In this study, 46 healthy French-speaking individuals divided into two groups of elderly people (n = 23), aged 65–75 (mean age = 69.6 ± 4.1), and young people (n = 23), aged 20–35 (mean age = 23.4 ± 2.7), were recruited. Because language knowledge is embedded in the social and cultural context, we restricted our participants to French speakers from Quebec. The elderly cohort was chosen from this specific age bracket because of the delicate transition to old age ( >65) and the increased prevalence of cognitive decline (from 4.97 to 24.19%) after the age of 80 (35). The study was approved by the ethics committee of the Institut universitaire de gériatrie de Montréal (IUGM) and all participants gave their written consent. Exclusion criteria were claustrophobia, hypertension or any cardiovascular disease, smoking, thyroid dysfunction, diabetes, taking any medication known to be vasoactive, as well as psychiatric or neurological illness. Participants were all right-handed according to the Edinburgh Handedness Inventory (36, 37). For the measurement of baseline cerebral blood perfusion, they were also asked to abstain from drinking coffee the day of acquisition (38, 39).

Participants were screened for their level of cognitive performance by standardized cognitive assessments including the Trail Making Test A/B (40), the Montreal Cognitive Assessment (MoCA; (41), and five subtests of the short-form Wechsler Adult Intelligence Scale (WAIS-III; (42–44), namely Vocabulary, Block Design, Similarities, Matrix Reasoning, and Direct and Inverse Digit Spans. In this way, it was possible to exclude those with mild cognitive decline according to age-corrected norms. These tasks assess phonological short-term memory storage as well as processing capacities and evaluate general intellectual ability, planning, visual exploration, attention, mental flexibility, and verbal inhibition.

The activation task represents a robust, well-studied lexical-semantic task: lexical decision (45). Stimuli were chosen from the specific categories of non-action words (nouns) denoting non-living objects in order to isolate the peripheral effects in networks associated with semantic processing. Words generated from a French lexical database (OMNILEX database from the Cognitive Psychology of Language Laboratory, University of Ottawa, Canada) were matched according to their lexical frequency, grammatical category (nouns), age of acquisition, orthographic structure, and length in letters. It is important to note that, in visual lexical decision tasks, word length affects reaction time (RT), with stable RTs for words four to six letters long (46). Concreteness of the words (abstract vs. concrete) was manipulated by the imageability index on a scale of 1–7 to investigate the effect of word imageability within a lexical decision test (n = 60 for each category). Pseudo-words were then created from the real words (n = 120) by changing two consonants. All items (words and pseudo-words) were then matched by bigram frequency and length in letters (Lexique database, Paris Descartes University, France). A pilot study of 15 young adults was done prior to the main study to eliminate outliers within each category. Participants were presented with words and pseudo-words on the screen and were instructed to answer whether or not the letter string constituted a real word. Each trial started with a fixation point (+) that appeared at the center of the screen and was followed by the stimulus. A blank screen provided time to answer with a yes/no button on the computer keyboard. The task was executed using E-prime software (version 2.1), which also recorded the RTs and correct responses.

The paradigm was designed in an event-related (ER) fashion. The ER design presented each stimulus at a specific time, allowing the investigation of the evoked hemodynamic response delayed by 2–3 s from stimulus-induced neuronal activity (47). Stimuli from different categories (e.g., experimental conditions; word vs. pseudo-word.) were presented in a random intermixed order for 4–11 s (Figure 1).

Anatomical MR images were obtained on a 3T Siemens Trio MRI (Siemens Medical Solutions, Erlangen, Germany) using a 32-channel receive-only head coil at the Unité de Neuroimagerie Fonctionnelle of the IUGM. A volumetric Magnetization Prepared Rapid Gradient Echo (MPRAGE) sequence was used to acquire a high-resolution T1-weighted 3D anatomical image, using the following parameters: TR = 2.3 s, TE = 2.91 ms, TI = 900 ms, flip angle = 9°, FOV = 240 × 256, voxel size = 1 × 1 × 1 mm³. This sequence was followed by an ASL sequence at rest, without sensory deprivation. The imaging sequence was a PICORE labeling geometry (20) and Q2TIPS tag duration control (48) to quantify the baseline CBF (CBF₀; (49). A post-label delay of 900 ms and label duration of 1500 ms were used, with repetition time (TR) and echo time (TE) of 3 s and 20 ms, respectively. The ASL signal is evoked by the local magnetization differences following the diffusion of the magnetically labeled blood to quantitatively measure blood perfusion. A single M₀ scan was also acquired to compute the blood perfusion parameters. This acquisition was done with the same parameters as the ASL sequence except for the TR, which was set to be very long (10 s) to yield a measurement of the fully relaxed magnetization. The whole acquisition including the MPRAGE took approximately 20 min.

A stereotaxic system (Brainsight, Rogue Research Inc.) was used to align anatomical images from each individual’s MRI and the patch holding the optical fibers. The registered positions of each optode were then mapped into normalized brain coordinates from Montreal Neurological Institute (MNI) template for group analysis.

Z-scores for each cognitive test were calculated from the normative reference data available for different age groups. Only participants whose results were above the normal guideline were kept for further analysis. The RTs and the accuracy of responses to the lexical-semantic task were analyzed using SPSS (IBM, New York, USA). A two-way repeated measures analysis of variance (ANOVA) was applied to RTs for correct answers as a dependent variable with the factors of age (young, elderly) and condition (word, pseudo-word).

Both signal processing (heart rate regression, intensity to concentration conversion, normalization, and smoothing) and statistical analysis (general linear model, GLM) were performed using an SPM8-compatible toolbox made in-house (50) based on NIRS-SPM v3.2; NIRS10.

Changes in optical density, ΔOD, were computed from emitted and received photon fluence Φ: ΔODt,λ=−lnΦt,λΦ0t,λ A heart rate analysis was done to eliminate channels without physiological signals. A coregistration of the source-detector positions on the MNI’s MRI template was done to ensure coherent optode positioning for group analysis. Optical signals were then transformed into hemoglobin concentrations C_HbO2 and C_HbR, applying the modified Beer–Lambert law with [ CHbO2(t)CHbR(t) ]=[ εHbO2λ1εHbO2λ2εHbRλ1εHbRλ2 ]1[ ΔOD(t,λ1)/(d.ℓDPF(λ1))ΔOD(t,λ2)/d.ℓDPF(λ2)) ]

where λ_DPF is the differential path-length factor accounting for the random photon trajectory between each source and detector (32).

Hemoglobin concentration changes were filtered with a Gaussian kernel (1.5 s FWHM) and high pass filtered by a second-order Butterworth filter with a cutoff frequency of 0.01 Hz. The significance of each effect of interest (abstract, concrete, and pseudo-word) was determined using the theory of Gaussian fields (47). A GLM was fit using a canonical hemodynamic response function (HRF). Contrasts over sessions (intrasubject) were analyzed using a fixed effects model, while testing for contrasts in the intersubject analysis was done by estimating the ratio of the random effects variance to the fixed effects variance. An expected Euler correction based on Lipschitz–Killing curvatures was applied to the threshold on the HbR/HbO₂ t-statistic images to account for the spatial correlation. The GLM method was based on the precoloring method of NIRS-SPM toolbox (51) for noise treatment.

To determine the value of the background absorption and scattering coefficients of the brain, a reflectance curve was fit for each source-detector pair of the time-domain system (34). The curve-fitting procedure was done by a non-linear optimization MATLAB function (Isqcurvefit) with fit parameters of absorption and reduced scattering coefficient (μ_a and μ_s, respectively) and amplitude to the theoretical temporal point spread function (TPSF). We applied the appropriate analytical model to fit the reflectance curve. This model was validated by applying a Monte Carlo simulation and with a priori information about the thickness of the first layer, including skin, skull, and CSF, obtained from the segmented anatomical MR images (using SPM8). With a high-resolution T1-weighted anatomical image, the maximum errors on the hemoglobin concentrations were expected to be no more than 15% (34).

To determine hemoglobin concentrations from optical parameters, we assumed that oxy- and deoxyhemoglobin and water were the dominant absorbers between the 690 and 850 nm wavelengths. The linear system describing the relationship between the extinction coefficient ε(λ) (taken from the literature) and the absorption coefficient μ_a(λ) (calculated from TRS measures) is given by μaλ=2.303⋅ελ⋅Cμaλ1μaλ2μaλ3μaλ4=εHbO2λ1εHbRλ1εH2Oλ1εHbO2λ2εHbRλ2εH2Oλ2εHbO2λ3εHbRλ3εH2Oλ3εHbO2λ4εHbRλ4εH2Oλ4 CHbO2CHbRCH2O We also assumed that biological tissues contained 70% water, reducing the above system to four equations with two unknowns. A pseudo-inversion of the equation with a least-square fit provided C, the hemoglobin concentration.

Coregistration: The anatomical images served two purposes in this study. First, at the individual level, we normalized the anatomical images to the MNI space including the subject’s fiducial coordinates. Then, to achieve better spatial resolution for fNIRS analysis, we projected the optodes’ positioning coordinates, collected from the Brainsight 3D camera, on the cortex (Figure 3). For group analysis, we needed to transform all images to the MNI template. These steps were performed using an in-house version of the algorithm of NIRS_SPM

The absolute CBF was measured in arbitrary units by the constant component of an ASL scan (ASL₀) using the standard quantitative approach described by Wong et al. (52). This value was then converted to physiologically relevant units (mL blood/100 g tissue per minute) using a general kinetic model fit to the ASL signal (20, 49). This model defines a relationship between the measured signal and CBF₀ assuming blood longitudinal relaxation time TI and fully relaxed magnetization M₀ are known CBF0=ASL06×106 ⋅ 2 ⋅ TI1 ⋅ exp (−T12T1b)⋅ M0b

where TI1 = 1400 ms, TI2 = 2000 ms, and T1b = 1932 ms at 3 T and the factor 6 × 10⁶ converts the units for CBF₀ to mL/min/100 g. The value for M_0b, the fully relaxed blood magnetization, was calibrated using that of the white matter measured in the M₀ scan (52): M0b=M0WM⋅1λ⋅expTE⋅1T2WM−1T2b with M_0WM the average value of the M₀ scan in a region of interest (ROI) selected from the segmented white matter, the brain-blood partition coefficient for water λ = 0.9 mL/g (53), TE = 12 ms, T_2WM = 70 ms, and T_2b = 275 ms at 3 T. The blood flow measurement was then regressed against the functional NIRS data to evaluate its impact.

There was no difference (p > 0.05) between the two age groups’ mean years of education (older = 16 ± 2.33 and younger = 16.95 ± 1.78). Both age groups were also matched for sex and consisted of 15 women and 8 men. They were compared on the neuropsychological measures of memory, vocabulary, and executive functions described earlier. The older adults performed worse on a number of subtests evaluating executive functions, but consistently with our hypothesis, the results of the vocabulary test showed no significant difference between the two groups (Table 1). Since responding correctly to the lexical-semantic decision task does not require planning or strategic changing skills, we expected these differences to have no impact on results.

Both groups performed equally accurately across conditions except on pseudo-words derived from concrete words [F(1, 55) = 6.1, p = 0.017]. Young participants were faster in all conditions except for concrete words. We applied a two-standard-deviation cutoff on the RTs of correct responses. RTs for correct trials are presented in Figure 4. Results from two 2-way ANOVAs showed no age × lexicality [F(1, 176) = 0.621, p > 0.05] or age × condition [F(1, 176) = 0.275, p > 0.05] interactions. There was a simple effect of lexicality [F(1, 179) = 11.4, p = 0.001], irrespective of age.

We calculated the absolute oxy- and deoxyhemoglobin concentrations ([HbO₂] and [HbR], respectively) as well as oxygen saturation (SatO₂) in both right and left prefrontal lobes covered by our four-channel TRS patch. Considerable intersubject variability was observed in the measures of hemoglobin concentrations, as previously reported in the literature (34). Group mean comparisons were inspected for the homogeneity of variance assumption. When the variance test of homogeneity was significant, the p-value of the unequal variance for one-tailed t-tests was reported. TRS measures of right and left frontal areas were analyzed separately. Older adults showed a resting [HbO₂] and [HbT] (HbT = HbO₂ + HbR) decrease in the left hemisphere compared to the right and an overall reduced [HbO₂] compared to young adults (Table 2). A one-tailed comparison revealed a decrease in SatO₂ in both left and right prefrontal lobes for the elderly group (p = 0.045 and 0.029, respectively). Because we observed a different trend in hemispheric changes in the measured hemoglobin concentrations across age, we applied a two-way ANOVA to investigate the effect of the baseline physiology of aging and laterality. The results showed no significant interaction between age and laterality.

Analysis of tagged ASL images revealed different group averages over the segmented white and gray matter anatomical images applied as explicit masks. Mean baseline CBF calibrated with the use of individual voxel values of the M₀ sequence was computed for both age groups and a significant difference was observed for global (p = 0.001) and gray matter (p = 0.02) blood perfusion (Figure 5). To investigate regional effects, we calculated mean CBF within regions of interest (ROIs) defined by the optical helmet. Temporal and frontal ROIs were created by applying a 35 mm diameter disk around optical channels covering the temporal and frontal regions. At rest, older adults had significantly lower blood flow (p = 0.01) in both temporal and frontal lobes than their younger counterparts.

Stimulus-dependent activation of all 58 channels was measured within the areas covered by the optodes. Significant changes were defined at p-values of less than 0.05. The activation map was then obtained via interpolation of the beta values calculated from the GLM model over the localized optodes coregistered on the MNI template, although a lack of spatial resolution made it difficult to identify activation areas with any precision.

The regions of activation for group average, obtained from the intersection of individuals’ activation maps, were impacted by variability induced in the positioning of optodes. Thus, a smaller group-level significant activation map was obtained, although the pattern of activation followed the same language network areas at the individual level. Brain regions activated by the lexical-semantic task were partially different for word and pseudo-word stimuli. In response to semantic word processing, a decrease in [HbR] concentration was observed in the elderly adults group in the left frontal region, at the intersection of the right inferior frontal gyrus (IFG) and the superior temporal gyrus (STG). In young adults, pseudo-word stimuli generated diminished [HbR] and increased [HbO₂] in the left IFG but an inverse response (decreased [HbO₂] and increased [HbR]) in the left inferior temporal (IT) and frontal lobes. In contrast to elderly participants, activated areas remained in the temporal sulci for other types of stimuli (concrete and abstract) but again in an inverse fashion.

An ANOVA for the main effect of age depicted significant [HbO₂] and [HbR] changes in both right and left hemispheres when older and younger adults were compared. Activation differences were mainly found in the bilateral dorsolateral prefrontal cortex (DLPFC) and IFG, and right posterior middle temporal and occipitotemporal gyri. Younger participants showed increased [HbR] and decreased [HbO₂] in the DLPFC and IFG in response to the semantic processing task (so-called inverse response). Conversely, pseudo-words led to a significant right ventral anterior premotor cortex decrease [HbR], which could be interpreted as an effort to analyze the stimulus by covert reading.

A correlation coefficient was calculated to examine the associations between age, performance and baseline physiology (ASL and TRS measures). Here, we sought relationships between RT and physiological measures while controlling for the effects of age and performance. The partial correlation coefficient between RT and [HbO₂], [HbR], SatO₂, and CBF was calculated by adjusting for age and performance scores. We presumed that both variables of correlation were linearly related to age and performance on the neuropsychological tests.

Inspection of the correlations between RT and CBF measures revealed no significant relationship. Pearson’s coefficient of correlation between RT and [HbO₂], [HbR] and SatO₂ dropped significantly once we removed the effect of age (from r (42) = 0.35 to r (40) = 0.06). It is interesting to note that, even though there were no significant correlations between the variables under investigation, the variations tended to move in opposite directions when younger and older adults were compared (see Figure 8). The tendency graphs revealed that slower responding elderly participants had slightly elevated baseline [HbO₂], [HbR], and SatO₂, whereas younger individuals showed the opposite pattern (i.e., slower respondents had lower baseline levels).