and performance correlations with brain signal variability. A, Univariate activation for task switching (red) and distractor inhibition (yellow); overlap in orange (for
visualization purposes, a strict threshold of p ⬍ 0.000000000001 was applied so that local maxima of activation patterns can be better identified.). B, Negative correlations between condition-
specific BOLD-signal variability and error costs: areas in red show significant negative correlation with error costs in task switching, areas in blue correlate negatively with error costs in distractor
inhibition, and overlap in magenta. BG, Basal ganglia; PCC, posterior cingulate corex; SFG, superior frontal gyrus.

lyzing trial-to-trial BOLD-signal variability in a condition- the new rule more easily, resulting in fewer errors. Neverthe-
specific manner, i.e., the DoR approach, and could show that less, it might for the exact same reason take longer to suppress
such antagonistic relationships between variability and flexi- irrelevant task rules triggered by a distractor stimulus, which
ble versus stable task performance can also be observed at the results in longer RTs during distractor inhibition. However,
neural level, driven in particular by neural variability in the IFJ this interpretation is at present speculative in nature and re-
region. quires additional research in the future.
We found no such association for inhibition-specific behav-
The relationship between behavioral variability and ioral variability and observed higher RT switch costs in partici-
cognitive processing pants with greater task switching-related behavioral variability.
In this study, we determined the ICV for the ongoing condition as In addition, the three indices of behavioral variability were not
a robust indicator of behavioral variability. The observed antag-
correlated, suggesting that they may in fact reflect different sys-
onistic relationship between this basic measure of variability and
tem properties. For example, the switching-specific ICV effect
the behavioral costs of cognitive flexibility versus stability is in
seems to indicate that increased variability during switching re-
line with our hypothesis derived from computational theories
flects inefficient implementation of switching-specific cognitive
(Durstewitz and Seamans, 2008; Ueltzhöffer et al., 2015), i.e., that
higher levels of variability facilitate cognitive flexibility but processes; this finding is in line with previous research finding
have detrimental effects on cognitive stability. Interestingly, higher levels of behavioral variability in cognitively impaired
this antagonistic effect is reflected in (lower) error rate costs samples (MacDonald et al., 2006a). In contrast, we interpret the
for task switching but in (higher) RT costs for distractor inhi- antagonistic relation of behavioral variability in the ongoing con-
bition. Note that this also holds for brain activation results dition to switch versus distractor costs as an indicator of basic
with respect to IFJ variability (see below). We speculatively system characteristics determining how efficiently the cognitive
interpret this as indicating that, with higher neural variability, system can transition from one task state to another (Durstewitz
brain network states representing different task rules are more and Seamans, 2008; Deco et al., 2009), which may be beneficial
readily available (Deco et al., 2009). When required to switch for some situations but detrimental for others (for a similar rea-
from one task to the other, the system can accordingly activate soning, see Dreisbach and Goschke, 2004; Dreisbach, 2006).
3984 • J. Neurosci., April 6, 2016 • 36(14):3978 –3987 Armbruster-Genç et al. • Differential Effects of Brain Signal Variability

Table 2. Coordinates of peak correlations between condition-specific BOLD-signal variability and performance
MNI coordinates
Brain region BA Hemisphere x y z Tmax k
Task switching BOLD-signal variability negatively correlating
with task switching costs in error rates
IFJ 9/45/8/6 Left ⫺34 6 40 3.49 10301**
MFG/superior frontal gyrus 9/8 Left ⫺36 32 34 3.85 *
ACC 24/32 Right 2 40 16 3.59 *
MPFC 9/10 Left ⫺2 62 2 3.72 *
SMA 32 Left ⫺2 10 46 4.21 *
Middle/posterior cingulate cortex 31/24 Right 6 ⫺38 46 5.65 *
Postcentral gyrus 43 Left ⫺44 ⫺22 24 4.29 *
IFG/insula 13/45/46/47 Right 34 26 12 3.77 6338†
Calcarine sulcus 17/18 Right 2 ⫺96 0 4.13 †
Thalamus Left ⫺16 ⫺28 2 4.33 †
Distractor inhibition BOLD-signal variability negatively correlating
with distractor inhibition costs in error rates
MPFC 10 Right 6 32 ⫺10 4.05 1007
0 60 8 3.76
IFG 44/45/36/47 Right 56 28 10 4.27 285
Precuneus 31/7 Right 2 ⫺50 44 3.84 406
Angular gyrus 39/40 Left ⫺52 ⫺62 36 3.60 280
Right 62 ⫺50 28 3.91 206
Cuneus/occipital cortex Right 14 ⫺86 42 3.97 774
Distractor inhibition BOLD-signal variability positively correlating
with distractor inhibition costs in RT
IFJ 9/8 Left ⫺42 12 40 3.99 137
For correlations between task-switching-related BOLD-signal variability and task-switching error costs local maxima are reported as the correlations formed from two extensive clusters: *one containing 10,301 voxels and the †other
containing 6338 voxels.

Figure 3. Positive relationship between RT costs and BOLD-signal variability during distractor inhibition and antagonistic relation between performance and BOLD-signal variability in left IFJ. A,
Positive correlation between RT costs and condition-specific BOLD-signal variability during distractor inhibition, p ⬍ 0.05 (corrected). B, Conjunction in left IJF (k ⫽ 8 voxel) between (1) negative
correlation of task-switching costs in error rates with BOLD-signal variability, as shown in Figure 2B and (2) positive correlation of distractor inhibition costs in RT and BOLD-signal variability. C, Scatter
plots illustrating the antagonistic relationship between BOLD-signal variability and behavior for task switching (left) and distractor inhibition (right) in the overlapping part of the IFJ (note that, after
exclusion of one multivariate outlier as detected by Mahalanobis distance, the result for distractor inhibition was still significant with r ⫽ 0.26, p ⫽ 0.02). **p ⬍ 0.01. ER, Error rate.
Armbruster-Genç et al. • Differential Effects of Brain Signal Variability J. Neurosci., April 6, 2016 • 36(14):3978 –3987 • 3985

Table 3. Coordinates of peak correlations between condition-specific BOLD-signal variability and behavioral variability
MNI coordinates
Brain region BA Hemisphere x y z Tmax k
Task-switching BOLD-signal variability positively correlating
with ICV of the ongoing condition
IFJ 9/6 Left ⫺34 8 40 3.75 236
ACC 32/24 Right 6 22 30 3.73 207*
Precentral and postcentral gyrus 4/6 Right 48 ⫺14 32 3.82 *
Middle temporal gyrus 22 Right 52 ⫺14 0 4.00 320
22/39 Right 56 ⫺58 14 3.72 169
Occipital cortex 19/18 Right 42 ⫺78 ⫺12 4.67 145
Caudate Left ⫺12 ⫺2 22 4.71 377†
Thalamus Left ⫺2 ⫺18 16 3.71 †
Task-switching BOLD-signal variability negatively correlating
with ICV of task switching
Precentral gyrus 6/4 Right 36 ⫺14 40 5.21 10,629‡
Precentral/postcentral gyrus 6/4/43 Left ⫺52 ⫺10 14 4.27 ‡
SMA 24/31 Right 2 ⫺2 50 3.82 ‡
Insula 13 Right 48 ⫺6 4 4.35 ‡
13/41 Left ⫺42 ⫺16 10 4.00 ‡
Inferior parietal cortex 40 Left ⫺63 ⫺30 28 3.33 ‡
For some areas, local maxima are reported as the correlations formed from coherent clusters: *contains 207 voxels, †contains 377 voxels, and ‡contains 10,629 voxels.

Beneficial effects of brain signal variability on performance Specifically, we show that variability in the IFJ antagonistically
Our results suggest that higher levels of brain signal variability modulates task performance during cognitive flexibility versus
lead to reduced error rates, i.e., higher effectiveness of perfor- stability, suggesting that neural variability in this area affects the
mance. This was shown for task switching and for distractor in- stability of task rule representations. The IFJ is a region (1) that
hibition, which is in line with previous research suggesting was previously (based on mean BOLD activity) shown to be im-
beneficial effects of higher brain signal variability in general (Gar- portant for task switching (Brass et al., 2005; Derrfuss et al., 2005)
rett et al., 2011). Beneficial effects on performance were shown and for distractor inhibition (Armbruster et al., 2012; Ueltzhöffer
for variability of neural processing in the MPFC, posterior cingu- et al., 2015), (2) whose activity and connectivity was shown to be
late cortex (PCC), calcarine sulcus, and right middle posterior modulated by individual behavioral flexibility in ways similar to
temporal gyrus. Especially for MPFC, there is evidence that brain those observed for neural variability in the present study (Arm-
signal variability increases from rest to task in healthy young bruster et al., 2012); and (3) whose activity and connectivity to
subjects (Garrett et al., 2013a), is higher in faster and more con- striatal areas during task switching is modulated by a genetic
sistent performers (Garrett et al., 2011), and that variability first polymorphism regulating the density of the dopamine D2 recep-
increases with higher task difficulty but then decreases when be- tor (DRD2; Stelzel et al., 2010). The current results underline the
havioral performance drops (Garrett et al., 2014). importance of the IFJ for this antagonistic relationship between
MPFC is acknowledged as a prominent region of the default task switching and distractor inhibition by showing that variabil-
mode network (DMN) for which reliable task-related activity ity of neural processing in IFJ has opposite effects on cognitive
decreases have been observed (Garrett et al., 2014). So it appears stability versus flexibility.
that MPFC deactivates and at the same time shows higher levels
of variability during task, both of which can be beneficial for task Physiological relevance of brain signal variability
performance (Raichle et al., 2001). With respect to the current We have shown in previous work that activity and connectivity of
results, this reasoning also holds for the PCC, another important IFJ during task switching are modulated by DRD2/ANKK1–
hub of the DMN (Anticevic et al., 2010). One explanation for this TaqIa, a polymorphism that regulates the density of DRD2
may be found in the observation that increased trial-to-trial vari- in the striatum: individuals with lower DRD2 densities, suppos-
ability goes along with greater spontaneous coherence between edly associated with increased dopamine activity and decreased
regions (Fox et al., 2006) and that the functional embedding of a prefrontal D1 receptor activation, show better performance de-
region is related to its variability (Mišić et al., 2011). Thus, we spite lesser IFJ activity and IFJ-to-striatum connectivity during
tentatively speculate that higher brain signal variability may re- task switching (Stelzel et al., 2010). We propose that, together with
flect greater connectivity with other brain regions, which could recent work suggesting that BOLD-signal variability also depends on
for example influence task-related patterns of deactivation and dopamine levels (Garrett et al., 2015; Guitart-Masip et al., 2015),
activation of DMN versus task-relevant brain networks, respec- these lines of research link the current findings to computational
tively (Fox et al., 2005). theories that model working memory as network attractor states
of prefrontal neuronal circuits (Durstewitz and Seamans, 2008).
Antagonistic effects of brain signal variability This can provide a sound framework for interpreting the physi-
on performance ological relevance of our results, because neural variability might
Previous work has shown that the amount of variability in brain be interpreted as more frequent or easier transitions from one
signals can vary between cognitive states (rest vs task; Garrett et network state to another. According to Durstewitz and Seamans
al., 2013a), task conditions (n-back with upright vs inverted faces; (2008), such transitions of network states should be more prom-
Misić et al., 2010), and difficulty levels (Garrett et al., 2014). A key inent when the network is in a DRD2-dominated state that is,
finding of the present study is that task context can also modulate according to their computational simulations, characterized by
how brain signal variability influences behavioral performance. rather unstable representations, lower SNR, and spontaneous
3986 • J. Neurosci., April 6, 2016 • 36(14):3978 –3987 Armbruster-Genç et al. • Differential Effects of Brain Signal Variability

transitions between representations. At the behavioral level, this control: reduced perseveration at the cost of increased distractibility.
should afford more flexible processing, and our results support J Exp Psychol Learn Mem Cogn 30:343–353. CrossRef Medline
Durstewitz D, Seamans JK (2008) The dual-state theory of prefrontal cortex
this prediction. Conversely, stable maintenance of representa-
dopamine function with relevance to catechol-o-methyltransferase geno-
tions and a higher SNR, which is a characteristic of D1- types and schizophrenia. Biol Psychiatry 64:739 –749. CrossRef Medline
dominated activity states should be beneficial for cognitive Faisal AA, Selen LP, Wolpert DM (2008) Noise in the nervous system. Nat
stability (Durstewitz and Seamans, 2008). Note that these mac- Rev Neurosci 9:292–303. CrossRef Medline
roscopic model assumptions are also compatible with more re- Forman SD, Cohen JD, Fitzgerald M, Eddy WF, Mintun MA, Noll DC (1995)
cent work suggesting that stable working memory maintenance is Improved assessment of significant activation in functional magnetic res-
achieved by dynamic microscopic processes, such as limit cycle onance imaging (fMRI): use of a cluster-size threshold. Magn Reson Med
activity (Balaguer-Ballester et al., 2011; Sreenivasan et al., 2014). 33:636 – 647. CrossRef Medline
Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME
As we argued previously, evidence points to the IFJ as a critical
(2005) The human brain is intrinsically organized into dynamic, anticor-
region for initiating network switches or, in other terms, as a related functional networks. Proc Natl Acad Sci U S A 102:9673–9678.
region that represents task rules and thus contains networks CrossRef Medline
whose transitions lead to shifts in the currently active task rule Fox MD, Snyder AZ, Zacks JM, Raichle ME (2006) Coherent spontaneous
(Armbruster et al., 2012). Accordingly, the robustness of task rule activity accounts for trial-to-trial variability in human evoked brain re-
representations may be related to brain signal variability in the sponses. Nat Neurosci 9:23–25. CrossRef Medline
IFJ and may thus influence task performance during both cogni- Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R (1998)
tive flexibility and cognitive stability in an antagonistic manner. Event-related fMRI: characterizing differential responses. Neuroimage
7:30 – 40. CrossRef Medline
To conclude, this study demonstrates that stochastic vari-
Garrett DD, Kovacevic N, McIntosh AR, Grady CL (2010) Blood oxygen
ability of neural processing—measured here on the basis of level-dependent signal variability is more than just noise. J Neurosci 30:
BOLD signal variability— can influence behavioral perfor- 4914 – 4921. CrossRef Medline
mance differentially depending on task context. On the one Garrett DD, Kovacevic N, McIntosh AR, Grady CL (2011) The importance
hand, we observed a generally beneficial effect of brain-signal of being variable. J Neurosci 31:4496 – 4503. CrossRef Medline
variability on the accuracy of performance, which is in line Garrett DD, Kovacevic N, McIntosh AR, Grady CL (2013a) The modulation
with previous research. In addition, neural variability in the of BOLD variability between cognitive states varies by age and processing
speed. Cereb Cortex 23:684 – 693. CrossRef Medline
IFJ area, i.e., in a critical hub region for cognitive flexibility,
Garrett DD, Samanez-Larkin GR, MacDonald SW, Lindenberger U, McIn-
supports flexibility but hinders cognitive stability. These re- tosh AR, Grady CL (2013b) Moment-to-moment brain signal variabil-
sults underscore the importance of considering the variability ity: a next frontier in human brain mapping? Neurosci Biobehav Rev
of neural processes when investigating the complex relation- 37:610 – 624. CrossRef Medline
ship between brain and behavior. Garrett DD, McIntosh AR, Grady CL (2014) Brain signal variability is para-
metrically modifiable. Cereb Cortex 24:2931–2940. CrossRef Medline
Garrett DD, Nagel IE, Preuschhof C, Burzynska AZ, Marchner J, Wiegert S,
References Jungehülsing GJ, Nyberg L, Villringer A, Li SC, Heekeren HR, Bäckman L,
Anticevic A, Repovs G, Shulman GL, Barch DM (2010) When less is more: Lindenberger U (2015) Amphetamine modulates brain signal variabil-
TPJ and default network deactivation during encoding predicts working ity and working memory in younger and older adults. Proc Natl Acad Sci
memory performance. Neuroimage 49:2638 –2648. CrossRef Medline U S A 112:7593–7598. CrossRef Medline
Armbruster DJ, Ueltzhöffer K, Basten U, Fiebach CJ (2012) Prefrontal cor- Guilford JP (1956) Fundamental statistics in psychology and education.
tical mechanisms underlying individual differences in cognitive flexibility New York: McGraw-Hill.
and stability. J Cogn Neurosci 24:2385–2399. CrossRef Medline Guitart-Masip M, Salami A, Garrett D, Rieckmann A, Lindenberger U, Back-
Balaguer-Ballester E, Lapish CC, Seamans JK, Durstewitz D (2011) Attract- man L (2015) BOLD variability is related to dopaminergic neurotrans-
ing dynamics of frontal cortex ensembles during memory-guided mission and cognitive aging. Cereb Cortex. Advance online publication.
decision-making. PLoS Comput Biol 7:e1002057. CrossRef Medline Retrieved February 16, 2016. doi:10.1093/cercor/bhv029. CrossRef
Beckmann CF, Smith SM (2004) Probabilistic independent component
Medline
analysis for functional magnetic resonance imaging. IEEE Trans Med
Heitz RP (2014) The speed-accuracy tradeoff: history, physiology, method-
Imaging 23:137–152. CrossRef Medline
ology, and behavior. Front Neurosci 8:150. Medline
Berch DB, Foley EJ, Hill RJ, Ryan PM (1999) Extracting parity and magni-
Kelly RE Jr, Alexopoulos GS, Wang Z, Gunning FM, Murphy CF, Morimoto
tude from Arabic numerals: developmental changes in number process-
SS, Kanellopoulos D, Jia Z, Lim KO, Hoptman MJ (2010) Visual inspec-
ing and mental representation. J Exp Child Psychol 74:286 –308. CrossRef
tion of independent components: defining a procedure for artifact re-
Medline
moval from fMRI data. J Neurosci Methods 189:233–245. CrossRef
Brass M, Derrfuss J, Forstmann B, von Cramon DY (2005) The role of the
inferior frontal junction area in cognitive control. Trends Cogn Sci Medline
9:314 –316. CrossRef Medline MacDonald SW, Nyberg L, Bäckman L (2006a) Intra-individual variability
Cousineau D (2005) Confidence intervals in within-subject designs: a sim- in behavior: links to brain structure, neurotransmission and neuronal
pler solution to Loftus and Masson’s method. Tutorial Quant Methods activity. Trends Neurosci 29:474 – 480. CrossRef Medline
Psychol 1:42– 45. MacDonald SW, Hultsch DF, Bunce D (2006b) Intraindividual variability
Deco G, Rolls ET, Romo R (2009) Stochastic dynamics as a principle of in vigilance performance: does degrading visual stimuli mimic age-related
brain function. Prog Neurobiol 88:1–16. CrossRef Medline “neural noise”? J Clin Exp Neuropsychol 28:655– 675. CrossRef Medline
Derrfuss J, Brass M, Neumann J, von Cramon DY (2005) Involvement of the Mahalanobis PC (1936) On the generalised distance in statistics. Proc Natl
inferior frontal junction in cognitive control: meta-analyses of switching Institute Sci India 2:49 –55.
and Stroop studies. Hum Brain Mapp 25:22–34. CrossRef Medline McIntosh AR, Kovacevic N, Itier RJ (2008) Increased brain signal variability
Dixon RA, Garrett DD, Lentz TL, MacDonald SW, Strauss E, Hultsch DF accompanies lower behavioral variability in development. PLoS Comput
(2007) Neurocognitive markers of cognitive impairment: exploring the Biol 4:e1000106. CrossRef Medline
roles of speed and inconsistency. Neuropsychology 21:381–399. CrossRef McIntosh AR, Kovacevic N, Lippe S, Garrett D, Grady C, Jirsa V (2010) The
Medline development of a noisy brain. Arch Ital Biol 148:323–337. Medline
Dreisbach G (2006) How positive affect modulates cognitive control: the Misić B, Mills T, Taylor MJ, McIntosh AR (2010) Brain noise is task depen-
costs and benefits of reduced maintenance capability. Brain Cogn 60: dent and region specific. J Neurophysiol 104:2667–2676. CrossRef
11–19. CrossRef Medline Medline
Dreisbach G, Goschke T (2004) How positive affect modulates cognitive Mišić B, Vakorin VA, Paus T, McIntosh AR (2011) Functional embedding
Armbruster-Genç et al. • Differential Effects of Brain Signal Variability J. Neurosci., April 6, 2016 • 36(14):3978 –3987 • 3987

predicts the variability of neural activity. Front Syst Neurosci 5:90. Sreenivasan KK, Vytlacil J, D’Esposito M (2014) Distributed and dynamic
Medline storage of working memory stimulus information in extrastriate cortex.
Monsell S (2003) Task switching. Trends Cogn Sci 7:134–140. CrossRef Medline J Cogn Neurosci 26:1141–1153. CrossRef Medline
Morey RD (2008) Confidence intervals from normalized data: a correction Stein RB, Gossen ER, Jones KE (2005) Neuronal variability: Noise or part of
to Cousineau (2005). Tutorial Quant Methods Psychol 4:61– 64. the signal? Nat Rev Neurosci 6:389 –397. CrossRef Medline
Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman Stelzel C, Basten U, Montag C, Reuter M, Fiebach CJ (2010) Frontostriatal
GL (2001) A default mode of brain function. Proc Natl Acad Sci U S A involvement in task switching depends on genetic differences in D2 re-
98:676 – 682. CrossRef Medline ceptor density. J Neurosci 30:14205–14212. CrossRef Medline
Rolls ET, Deco G (2010) The noisy brain: stochastic dynamics as a principle Ueltzhöffer K, Armbruster-Genç DJ, Fiebach CJ (2015) Stochastic dynam-
of brain function. Oxford, UK: Oxford UP. ics underlying cognitive stability and flexibility. PLoS Comput Biol 11:
Schmiedek F, Lövdén M, Lindenberger U (2009) On the relation of mean e1004331. CrossRef Medline
reaction time and intraindividual reaction time variability. Psychol Aging Ward BD (2000) Simultaneous inference for fMRI data. Available at:
24:841– 857. CrossRef Medline https://afni.nimh.nih.gov/pub/dist/doc/manual/AlphaSimpdf.