1.

Introduction

The relationship between mood and cognition, as well as the neural mechanisms supporting it, has long attracted researchers in psychology and neuroscience.^{1, 2} Behavioral experiments have shown that even mild variations in mood state influence various cognitive functions such as working memory (WM) and cognitive fluency.^{1, 3, 4} Recent neuroimaging studies have implicated the prefrontal cortex (PFC) as one of the key regions that converges mood and cognition in the brain.^{1, 2} For instance, functional magnetic resonance imaging (fMRI) studies have reported that experimentally induced negative moods affect PFC activity during cognitive tasks such as WM and Stroop tasks.^{5, 6} In addition, near-infrared spectroscopy (NIRS) studies focusing on moods under natural circumstances indicate that people having a high level of fatigue or sleepiness showed decreased PFC activity in response to verbal fluency tasks.^{7, 8} However, little is known about how PFC activity during cognitive tasks and natural moods, which can vary across days or weeks, are coupled within individuals. Thus, tracking within-individual fluctuations in natural mood and PFC activity during cognitive tasks is necessary to deepen our knowledge about their relationship.

In our previous study, we used optical topography (OT), which is a noninvasive functional imaging-technique based on NIRS,^{9, 10, 11} to show that a variation in the negative mood state across participants is correlated with their PFC activity for a verbal WM task.¹² OT measures hemodynamic responses in the cerebral cortex under near-natural situations (e.g., sitting position) and allows us to minimize mood modulation due to the experiment itself. In the previous study, the participants’ natural moods were assessed by using the Profile of Mood States (POMS), a self-reporting questionnaire.^{13, 14} As the POMS is used to evaluate a responder's typical mood states rather than personality traits,¹⁵ one can assume that the observed correlation would reflect a state-dependent effect. However, because of the experimental design used in the previous study, i.e., across-subject design, we could not deny the possibility of the contribution of individual trait factors (e.g., personality traits or dispositional moods) in the results.

One solution to dissociate the state-dependent effect found in the previous study¹² from certain trait factors is examining whether PFC activity changes in correlation with his/her mood state using a within-individual design. In this study, we repeated three experiment sessions for each participant at 2-week intervals to trace the time-to-time mood fluctuations within individuals, which enabled us to identify the contribution of each participant's depressed mood state to the variations in their PFC activity independent of trait factors.

2.

Materials and Methods

2.4.

OT Measurement

The PFC activity was measured using an OT system (ETG-7100, Hitachi Medical Corporation, Japan). The system is based on a continuous wave (cw) NIRS technique.^{9, 10, 11} Although other techniques such as time-resolved and frequency-domain techniques^{17, 18, 19} are possibly advantageous for accurate estimation of brain tissue oxygenation and for depth-resolved analysis,^{20, 21, 22} we used the cw NIRS technique because it has proved useful in previous practical studies.^{23, 24, 25, 26, 27, 28}

The OT system light sources were cw laser diodes with a wavelength of 695 or 830 nm. The average power of each source was 2 mW (for both wavelengths), and the two wavelength lights were irradiated on the skin through an incident optical fiber bundle (1.5-mm diameter). The signal intensities of the transmitted light were detected through a detection optical fiber bundle located 30 mm from the incident position at a sampling rate of 100 ms. We used 15 incident optical fiber bundles (sources) and 15 detection optical fiber bundles (detectors), which were arranged alternately, separated by 30 mm, in a 3×10 lattice pattern and embedded in a soft silicon holder. This resulted in a configuration with 47 measurement positions (defined as channels: chs), each corresponding to a midpoint of source-detector pair (Fig. 1). The holder was placed on each participant's forehead, and the optical fiber bundles contacted the skin on the forehead. Although we used a different system (ETG-4000, Hitachi Medical Corporation, Japan) in our previous study,¹² the basic specifications were common to both systems. The only difference between the two measurements was the width of the measurement area; the present study measured 47 chs (6 × 27 cm) using a 3×10 optode arrangement while our previous study measured 52 chs (6 × 30 cm) using a 3×11 optode arrangement. The reduced width in the present study was due to a limitation in the system software. In addition, the primary activation areas identified in the previous study¹² were in the range of the measurement area of the present study.

Fig. 1

Arrangement of measurement positions (47 channels) in MNI space, estimated using probabilistic registration method (Ref. 30).

To estimate the locations of the OT channels in the Montreal Neurological Institute (MNI) space, we used the probabilistic registration method.^{29, 30} Prior to the experiment, we corrected the sample data for the three-dimensional (3D) coordinates of the 30 optode locations and scalp landmarks (in accordance with the international 10 to 20 system: Fp1, Fp2, Fz, T3, T4, C3, C4) for the 11 participants. The data were recorded with a 3D-magnetic space digitizer (3D probe positioning unit for OT system, EZT-DM101, Hitachi Medical Corporation, Japan).

3.

Results

Behavioral data (accuracy and reaction time: RT) for the WM tasks are listed in Table 1. An analysis of variance (ANOVA) with session order (1st, 2nd, and 3rd) and WM task (verbal and spatial) as within-participant factors revealed that there was no significant main effect of session order for both accuracy (p = 0.47) and RT (p = 0.29). On the other hand, a significant main effect of WM tasks was indicated for RT (p = 0.005), which means that RT for the verbal WM task was longer than that for the spatial WM tasks. However, the verbal WM task tended to show higher accuracy than the spatial WM task, though it was not significant (p = 0.14). The mean and standard deviation (SD) of POMS_D were 2.47±2.67 for the 1st session, 1.88±2.23 for the 2nd session, and 2.65±2.91 for the 3rd session.

Table 1

Mean accuracy and RT for WM tasks.

To examine the basic activation pattern during WM tasks, we first conducted across-participants t-test of the activation values (Act_V and Act_S) for each session [Fig. 2a]. Channels in the bilateral dorsolateral PFC showed significantly positive values in all cases (p < 0.05). In particular, ch22 in the right hemisphere and ch26 in the left hemisphere consistently showed prominent activation (p < 0.01) for all sessions for both WM tasks [marked channels in Fig. 2a]. For these main activation channels, a session dependency (order effect) on the activation values was tested using the repeated-measures ANOVA with session order (1st, 2nd, and 3rd) and WM task (verbal and spatial). This analysis showed no main effect of the session order (ch22: p = 0.96, ch26: p = 0.28) with no interaction between session order and WM task (ch22: p = 0.59, ch26: p = 0.76), indicating the same activation pattern was reproduced for all sessions regardless of the WM type. The similarity of the time courses in activation signals among sessions is shown in Fig. 2b. In addition, the main effect of WM tasks indicated no significant effect in the two channels (ch22: p = 0.90, ch26: p = 0.15).

Fig. 2

Reproducibility of cortical activation for verbal and spatial WM tasks. (a) t-maps of activation values among participants. Channels with significant activation (determined with one-sample t-test against zero) are indicated with color scale shown below (two-tailed, p < 0.05). (b) Mean time courses of reconstructed oxy-Hb signals for representative channel (ch22 in Fig. 1). Yellow bars perpendicular to the x-axis indicate activation period (5-s duration). Error bars indicate standard errors across participants.

To reveal the relationship between within-individual changes in depressed mood state and PFC activation for WM tasks, we calculated the Spearman rank correlation coefficient (ρ) between ΔPOMS_D and ΔAct_V (or ΔAct_S) for each channel. The ρ-maps for Δ1st to 2nd and Δ2nd to 3rd are shown in Fig. 3a. We regarded the channels, in which the p-values were less than 0.05 for both Δ1st to 2nd and Δ2nd to 3rd, as significantly correlated. This result indicated significantly negative correlations between ΔPOMS_D and ΔAct_V for channels mainly located in the left dorsolateral PFC (ch26 and ch35) and around the left pre-motor and supplementary motor cortex (ch8 and ch19). In contrast, the correlation coefficients between ΔAct_S and ΔPOMS_D did not reach statistical significance (the higher p > 0.21). These results were consistent with our previous results.¹² The distributions of ρ-values among¹² channels are shown with histograms [Fig. 3b] to demonstrate the discrepancy between ΔAct_V and ΔAct_S.

Fig. 3

Relationship between variations in depressed mood state (POMS_D) and those in activation values for verbal and spatial WM tasks (Act_V and _Act_S). (a) Statistical ρ-maps indicating correlation coefficients (Spearman ρ) between ΔPOMS_D and ΔAct_V (or ΔAct_S). Channels that showed correlation coefficients (p < 0.05) in both Δ1st to 2nd and Δ2nd to 3rd are marked with solid circles. In addition, channels that showed correlation coefficients (p < 0.10) in both Δ1st to 2nd and Δ2nd to 3rd are marked with dashed circles. (b) Histograms illustrating distributions of ρ-values for each WM task. Blue vertical line denotes median ρ-value for verbal WM task, and red vertical line denotes median ρ-value for spatial WM task among 47 channels.

Figure 4 shows the negative correlation between ΔPOMS_D and ΔAct_V in the left dorsolateral PFC (ch26 and ch35), where clear WM-related activity was demonstrated (Fig. 2). The p-values for the correlation coefficients were 0.0315 and 0.0292 for Δ1st to 2nd and Δ2nd to 3rd, respectively. These p-values mean that the probability of obtaining these correlation coefficients is 0.00092 (0.0315×0.0292), which is below the corrected p-value of 0.05 for all 47 channels (0.05/47 = 0.00106). This indicates that decreased depressed mood score (POMS_D) is correlated with increased PFC activity for the verbal WM task (Act_V).

Fig. 4

Scatter plot showing relationship between ΔPOMS_D and ΔAct_V in left dorsolateral PFC (mean of ch26 and ch35). (a) Regression line of data for Δ1st to 2nd. (b) Regression line of data for Δ2nd to 3rd.

The within-individual fluctuations of POMS_D and Act_V (mean of ch26 and ch35) are shown in Fig. 5 for all participants, where the POMS_D is denoted on the reversed Y-axis. Although the sensitivities might be different, the basic change pattern for Act_V and reversed POMS_D across sessions appears to be similar in most participants.

Fig. 5

Monthly fluctuations in POMS_D and Act_V (mean of ch26 and ch35) shown for every 2 weeks for all participants. Graphs with small “f” in top-left corner indicate data from females.

4.

Discussion

We used OT to show that fluctuations in the depressed mood state within individuals are correlated with their PFC activity during a verbal WM task. The results are consistent with those of our previous study¹² and confirm the state-dependent feature of PFC activity, which cannot be explained by the effects of certain trait differences among individuals.

The mean activation patterns for WM tasks were well reproduced, as shown in Fig. 2. The central channels of the activation were located in the dorsolateral PFC, which is consistent with other fMRI studies.³⁶ Moreover, the strong similarity between the activation patterns found in the present study with those found in our previous study¹² suggests that OT measurement is highly reliable for detecting cortical activity for these WM tasks. We found no main effect of session order for the PFC activity, suggesting that there was no simple order effect in our measurements. In addition, the oxy-Hb signals in a representative channel demonstrated overlapped time courses among three sessions. These results support that relevant cortical activity for WM functions can be extracted using ICA.³⁵ Regarding differences between verbal and spatial WM tasks, the activation regions for the verbal WM task appeared to be larger than those for the spatial WM task. However, no significant main-effect of WM tasks was found in the channels that showed significant correlation coefficients between ΔPOMS_D and ΔAct_V (ch8: p = 0.94, ch19: p = 0.12, ch26: p = 0.15, ch35: p = 0.07). In addition, the behavioral data did not reveal significant differences between the two tasks in terms of difficulty. The RT for the verbal WM task was longer than that for the spatial WM tasks, which was possibly due to a difference in the task strategy, as participants showed higher accuracy for the verbal WM task than for the spatial WM task.

The basic activation patterns were thus similar in both WM tasks, but the correlation analysis revealed a contrast between tasks in relation to the depressed mood state (POMS_D). The negative correlation between ΔPOMS_D and ΔAct_V were well reproduced in the two independent differences between successive sessions (Δ1st to 2nd and Δ2nd to 3rd), while there were consistently no significant correlations between ΔPOMS_D and ΔAct_S. This result suggests that the PFC activity induced by a verbal WM task is selectively related to an individual's depressed mood state. In addition, the significant channels were mainly located in the left dorsolateral PFC, which is consistent with previous results.¹² Thus, we found that the interindividual variation in PFC activity during a verbal WM task¹² is not simply an indirect reflection of individual differences in trait factors; it actually reflects time-to-time fluctuations in the depressed mood state of an individual. Although the inherent limitation of transcranial NIRS requires that we consider the effect of hemodynamic changes in the extracerebral tissue, such as changes in skin blood flow,^{37, 38} our task paradigm did not impose intense physical or psychological demands, which could induce systemic changes. Moreover, the effect of extracerebral hemodynamic changes could not account for our finding that there is selective negative correlation between ΔPOMS_D and ΔAct_V in a localized area.

We plotted the within-individual fluctuations of POMS_D and Act_V (Fig. 5) to indicate the feasibility of using OT signals for assessing the depressed mood state of healthy participants. While Act_V apparently fluctuated in parallel with POMS_D within individuals, we could not estimate its effectiveness or validity because the depressed mood state is a psychological construct and difficult to define. To uncover new ways of using optical topography to obtain a potential state marker for an individual's mood state, we need to further clarify the interactions between the depressed mood state and PFC activity on the basis of neurophysiology, such as by using a pharmacological approach.^{39, 40, 41}

5.

Conclusion

We demonstrated that prefrontal cortex activity during a verbal working memory task varies depending on the participant's depressed mood state, independent of trait factors. This suggests that using optical topography to measure PFC activity during a verbal WM task can be used as a potential state marker for an individual's depressed mood state.