2.1.

Subjects

Fourteen healthy subjects (9 men, 5 women, age: $33.4\pm 10.5$ years, range: 24 to 57 years) were asked not to smoke, eat, or consume any stimulants such as caffeine or energy drinks for 2 h before the start of the measurements. The study was approved by the Ethical Committee of the Canton of Zurich. Informed consent was obtained from the subjects prior to each measurement.

2.2.

Experimental Protocol and Experimental Setup for Wide-Field Visual Color Stimulation

Following a randomized crossover design, each subject was measured during three different experimental conditions: red, green, or blue. Each task was performed on a separate day to avoid potential carry-over effects but at the same time each day for each individual subject to exclude chronobiological effects. The mean time of measurement was 12:24 with a standard deviation of 2.36 h.

Each measurement lasted 33 min, i.e., 8-min baseline in darkness (interval 1), 10-min intermittent wide-field visual color stimulation (interval 2), and 15-min recovery in darkness (interval 3). The visual stimulation followed an event-related design of 15 alternating light-on (20 s) and light-off periods. The length of the light-off periods varied randomly [mean $21.1\pm 3.3\mathrm{s}$, range 17 to 27 s, Fig. 1(d)] to prevent habituation. During the measurements, the subjects sat opposite a screen (width: 2 m, height: 3 m, distance from subject to screen: 1.2 m) illuminated by four light-emitting diode beams (PAR 56 CAN RGB 05 BS, Cameo, Adam Hall GmbH, Neu-Anspach, Germany), which created a homogenous color. They were reproducibly controlled by a DMX mixer. The illuminance of the light at the eye level of the subjects was 20 lx for each color as measured by the LT300 luminance meter (Extech Instruments, Nashua, New Hampshire). The wavelengths of the maximum intensity determined by the MAYA 2000-Prospectrometer (Ocean Optics, Inc.) were 682 nm (red), 515 nm (green), and 465 nm (blue).

Fig. 1

(a) Placements of the NIRS-optode (right and left PFC at positions Fp2 and Fp1; right and left VC at positions ${\mathrm{O}}_2$ and ${\mathrm{O}}_1$), (b) sensitivity profile of NIRS-sensors on the brain, (c) experimental setup with position of the subject and the color screen, and (d) experimental protocol. (e) and (f) Overview of the block-averaged (group-level) stimulus-evoked changes in cerebral hemodynamics/oxygenation and systemic physiology ($\text{median}\pm \text{standard error}$ of the median). Black bar and gray area: time interval during visual stimulation.

Subjects were instructed not to move their body or head during the measurement to prevent movement artifacts (MAs).

2.3.

Systemic-Physiology-Augmented Functional Near-Infrared Spectroscopy: Equipment

The following devices were applied to simultaneously assess changes in cerebral hemodynamics and oxygenation, and systemic physiological activity: (i) a multichannel frequency-domain NIRS system (Imagent, ISS Inc., Champaign, IL), (ii) a gas analyzer (Nellcor N1000, Covidien, Dublin, Ireland), (iii) a continuous noninvasive blood pressure (NIBP) monitor (SOMNOtouch, Somno Medics, Randersacker, Germany), and (iv) a skin conductance measuring device (Mind-Reflection, Audiostrobe Ltd., UK).

The ISS Imagent frequency-domain NIRS system determined with 50-Hz resolution absolute values of the tissue oxygen saturation (${\mathrm{StO}}_2$), oxyhemoglobin ($[{\mathrm{O}}_2\mathrm{Hb}]$), deoxyhemoglobin ([HHb]), and total hemoglobin ([tHb]) concentration by the frequency-domain multidistance (FDMD) method⁶⁰ calculated by the ISS software. The Imagent employs 16 laser diodes at 834 nm and 16 at 760 nm, and 4 highly sensitive photomultiplier tubes as detectors. Optodes 1 and 2 had source–detector separations of $d=2.0$, 2.5, 3.5, and 4.0 cm. Optodes 3 and 4 had $d=2.5$, 3.0, 3.5, and 4.0 cm. Their multidistance geometry minimized the influence of extracerebral tissue such as scalp^61,62 and reduced MAs.⁶³ The correlation coefficient of the slopes enables identifying tissue inhomogeneities.⁶⁴ Optodes 1 and 2 were placed over the left (${\mathrm{Fp}}_1$) and right (${\mathrm{Fp}}_2$) PFC (LPFC and RPFC, respectively), and optodes 3 and 4 over the right (${\mathrm{O}}_2$) and left (${\mathrm{O}}_1$) VC (RVC and LVC, respectively) [Fig. 1(a)] according to the international 10 to 20 system⁶⁵ based on manually determined positions of standard landmarks (inion, nasion, and periauricular points). The latter covered the primary VC (V1) and the secondary VC (V2) [Fig. 1(b)]. The attachment of the opcodes to the head was done is such a way to avoid any discomfort and stress for the subjects, i.e., attaching the optodes too tightly was avoided, and it was ensured that no sharp edges of the optodes caused discomfort or pain.

A Nellcor N1000 gas analyzer measured the partial pressure of exhaled ${\mathrm{CO}}_2$ (${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$) noninvasively with a probe positioned directly below the right nostril of the subject [Fig. 1(a)].

The SOMNOtouch continuous NIBP device measured and determined the following parameters: mean arterial blood pressure (MAP), systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse pressure (PP), pulse transit time (PTT), heart rate (HR), high-frequency (HF) (0.15 to 0.4 Hz) component of the HR variability (HF-HRV), and the low-frequency (LF) (0.04 to 0.15 Hz) component of the HRV (LF-HRV).

To determine the status of the ANS, the skin-conductance biofeedback device measured the electrodermal activity (EDA) and the skin conductance level (SCL) at 2 Hz. Electrodes were attached to the distal phalange of the index and middle fingers of the right hand.

The methodical approach employed in our study of measuring changes in systemic physiology as well as local perfusion and oxygenation in the head we termed “systemic-physiology-augmented functional near-infrared spectroscopy” (SPA-fNIRS). The term refers to use systemic physiology measurements in order to assist, complement, improve, i.e., augment, the “traditional” fNIRS measurements.

2.4.

Assessment of Psychological Changes

The mood state of the subjects before and after the experiment was determined by the multidimensional mood questionnaire (MDBF),⁶⁶ yielding the good or bad mood (GS), the vigilance (WM), and the nervousness (RU) of the subject. For these three parameters, the postexperiment minus preexperiment differences were calculated ($\mathrm{\Delta }\mathrm{GS}$, $\mathrm{\Delta }\mathrm{WF}$, and $\mathrm{\Delta }\mathrm{RU}$). Deviations from a median of zero and influences of the subject’s gender and light color were assessed by a Wilcoxon signed rank test. The influence of age was tested using linear regression analysis.

2.5.

Signal Preprocessing

All signal preprocessing was performed in MATLAB (R2013b, MathWorks, Natick). MAs in the ${\mathrm{StO}}_2$, $[{\mathrm{O}}_2\mathrm{Hb}]$, [HHb], and [tHb] signals were removed by the movement artifact reduction algorithm (MARA).⁶⁷ The most prevalent MAs were spikes and sudden baseline shifts. In total, 3.3% of the fNIRS signal time was defined as artifacts, which were corrected by applying the MARA approach. MARA was evaluated as a useful^68–71 or moderately useful^72,73 MA correction technique. All parameters for MARA were selected for each signal separately and optimized to obtain an effective removal of MAs without disturbing the signal.

The fNIRS signals were downsampled to 2 Hz to remove high-frequency physiological and measurement noise by applying a low-pass filter to prevent aliasing. To further remove high-frequency noise, a moving average with a span of 2.4 s was applied.

The capnography signal was resampled to 2 Hz and the envelope calculated, from which the ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ was determined. To remove high-frequency noise, a robust local regression using weighted linear least squares and a second degree polynomial model (RLOES) with a bin of 20 samples were applied. This approach removes random noise while preserving the physiologically relevant high-frequency content. ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ is directly related to the arterial ${\mathrm{pCO}}_2$ (${\mathrm{PaCO}}_2$).⁷⁴ The respiration rate (RR) was extracted from the capnography signal by (i) detecting the maxima of every breath employing our peak detection algorithm,⁷⁵ (ii) calculating the time differences between successive peaks ($\mathrm{\Delta }T$), and (iii) resampling $\mathrm{\Delta }T$ to obtain an equidistantly sampled signal of the RR by a piecewise cubic Hermite interpolating polynomial. In addition, to quantify the coupling between RR and HR, the pulse-respiration quotient (PRQ) was calculated ($\mathrm{PRQ}=\mathrm{HR}/\mathrm{RR}$).

All other systemic physiology data, except the skin conductance, were also denoised by the RLOES method.

From the LF- and HF-HRV signals, the LF/HF ratio (LF/HF) was calculated. Furthermore, the cardiac output (Q) was calculated according to $\mathrm{Q}=(\mathrm{PP}\times \mathrm{HR})/(\mathrm{SBP}+\mathrm{DBP})$.⁷⁶

2.6.

Signal Processing and Statistical Data Analysis

For the final analysis, the following signals were included: (i) fNIRS-signals, i.e., $[{\mathrm{O}}_2\mathrm{Hb}]$, [HHb], [tHb], ${\mathrm{StO}}_2$, and Mayer wave amplitude (MWA), and (ii) systemic physiological signals, i.e., HR, RR, PRQ, ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$, MAP, PP, Q, PTT, SCL, HF, LF, and LF/HR.

To assess the changes elicited by the light stimulation, (i) stimulus-evoked changes were block averaged and (ii) neurosystemic functional connectivity was analyzed. Statistical significance was evaluated at subject and group level.

For the block averaging, each of the segments had a length of 35 s starting 5 s before the stimulus. Each single trial was normalized to the baseline value and linearly detrended. Then, the medians in the intervals 1 to 4, 6 to 9, 11 to 14, 16 to 19, 21 to 24, 26 to 29, and 31 to 34 s were calculated. Thus, the stimulus period was covered by four intervals between 6 and 24 s. For each trial, subject, and signal, it was tested whether the block-averaged stimulus-evoked changes were significant by a Wilcoxon signed rank test (subject-level analysis). False discovery rate (FDR) correction for multiple comparisons was applied. The $p$ and $h$ values of the statistical tests were stored for the subsequent group level statistical analysis. Here, all block averages of the stimulus-evoked changes for each subject and task were analyzed by quantile regression models using the $R$ quantreg-package.⁷⁷ Quantile regression is more robust against outliers and does not assume a normal distribution of errors compared to mean and least squares regressions. Confidence intervals for the regression were calculated by bootstrapping⁷⁸ with 100,000 bootstraps. It was tested whether the stimulus-evoked changes were significant and whether they depend on the factors color of light, age, and gender. The fNIRS signals were separately analyzed for the VC and PFC, but left and right VC or PFC were combined to increase the power of the statistics.

The inputs for the quantile regression were the block averages of each subject and task, where nonsignificant changes in the single block averages were weighted by zero. This step ensured that only statistically significant subject level data were included in the group-level analysis, thus reducing intersubject variability that confounds the potential color-type dependency.

To analyze the neurosystemic functional connectivity, for each person and each trial, a correlation matrix was computed by comparing each block average of each signal to any other signal by the Spearman correlation coefficient ($r_{\mathrm{s}}$) and its statistical significance ($p<0.05$, with FDR correction for multiple comparisons). If the correlation was not statistically significant, the value was deleted in the matrix. Then, the correlation matrices for all trials and subjects were averaged. Additionally, matrices were calculated only for the fNIRS signals and only for the systemic physiological signals. Then, all correlation matrices were analyzed by transforming the respective matrix into a network where the nodes represent the block-averaged stimulus-evoked signals and the edges represent the connection strengths in terms of the correlation coefficients. The complex network properties, i.e., the assortativity (i.e., the weighted assortativity coefficient, $r^{\mathrm{w}}$), transitivity (i.e., the weighted transitivity, $T^{\mathrm{w}}$), density ($D$), and efficiency (i.e., the weighted global efficiency, $E^{\mathrm{w}}$) were calculated using the MATLAB Brain Connectivity Toolbox.⁷⁹ The complex network parameters for each color were compared using a nonparametric Wilcoxon rank sum test.

Each complex network parameter is related to specific properties of the network created by the correlation matrices.⁷⁹ $r^{\mathrm{w}}$ quantifies the correlation strength between the strengths of all nodes of the network at two opposite ends of a link. The higher the value, the higher the number of nodes in the network, which tend to link to other nodes with the same or similar strength. $T^{\mathrm{w}}$ is associated with the degree to which nodes in a network tend to cluster together. It is a classical version of the clustering coefficient. The higher the value, the higher the prevalence of clustered connectivity around individual nodes. $D$ is a measure of the mean network degree, i.e., the relation between the potential connections in a network with the actual connections. A high value indicates that many nodes are connected with each other. Finally, $E^{\mathrm{w}}$ is related to the average inverse shortest path length between all pairs of nodes in the network. It is inversely related to the characteristic path length. The higher the value, the more interconnected the nodes are.

3.1.

Changes During All Three Color Stimuli

The results are depicted in Figs. 1(e) and 1(f).

3.1.4.

Neurosystemic functional connectivity

Stimulus-evoked changes in cerebral parameters and systemic physiology correlated statistically significantly for seven pairs of variables: $[{\mathrm{O}}_2\mathrm{Hb}]$ (RVC) versus MWA (RPFC), $r_{\mathrm{s}}=0.505\pm 0.474$; [HHb] (LPFC) versus LF, $r_{\mathrm{s}}=0.424\pm 0.383$; [tHb] (LPFC) versus MWA (RPFC), $r_{\mathrm{s}}=-0.399\pm 0.419$; Q versus MWA (RPFC), $r_{\mathrm{s}}=-0.379\pm 0.440$; [HHb] (RVC) versus MWA (RPFC), $r_{\mathrm{s}}=-0.358\pm 0.393$; [HHb] (LVC) versus HF, $r_{\mathrm{s}}=0.337\pm 0.374$; and [HHb] (LVC) versus Q, $r_{\mathrm{s}}=-0.311\pm 0.320$. This demonstrates the neurosystemic functional connectivity.

Additionally, functional connectivity was shown between the cerebral signals: [HHb] (LPFC) versus ${\mathrm{StO}}_2$ (LPFC), $r_{\mathrm{s}}=-0.921\pm 0.093$; [HHb] (RPFC) versus ${\mathrm{StO}}_2$ (RPFC), $r_{\mathrm{s}}=-0.920\pm 0.088$; [HHb] (RVC) versus ${\mathrm{StO}}_2$ (RVC), $r_{\mathrm{s}}=-0.916\pm 0.099$; [HHb] (LVC) versus ${\mathrm{StO}}_2$ (LVC), $r_{\mathrm{s}}=-0.904\pm 0.145$; $[{\mathrm{O}}_2\mathrm{Hb}]$ (RVC) versus ${\mathrm{StO}}_2$ (RVC), $r_{\mathrm{s}}=0.846\pm 0.087$; $[{\mathrm{O}}_2\mathrm{Hb}]$ (LPFC) versus [tHb] (LPFC), $r_{\mathrm{s}}=0.820\pm 0.107$; and $[{\mathrm{O}}_2\mathrm{Hb}]$ (LVC) versus ${\mathrm{StO}}_2$ (LVC), $r_{\mathrm{s}}=0.813\pm 0.178$. Relatively strong and nontrivial correlation between systemic physiology signals were observed for the following pairs: PP versus PTT, $r_{\mathrm{s}}=-0.607\pm 0.199$; MAP versus PTT, $r_{\mathrm{s}}=-0.550\pm 0.212$; PP versus Q, $r_{\mathrm{s}}=0.537\pm 0.208$; ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ versus MAP, $r_{\mathrm{s}}=0.475\pm 0.400$; HR versus PP; $r_{\mathrm{s}}=-0.416\pm 0.363$; ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ versus PP, $r_{\mathrm{s}}=0.410\pm 0.310$.

All 72 significant correlations are visualized in Figs. 2(c)–2(d).

Fig. 2

Visualization of the correlation analysis. (a) Exemplary correlation matrix of one subject and trial. (b) Combining the correlation matrices of each subject and trial results in a multidimensional correlation matrix. (c) Averaging these 3-D matrices generates the group-averaged correlation matrix. (d) Statistical analysis identifies the significant correlations shown here.

3.2.

Dependence of Changes on Color of Light, Gender, and Age

3.2.4.

Neurosystemic functional connectivity

Figure 3 shows the results for the complex network parameters ($r^{\mathrm{w}}$, $T^{\mathrm{w}}$, $D$, and $E^{\mathrm{w}}$). The parameters were not found to show significant color dependence.

Fig. 3

Results of the complex network analysis. The values for the assortativity, transitivity, density, and efficiency were plotted for the three colors and the three types of correlation analyses performed: (a) cerebral hemodynamics/oxygenation + systemic physiology (i.e., eurosystemic functional connectivity), (b) cerebral hemodynamics/oxygenation, and (c) systemic physiology.

4.1.

Cerebral Tissue Hemodynamics and Oxygenation

The results show that the two cortices assessed, i.e., PFC and VC, react differently to short-term color stimuli. While the hemodynamic responses in the VC were stronger than in the PFC, independent of color, the PFC showed a significant hemodynamic response only to blue. These differing responses in the two cortices reflect different underlying processes.

4.2.

Cardiorespiratory Activity

Color-specific effects in the cardiovascular system and the respiratory system were identified for RR, PRQ, HF, LF, LF/HF, and MWA (RPFC).

There were strong increases in RR for blue and red light and a decrease for green. A negatively correlated change in ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ would have been expected and such a trend, although nonsignificant, is visible in Fig. 1(f). This negative correlation is well known, i.e., for a higher RR more ${\mathrm{CO}}_2$ is exhaled and ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ falls (or vice versa).⁹⁸ An increase in RR and decrease in ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ is a typical pattern during an increase in arousal⁹⁹ but is not directly linked to the emotional state.¹⁰⁰ However, during psychological stress (i.e., negative valence with high arousal), characteristic changes in RR and/or respiratory tidal volume take place.^101,102 The arousal-inducing effect of blue light was also shown by EEG:¹⁰³ A 135-s exposure to light reflected from blue paper elicited a stronger decrease in the $\alpha$ attenuation coefficient than light reflected from red paper.

Blue light elicited a strongly significant decrease in PRQ ($=\mathrm{HR}/\mathrm{RR}$) compared to green, which can be explained mostly by an increase in RR (Fig. 2). An opposite effect was observed by Gerard¹⁰⁴ in long-term light exposure of 10 min. Previously, a 2-h exposure to blue light was shown to increase HR,¹⁰⁵ whereas a 5-min exposure to blue, red, and white light decreased HR.¹⁰⁶ Thus, the temporal structure of the exposure may be relevant.

Concerning HRV, red light elicited an increase in HF and green a decrease in LF/HF. Traditionally, the LF is considered to be related to the sympathetic and the HF to the parasympathetic part of the ANS.^107,108 But in fact both components are influenced by the sympathetic and parasympathetic ANS in parallel,^109–111 and the LF/HF ratio does not relate directly to the sympathovagal balance.¹¹¹ The LF component and thus LF/HF are also affected by changes in respiration.¹¹² The interpretation of HF and LF/HF by themselves is, therefore, not straightforward. An increase in LF/HF was reported for red and blue light, albeit nonsignificantly for blue during 10-min exposure.¹¹³ These results are not directly comparable to our study due to the much longer exposure.

The age-dependent effects observed for the stimulus-evoked changes in the systemic parameters [RR, PRQ, LF, LF/HF, and MWA (LPFC)] are expected due to the age dependence of cardiovascular reactivity,^114–116 activity and reactivity of the ANS,^117–120 and respiration patterns and reactivity.^121,122 Gender effects found for RR, HR, PRQ, ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$, PP, Q, and MWA (RPFC) are also known for the cardiovascular reactivity,^{116,123–127} activity and reactivity of the ANS,^128,129 and respiration patterns and reactivity.^121,122,130

In a follow-up paper, we will report the individual responses in systemic physiological signals observed in the experiment, highlighting that the magnitude and sign of them depends strongly on the individual subject and experimental trial.

4.3.

Electrodermal Activity

It is rather unexpected that stimulus-evoked changes in EDA were not related to color of light, age, or gender, given the strong correlation of EDA with attention, arousal and emotion,^131–135 and the expected color-dependent differences in these psychological factors. This absence may be explained by the fact that we investigated the trend of the EDA, i.e., the tonic SCL and not the rapidly varying changes, i.e., the skin conductance response (SCR). Both components constitute the EDA signal and are related to the sympathetic activity of the ANS.¹³⁶ SCL showed a high intersubject variability, possibly due to different individual psychological states and traits, thus potentially preventing the detection of significant correlations.^137–143 In the future, we recommend applying methods to reduce this intersubject variability.¹⁴⁴

Previous studies found a decrease in the SCL during a 10-min blue-light exposure,¹⁰⁴ a stronger SCR response to red or green light of 1 min compared to yellow or blue,¹⁴⁵ and a stronger SCR for red compared to green light.¹⁴⁶ Interestingly, there were also individual subjects who showed the opposite effect compared to the group.

4.4.

Neurosystemic Functional Connectivity

Although there were several parameters with significant neurosystemic functional connectivity, these were not color dependent. Color dependence, although present in the PFC and in specific systemic physiological signals (RR, HR, PRQ, MAP, HF, LF/HF, and MWA), differs substantially between individuals and is, therefore, masked. To the best of our knowledge, our study is the first ever to assess the neurosystemic functional connectivity.

4.5.

Psychological State

Our study was the first to use the MDBF to determine psychological changes associated with colored-light exposure. The psychological state was not affected by the color of light, time of measurement, age, or gender. One possible reason why no correlation was found might be that the MDBF assessment was performed before and after the experiment, i.e., $>15\min$ before/after the light exposure. Any color-specific effect may have been more prominent during the light exposure. Thus, our finding does not rule out a possible color-dependent effect.

Previously, several color-dependent psychological effects were reported: a lower anxiety state was found for blue and green compared to red and yellow wide-field stimulation in healthy subjects.¹⁴⁷ Red and green light (duration: 1 min) elicited a stronger SCR, indicating arousal, than yellow or blue.¹⁴⁵ The subjective rating of pleasurableness of a color is a nonlinear function of its intensity and saturation.¹⁴⁸ Colors in the green–blue spectral region were rated as more pleasant than those in red–yellow.

4.6.

Strengths and Limitations

4.7.

Implications for Functional Brain Activation Studies

There are several methods that employ cerebral hemodynamics as a biomarker of brain activity. In addition to fNIRS and PET, the most widely applied today is fMRI. Our results demonstrate that even nonstrenuous, perception-based functional brain activity leads to changes in systemic physiology, in particular the MAP, which has previously been shown to influence cerebral hemodynamics. Other relevant parameters, such as HR and RR, were also changed. The latter implies that there should also have been a change in ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ in response to the color stimulations; this effect was only visible at an individual level, however, and was not significant at a group level. ${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$ is the parameter known to affect cerebral blood flow most strongly.¹⁵¹ These changes in several systemic physiological variables could very well be important confounders in functional activation studies using fMRI, PET, and fNIRS. Methods to disentangle the different influences on cerebral hemodynamics are urgently needed.¹⁵⁸ To detect these possible confounders, an SPE approach as employed in this study is required.