Association of Compensatory Mechanisms in Prefrontal Cortex and Impaired Anatomical Correlates in Semantic Verbal Fluency: A Functional Near-Infrared Spectroscopy Study

Article information

Psychiatry Investig. 2024;21(10):1065-1075

Publication date (electronic) : 2024 September 11

doi : https://doi.org/10.30773/pi.2023.0447

Hae-In Kim¹, Sungman Jo¹, Minjeong Kwon¹, Ji Eun Park^1,2, Ji Won Han², Ki Woong Kim^,1,2,3,4

¹Department of Brain and Cognitive Sciences, Seoul National University College of Natural Sciences, Seoul, Republic of Korea

²Department of Neuropsychiatry, Seoul National University Bundang Hospital, Seongnam, Republic of Korea

³Department of Health Science and Technology, Graduate School of Convergence Science and Technology, Seoul National University, Seoul, Republic of Korea

⁴Department of Psychiatry, Seoul National University College of Medicine, Seoul, Republic of Korea

Correspondence: Ki Woong Kim, MD, PhD Department of Neuropsychiatry, Seoul National University Bundang Hospital, 82 Gumi-ro 173beon-gil, Bundang-gu, Seongnam 13620, Republic of Korea Tel: +82-31-787-7432, Fax: +82-31-787-4058, E-mail: kwkimmd@snu.ac.kr

Received 2023 December 20; Revised 2024 May 16; Accepted 2024 July 7.

Abstract

INTRODUCTION

Semantic verbal fluency (SVF) is a type of cognitive performance to measure verbal ability and other accompanying cognitive processes. Including a language ability which involves the production of words, SVF utilizes various types of cognitive abilities such as executive function to select the appropriate words related to specific categories, mental flexibility to switch between categories, and semantic memory function to retrieve semantic information [1,2]. These abilities are frequently impaired in the early stage of dementia [3]. Raoux and colleagues reported that switching scores were reduced in future Alzheimer’s diseases (AD) patients and kept declining during preclinical phase of AD [4]. Clark and colleagues [5] proposed that a decreased ability to switch between semantic categories may foretell later global cognitive decline in non-demented older adults as well. Consequently, it is postulated that a disrupted SVF network can predict impairment in global cognitive functioning. Indeed, the test for SVF is widely regarded as a reliable tool for screening dementia including AD [3,6]. In this regard, identifying declines in SVF could aid in the early diagnosis of AD.

From neuroimaging studies with healthy individuals, anatomical correlates underlying SVF have been mostly observed in left frontal and left temporal lobe. Left frontal lobe as a whole works as a crucial role in SVF, but the role of left inferior frontal gyrus (IFG) was more specifically and consistently demonstrated as it directly reflects the semantic-related cognitive process compared to the other SVF associated frontal regions [7-9]. Regarding the specific role of left frontal inferior gyrus, it involves in semantic processing both during language comprehension and production as well as selection of semantic knowledge [1,10]. Retrieval and storage of semantic knowledge are suggested to be the roles of the left temporal regions [7]. Accordingly, neural correlates of SVF also have been investigated extensively using neuroimaging techniques like functional magnetic resonance imaging method (fMRI) or functional near-infrared spectroscopy (fNIRS). SVF has been shown to reliably elicit bilateral functional hemodynamic responses in frontal lobe but pronounced left hemisphere within middle frontal gyri, IFG, fronto-temporal regions, and left temporal gyrus in cognitively normal adults [1,8,9,11].

Previous structural neuroimaging studies of individuals with ischemic stroke or traumatic brain injury who had a direct lesion in the left hemisphere have reported that structural abnormalities were associated with SVF deficits [12-14]. Further, impaired SVF in aged adults and AD patients was in commonly related to the bilateral cerebral atrophy besides the atrophy in the major correlates of SVF, and in the extra parts of cerebral regions (i.e., Parietal cortices) [15-17]. Along with these findings, recent several studies with fNIRS have found that individuals with impaired SVF, such as mild cognitive impairment (MCI) or AD showed a shared abnormal pattern of prefrontal activation during the SVF task (SVFT), which is loss of left lateralization [18-21]. There were also different patterns of prefrontal activation during the SVFT from to the normal individuals with no SVF impairment and from to each clinical group. In the individuals with AD, who have advanced cognitive impairment, prefrontal regions were less activated during the SVFT in both hemispheres than in cognitively normal individuals [18,20,22]. This reduced activation in bilateral frontal lobes caused absence of lateralization to the left hemisphere. However, in those with mild SVF impairment, left IFG and other prefrontal regions related to SVF were hypo-activated while increasing the hemodynamic activity in dorsolateral prefrontal cortex in either right or left hemisphere than in cognitively normal individuals [19,21,23]. This suggested that the inefficient utilization of primary neural network due to neuronal loss may be compensated by activation of other non-dominant to SVF regions in prefrontal cortices in early stage of cognitive impairment [24-26].

Yet, Price and Friston [27] reported that the capacity to compensate in neural resources is limited and the ability to manage the resources were inversely proportional to the severity of cognitive impairment. As neural damage progresses, the decline in both cognitive processing efficiency and capacity hinders neuronal recruitment and diminishes compensation capabilities, leading to overall poorer performance. Additionally, the presence of lateralization and compensatory activation is notably found in the frontal lobes, rather than in the temporal lobes in overall span of cognition. Therefore, individuals with modestly impaired SVF but no noticeable cognitive decline have ability to compensate for the effects of neuronal loss in the IFG areas by recruiting additional neural resources to superior frontal areas as substitutes for performing the SVFT. However, no study to date has directly demonstrated an association between the degree of neuronal loss in the SVF-dominant region in prefrontal cortex (IFG) and the level of functional activation in the dorsolateral prefrontal cortex during the SVFT in the individuals with cognitively normal but subtle SVF deficits.

The aims of this study are 1) to determine the pattern of prefrontal activation during the SVFT. We chose prefrontal regions because the presence of lateralization and compensatory activation is known to be notably found in the frontal lobes, rather than in the temporal lobes regardless of cognitive decline level [28] and 2) to examine the influence of cortical thickness in the SVF dominant prefrontal cortex and the level of prefrontal activation during the SVFT in individuals with mild level SVF impairment in cognitively normal individuals.

METHODS

Task paradigm

Participants were instructed to sit on a comfortable chair and to avoid movement as much as possible. This study consisted of 3 sessions with resting phases for recording resting-state Blood-Oxygen-Level-Dependent (BOLD) signals and task phases of each task, a SVFT and control task for recording task-elicited BOLD signals (Figure 1A). There were around 15 s of intervals between each session for reading the instruction displayed on a monitor for next session. A session consisted of 4 blocks. There also were a several seconds of interval less around 10 s between each block for reading the instruction displayed on a monitor for next block. Each block is equivalent to 2 periods of resting, one SVFT, and one control task. A resting period was assigned before and after the SVFT, and the control task was assigned after the second resting period ended. In this order, a session started with resting and ended with the control task. Resting-state BOLD signals were recorded during each resting period for 30 s, respectively while the participants stayed still their eyes open fixing on a fixation cross symbol displayed on a monitor. The task-elicited BOLD signals were recorded during a SVFT and control task for 30 s each. In each session, the entire BOLD signals were recorded for 120 s. Three experimental sessions were 6 min (instruction time is not included).

Figure 1.

Experimental paradigm and fNIRS channels. A: Task paradigm for the fNIRS measurement. B: Configuration of the probeset with positions of the measured fNIRS channels located between each transmitter (blue-colored numbers) and detector (red-colored numbers) over the prefrontal cortex in accordance with the International 10–20 System. Boxes with orange colors indicate superior frontal regions in the left and right hemispheres and the ones with light orange indicate the midline superior frontal regions. Blue-colored boxes indicate rostral middle frontal cortices in the left and right hemispheres. SVFT, semantic verbal fluency task; Control, control task; fNIRS, functional near-infrared spectroscopy.

SVFT had different categories for a block in each session: animals, vegetables, market items. The control had the same simple category for a block in every session, weekdays (Figure 1A). During the SVFT, participants had to generate as many words that related to the given semantic category as possible within 30 s. The task measures how much information can be retrieved from the categorization and memory repository of text. During the control task, participants had to slowly produce and repeat the names of weekdays for 30 s. Control task was designed to compare the induced brain activation by the control task and the SVFT, the task which requires a higher level of cognition and manifests semantic knowledge and fluency. Resting was performed to establish participants’ baseline brain activation.

fNIRS

We recorded the changes in prefrontal hemodynamic activity with BOLD signals over the experiment with a continuous wave system 27-channel Brite24 (Artinis Medical Systems B.V., Elst, The Netherlands). Changes in the concentrations of oxyhemoglobin (HbO2) and deoxyhemoglobin (HbR) across the channels were recorded in the units of micromolar-millimeter (μM×mm) at 10 Hz of sampling rate by 2 near infrared wavelengths, 760 nm and 850 nm, respectively.

We utilized the fNIRS device consists of 10 light transmitters and 8 detectors with a 3.0 cm of inter-optode distance, and the fNIRS data were collected at 27 measurement channels located between each transmitter and detector (Figure 1B). The device was adjusted based on landmarks from the International 10–20 system. Specifically, for the most superior side of the probeset, channel 16 and 17 were positioned between the marker FCz and Fz, and 16 to the right and 17 to the left. Moreover, for the most inferior side of the probeset, the corresponding channels were positioned between the FPz and AFz, and 6 to the right and 18 to the left. This spatial information was used to calculate projections of the superficial optode to the cortex and position channels onto a standard brain template in standard stereotactic brain coordinate systems of montreal neurological institute using AtlasViewerGUI, a visual graphical user interface contained within Homer2 (https://homer-fnirs.org/) [41,42]. Afterwards, label assignment for the corresponding channels was defined by the automated anatomical labeling (AAL) [43], and fNIRS channels recorded BOLD signals within the superior frontal and middle frontal cortex according to the AAL label assignment. Hereby, the ROI on the right superior frontal cortex covered channels #5, 6, 7, 8, 9, 10, 11, and 16. Similarly, the ROI on the left superior frontal cortex covered channels #14, 17, 18, 19, 21, 22, 23, and 24. The ROI on the right middle frontal cortex covered channels #1, 2, 3, and 4, and the left middle frontal cortex covered channels #20, 25, 26, and 27. The ROI on the midline superior frontal regions covered channels #12, 13, and 15 (Figure 1B).

The fNIRS data were preprocessed using the open source packages, MNE-Python version 1.6 (www.martinos.org/mne/stable/index.html) and MNE-NIRS version 0.5.0 (https://mne.tools/mne-nirs/stable/index.html). The optical intensity signals were first transformed into the time series of HbO2 and HbR concentration changes using the modified Beer-Lambert law [44]. A bandpass filter was applied to converted signals with cutoff frequencies of 0.01 Hz and 0.09 Hz to remove slow drifts and high frequency fluctuations of motion and physiological noise [45-47]. We investigated only the HbO2 data as validated BOLD signals in the present study because HbO2 offers better signal-to-noise ratio and demonstrates a stronger relationship with the BOLD compared to HbR, and thus has better reliability and sensitivity to verbal-related changes in cerebral blood flow [48-50].

After preprocessing, we used task blocks as epochs for analyzing HbO2 concentration changes during the resting state and tasks. We obtained 6 epochs from resting state HbO2 concentration and 3 epochs from SVFT and control task, respectively. Each epoch had the same length of a task block, which is 30 s. To find the HbO2 concentration changes related to each task and resting state, we calculated weighted averaging for HbO2 concentration across time points of epochs in tasks and a resting state by block averaging. Block-averaged HbO2 concentration in each epoch was then corrected its baseline to a zero. The baseline-corrected data were then averaged across channels, and participants. Previous fNIRS studies have suggested a lag in hemodynamic activity like earliest activation starting from 5 s after task onset and sharp activation at around 5–10 s after onset [45,51]. Therefore, to observe the level of activation in a full length as possible, we included the time course of 10 s after the epoch was ended. The purpose for this analyzing strategy was to observe any visual differences among the change of HbO2 concentrations of resting state, control task, and SVFT (Figure 2) not to assess the response value to the task. Instead, we utilized the general linear model (a model-based statistical analysis tool) to identify hemodynamic responses to the tasks because the estimation of the response tends to be more accurate and robust compared to the block-averaging technique as it can derive hemodynamic response function (HRF) considering the entire time course of HbO2 fluctuation in addition to the task and resting periods [52,53]. The HRF is used to serve as a reference to estimate the changes in HbO2 signals during the task [54]. The formula is as follows:

Figure 2.

Group-averaged hemodynamic responses in resting (A), semantic verbal fluency task (B), and control task (C).

(1) Y=Xβ+ε,

(2) X=h(t)⊗s(t)

where Y represents the temporal profile of the measured HbO2, β is the estimated amplitude of the changes in HbO2, and ε represents the residual owing to the difference between the measured signals and the predicted model. X is the stimulation-specific predicated response, which is expected to match the temporal profiles of the measured hemodynamic signal, HRF; h(t) represent the canonical HRF, and s(t) is the stimulation-specific boxcar function for a given task. A convolution matrix of h(t) and s(t) provides the HRF. Fitting the equation (2) calculates β, statistical t-value representing the statistical significance of the changes in HbO2 (HRF) with respect to the baseline at each respective channel [55]. We modeled the baseline drift with a 2nd order polynomial. After that, we obtained the group-level β values at each channel which were averaged across participants by group-level hemodynamic analyses. All analyses were performed using the MNE-Python version 1.6 and MNE-NIRS version 0.5.0 [56].

RESULTS

Their mean age and educational level of the 54 participants was 71.6±5.4 years (range: 60–80) and 12.3±3.7 years (range: 6–18), respectively. About two thirds of them (65.6%) were women. The neuropsychological characteristics of the participants are summarized in Table 1. All participants were diagnosed with MCI and scored -1.5 standard deviation or less on one or more cognitive tests compared with age-, sex-, and education-adjusted normative data of cognitively normal Korean elderly. Considering that the SVF test z scores of the participants ranged from -2.57 to 1.64, the SVF of the participants ranged from moderately impaired to normal.

Table 1.

Neuropsychological performance of the participants (N=54)

As shown in Figure 2, the block-averaged concentration change of HbO2 was greatest in the SVFT phase, followed by the control task phase. During each task phase, HbO2 began to increase about 10 seconds after the start of the task and began to decrease about 10 seconds after the end of the task because of delayed hemodynamic response [45,51].

As summarized in Table 2, the main effect of task was significant in 23 channels by rmANOVA. In post hoc comparisons, the averaged change in β coefficient of HRF in the SVFT phase was significantly greater than that in resting phase in 21 channels, and greater than both the resting and control task phases on 5 channels (3 channels in left superior frontal regions [channel 17, 21, and 24] and 2 channels on midline superior frontal regions [channel 13 and 15]). These 5 channels were categorized as channels with SVA-S HbO2 as they showed the greater level of SVFT HRF β coefficient compared to the other tasks.

Table 2.

Comparisons of hemodynamic responses in resting, control task, and semantic verbal fluency task phases

As shown in Table 3, SVA-S HbO2 activation, the difference of HRF β coefficient between that of SVFT and control task, in these 5 channels in superior frontal cortex was inversely correlated with cortical thicknesses of IFG areas, pars triangularis and pars opercularis. The SVA-S HbO2 activation in channel 13, 15, and 24 was also inversely correlated with cortical thickness of pars orbitalis. The correlation was statistically significant between channel 17 and pars triangularis’ thickness (r₍₅₄₎=-0.304, p=0.025).

Table 3.

Correlations of left inferior frontal cortical thickness with hemodynamic responses that were higher in the semantic verbal fluency task phase compared to both resting and control task phases

DISCUSSION

In the current study, we examined the neural processing during the SVFT and neural activation specific to semantic verbal ability in individuals with mild SVF impairment using fNIRS. We also explored the association between left IFC and function in prefrontal areas which are known to be working for SVF, combining the HbO2 activation solely reflects the SVF and MRI data of grey matter cortical thickness in those individuals.

We targeted the individuals who have normal or similar to normal but have objectively mild impairment in SVF as we intended to find the response of neural utilization in the presence of the atrophy of SVF-related regions. As described in Table 1, the results of neuropsychological assessments clearly support that the group represents the older adults with almost no cognitive decline but mildly declined SVF. Therefore, we can be reasonably confident that outcomes from our data demonstrated the characteristics of target population, normal individuals with mild deficit of SVF.

Our findings suggested can be summarized as follows: 1) individuals with normal cognition and mild SVF impairment showed the prefrontal activation during SVFT. The activation level was higher than that of resting state, and that of the control task, which was to repeatedly produce the name of weekdays. The averaged concentration changes of HbO2 during the tasks showed that the SVF the highest, followed by control task, and resting (Figure 2). 2) As shown in Table 2, participants showed a difference in β coefficient of HRF in activated regions between the task conditions, i) bilateral activation on the entire prefrontal regions during SVFT compared to the resting state and ii) activation exclusive to SVF (SVA-S activation) on the midline and left superior areas in prefrontal cortex. 3) SVA-S activation of those areas are influenced by reduced left pars-triangularis thickness in the individuals with modest level of impairment in SVF.

Numerous previous studies have demonstrated that healthy controls or younger adults exhibit an elevation in HbO2 activation corresponding to increased cognitive demands, demonstrating efficient utilization of neural resources when engaging in more challenging tasks [57-59]. In line with this, our findings, illustrating an increase in HbO2 from a control task to the SVFT, suggest that participants in our study can enhance HbO2 levels in response to escalating cognitive demands. Two possible explanations can be considered. Firstly, individuals in the early stages of cognitive decline may still possess the capacity to discriminate cognitive levels between the repetition of ordinary words (weekdays) and the generation of words within specific categories, effectively deploying neural resources accordingly. Alternatively, participants may have found the SVFT relatively manageable, allowing for the effective utilization of a slightly reduced number of neurons to meet the increased cognitive demand.

Furtermore, we successfully confirmed the HbO2 activation specifically associated with SVF, referred to here as SVA-S activation. Even though we observed the hemodynamic activity during the SVFT, we believed that this activation did not purely reflect the response to semantic verbal ability. To identify the ‘true’ HbO2 activation related to semantic fluency, we subtracted the effect of word generation, which lacks semantic knowledge and other traits of semantic fluency such as mental flexibility or semantic memory retrieval, using HbO2 activation during the control task. While many functional studies using fNIRS or fMRI have reported HbO2 activation levels during the SVFT compared to the non-task (resting) phase [18,20,23], we eliminated potential confounding effects and obtained results expected to be genuinely elicited by the task.

Consequently, the activation pattern identified in the SVA-S activation suggests that the midline superior and left superior prefrontal regions may represent the neural correlates underlying semantic fluency in individuals with mild SVF deficits (Table 2). In contrast to the semantic fluency correlates in healthy younger adults, where left inferior, medial, and middle frontal regions are involved [8,9,11], our finding diverge as ours, channel #13, 15, 17, 21, 24 are associated with superior frontal regions. Herrmann et al. [60] reported reduced activation in IFG during the verbal fluency task in elderly subjects compared to younger individuals. Heinzel and colleagues [61] also demonstrated that age negatively correlates with activation in the IFG and positively correlates with dorsolateral frontal activation. In this context, the regions with semantically working SVA-S activated undergo cortical reorganization as cognition declines.

Subsequently, we observed the association between prefrontal HbO2 SVA-S activation in channel #17 and reduced cortical thickness in pars-triangularis in the context of similar behavioral performance. The pars triangularis, the anterior-ventral parts (BA 45) of IFG, has consistently been linked to semantic linguistic processes, playing a role in the retrieval of words prompted by semantic cues [10,62,63]. Considering its role in semantic fluency, our correlation between pars-opercularis and SVA-S dorsolateral prefrontal HbO2 activation which is conjectured to reflect the genuine of semantic knowledge is highly reliable. Also, this hints that thinner cortices in anterior part of IFG, especially in pars-triangularis in our study, let recruiting supplementary neurons for processing semantic knowledge from other regions as compensatory mechanisms, where the activation in those regions initially not expected to be involved in the task. Park and Reuter-Lorenz [64] proposed the STAC (Scaffolding Theory of Aging and Cognition) which views this usage of additional neural resources in prefrontal cortex is an adaptive response that engages in compensatory scaffolding to the declined neural structures and function in order to execute a particular cognitive goal, SVFT in our study, as similar to or slightly worse than those with thicker cortices in left IFG. Therefore, this alternate response to the degradation of primary task network mitigates decline in cognition inferred to be occurred in the in older adults who are cognitively normal or comparable to normal.

This study has several limitations. First, the participants lack the control group to compare with. The absence of a normal SVF control group makes it challenging to compare the degree of abnormal activation and structure in our study participants against to those with normal SVF. Previous case-control studies comparing healthy controls with individuals with mildly impaired SVF, however, have demonstrated that SVF reliably elicits bilateral activation in the frontal lobe, with pronounced left middle frontal gyri and left IFG in healthy younger individuals. Conversely, older adults or those with MCI exhibited reduced activity in the left IFG and increased activity in the right middle frontal or left superior frontal cortices [8,19-21,23]. Although we did not collect data from healthy controls in our study, we observed an altered pattern of prefrontal activation that aligns with findings from studies comparing SVF impairment groups with healthy controls. In order to comprehensively investigate the extent of abnormality of clinical groups, future research should consider including a healthy control population.

In addition, the ROIs are limited. Due to the limitation of fNIRS measurements to the dorsolateral prefrontal regions, we could only examine the influence of the left IFG on superior frontal areas. Given that semantic fluency performance may engage regions outside the frontal areas [11,65], participants may display altered activations in other brain regions. Future studies will explore similar associations throughout other cortical and subcortical regions. Finally, this study did not include a connectivity analysis. Considering the local dysfunction in prefrontal brain areas associated with the physiological mechanisms underlying SVF impairment, it is crucial to emphasize the functional connectivity of these regions in individuals with impaired SVF. Future research should focus on investigating the network-level dynamics within the prefrontal areas to gain a deeper understanding of their role in SVF impairment.

Nevertheless, this study is significant in that it is the first, to our knowledge, to establish a relationship between grey matter thickness abnormalities in the prefrontal cortex, a region well-known to be associated with SVF, and abnormalities in functional activation related to SVF. Additionally, the study holds importance as it utilized grey matter as a measure, known to better reflect the BOLD signal compared to white matter [66-68] to find this relationship.

In conclusion, our study utilized fNIRS to investigate neural processing during the SVFT and the specific activation associated with semantic verbal ability in cognitively normal individuals with mildly impaired SVF. We also explored the intricate relationship between prefrontal structure and function, focusing on areas known to contribute to the processing of semantic knowledge. Notably, our findings suggest that individuals with normal cognition may sustain SVF ability through a recruitment of additional neuronal networks in left superior frontal regions to compensate for malfunctioning primary networks, pars-triangularis. This implies a potential adaptive strategy in the face of cognitive decline, allowing for the preservation of SVF ability even with a diminished number of neurons in the task network.

Article information Continued