The PICO (population, intervention, comparison, and outcome) model was used to outline the inclusion and exclusion criteria for this systematic review (Eriksen and Frandsen, 2018). The population being studied included healthy younger, middle-aged, and older adults (i.e., participants aged 18 years and older). As this review focuses on healthy adults, studies that examined participants who are obese or patient populations including those with chronic neurological disorders were excluded. Intervention type included cognitive, physical, or a combination of both while fNIRS was used to assess changes in cerebral oxygenation during task performance. The studies must have also compared a high- and low-fit group or an active and control group that were created based on direct or estimated measures of CRF. Cross-sectional and longitudinal studies were included whereas those without full-texts, systematic reviews, and animal studies were excluded.

Following PRISMA guidelines, the systematic literature search was conducted in July 2021 using four online databases: Medline, Embase, SPORTDiscus, and Web of Science. There were no restrictions set for language or publication year. The search terms encompassed Medical Subject Headings (MeSH) and keywords that described fNIRS, the hemodynamic response as well as CRF levels (Supplementary Table 1). For example, the following keywords and MeSH terms were used: “Spectroscopy, Near-Infrared” AND “Cardiorespiratory fitness” OR “V̇O_2max” AND “[HbO₂]” OR “[HHb].”

The resulting studies were imported and managed in Covidence (Melbourne, Australia) where duplicates were automatically removed and visually inspected by TS. Titles and abstracts were then independently screened by OD and TS. SF resolved all conflicts and relevant studies were retained for the full-text review. SF, TS, and OD reviewed the full-texts and disagreements were resolved by a consensus between authors. The reference lists of the included full-texts were hand-searched as well as recent publications since the initial search for additional studies meeting the inclusion and exclusion criteria.

The Joanna Briggs Institute (JBI) checklist for analytical cross-sectional studies was used to assess the methodological quality of the included studies (Ma et al., 2020). It is suitable for non-randomized experimental studies because it evaluates eight criteria: (1) Inclusion criteria in the sample were clearly defined; (2) The study subjects and the setting were described in detail; (3) The exposure was measured in a valid and reliable way; (4) Objective, standard criteria were used for measurement of the condition; (5) Confounding factors were identified; (6) Strategies to deal with confounding factors were stated; (7) The outcomes were measured in a valid and reliable way; and (8) An appropriate statistical analysis was used. The overall appraisal to determine whether a study is of sufficient quality to include, exclude, or seek further information, was decided based on a consensus between TS and SF.

Data extraction was conducted by TS and SF. The form included information on study and participant characteristics, study design, CRF measures, and fNIRS measures. Significant differences between CRF scores were identified across both direct and estimated measures. The hemodynamic response was extracted based on the study’s primary outcome measure. In addition, the type of task during which the hemodynamic response was measured (e.g., cycling, Stroop task) was analyzed while accounting for variables such as age, sex, and exercise intensity.

The initial database search resulted in 2,592 studies and 1,495 after duplicates were removed. After screening the titles and abstracts, 1,450 records were excluded. The full-texts of 45 studies were then reviewed and 31 studies were excluded due to the wrong study design (n = 18), wrong comparator (n = 5), wrong outcomes (n = 4), abstract only (n = 2), wrong population (n = 1), and no full-text available (n = 1) (Figure 1). No additional papers were included following a hand-search. Fourteen studies were retained for analysis. The publication period ranged from 2010 to 2020 and all the studies were cross-sectional except for one which was a pre-/post-intervention design. Four studies took place in France, followed by three in Japan, two in Australia and Canada, and one in Belgium, Germany, and the United States, respectively. Study characteristics for all 14 included studies are shown in Table 1.

Seven studies examined the effects of CRF during an exercise task, five studies evaluated the effects during a cognitive task, one study examined a cognitive-motor dual-task, and one study examined a visual task. CRF was assessed using direct and estimated measures. Ten studies used a cycle ergometer test, two studies used a treadmill test, one study used self-report questionnaires, and one study did not specify which test was used but provided a V̇O_2max cut-off score for each group (Shibuya and Kuboyama, 2010). In addition, the hemodynamic response was measured using both [HbO₂] and [HHb] in 10 studies whereas two studies exclusively examined [HbO₂], and two studies only examined [HHb]. These measurements were taken in the PFC across 11 studies, motor cortices in two studies, and visual cortex in one study.

A total of 530 participants were examined and sample sizes ranged from 11 to 66. Seven studies consisted exclusively of male participants whereas three studies only enrolled females, and four studies evaluated both males and females. The age groups being examined varied across studies including five that evaluated younger adults (18–26 years old), four examined older adults (≥60 years old), three studies compared older and younger adults, and two studies evaluated middle-aged adults (40–60 years old). Amongst the studies that compared older and younger adults, high and low levels of CRF were only established in the older adult group in two studies (Fabiani et al., 2014; Agbangla et al., 2019a). Participant characteristics for included studies are indicated in Table 2.

In the included studies, various terms were used to describe CRF levels. More specifically, the studies evaluating CRF during exercise tasks used terms such as moderately active and sedentary (Brugniaux et al., 2014; Caen et al., 2019), short-term and long-term training (Buzza et al., 2016, 2020), trained and untrained (Oussaidene et al., 2015), endurance athlete and control (Seidel et al., 2019), and athlete and non-athlete (Shibuya and Kuboyama, 2010). In studies measuring cerebral oxygenation during cognitive tasks, the terms higher and lower fit were used to characterize CRF level.

Two studies examined the effects of CRF level on [HbO₂] activation during the Stroop task (Dupuy et al., 2015; Hyodo et al., 2016), one study used the n-back task (Agbangla et al., 2019a), control counting and random number generation (Albinet et al., 2014), and trail making tests part A and B (Mekari et al., 2019). Four studies measured [HbO₂] and [HHb] (Albinet et al., 2014; Dupuy et al., 2015; Hyodo et al., 2016; Agbangla et al., 2019a) whereas one study only measured [HbO₂] (Mekari et al., 2019). All studies measured PFC activation in older adults and divided the participants into high and low fit groups based on V̇O_2max measured by cycle ergometer tests except for one study that used a self-report questionnaire in older adults (Agbangla et al., 2019a).

FNIRS recordings during exercise tasks lasted between 2 and 32 min, which was substantially longer than studies using cognitive tasks where brain activity was recorded for 30–150 s. Physiological and/or motion filters were used to remove artifacts from noisy data in all studies evaluating cerebral oxygenation during cognitive and visual tasks. In contrast, the study evaluating dual-tasks stated that they did not use any filters but artifacts were identified through visual inspection and replaced by interpolation of adjacent data (Goenarjo et al., 2020a). Of the studies examining exercise tasks, filtering methods were not specified except for one study that used short-distance channels to eliminate physiological artifacts (Seidel et al., 2019).

All but one study used continuous wave fNIRS, which are the most commercially available devices (Scholkmann et al., 2014). Fabiani et al. (2014) used a frequency domain device. Accordingly, thirteen studies measured relative changes in cerebral oxygenation by subtracting baseline values from task-evoked changes. A linear regression approach was used by one study to assign a slope coefficient to [HbO₂] and [HHb] for the entire response signal (Albinet et al., 2014). Channel configuration varied between studies with four using three channels (Buzza et al., 2016, 2020; Mekari et al., 2019; Goenarjo et al., 2020b), three studies using two (Shibuya and Kuboyama, 2010; Albinet et al., 2014; Caen et al., 2019) or eight channels (Brugniaux et al., 2014; Oussaidene et al., 2015; Agbangla et al., 2019a), and one study using 16 (Dupuy et al., 2015), 22 (Seidel et al., 2019), 32 (Fabiani et al., 2014), or 48 (Hyodo et al., 2016) channels, respectively. Across all studies, the lower wavelength ranged from 690 to 794 nm and the upper wavelength ranged from 830 to 905 nm, which was largely dependent on the fNIRS device.

CRF was measured using direct measures of V̇O_2max, V̇O_2peak, ventilatory threshold, or peak power output in 11 studies and self-report questionnaires in three studies. The study using a ventilatory threshold justified using this measure because it was more convenient for older adults (Hyodo et al., 2016). Amongst the studies using direct measures, four studies divided the participants into high and low fit groups based on a median split or by excluding the middle group of CRF scores and analyzing the upper and lower thirds (Albinet et al., 2014; Fabiani et al., 2014; Mekari et al., 2019; Goenarjo et al., 2020a). The remaining studies divided groups based on published age and gender norms of CRF. In addition to direct measures of CRF, five studies assessed whether participants had a background in different types of physical activity or training during the recruitment stage (Albinet et al., 2014; Brugniaux et al., 2014; Oussaidene et al., 2015; Caen et al., 2019; Seidel et al., 2019). This included a 7-point physical activity rating used to estimate V̇O_2max (Albinet et al., 2014), a self-report physical activity questionnaire (Brugniaux et al., 2014), a simple question about past involvement in recreational sports (Caen et al., 2019), or questions that determined the number of hours per week of physical activity (Oussaidene et al., 2015; Seidel et al., 2019).

The three studies using self-report questionnaires to determine CRF consisted of the NASA/Johnson Space Center physical activity questionnaire, which required older adults to rate their physical activity on a scale from 0 to 7 and was adjusted for their age, body mass index (BMI), and sex (Agbangla et al., 2019a). In comparison, two studies used self-reported physical activity logs to track training minutes of moderate to vigorous exercise per week (Buzza et al., 2016, 2020).

Based on the JBI quality assessment checklist, each of the 14 studies was of sufficient quality to be included in this review (Table 3). In all studies, the exposure was clearly described, the outcomes were clearly defined, and statistical analyses were appropriately chosen. All studies described the participant characteristics in sufficient detail, but two studies did not clearly describe the setting. Despite this, it can be assumed that these studies were conducted in a university lab setting due to the specialized equipment involved in CRF testing. Objective criteria were used to measure CRF in 13 studies whereas one study indicated the cut-offs between high and low groups but did not indicate what test was used to measure V̇O_2max in each group (Shibuya and Kuboyama, 2010). Lastly, seven studies identified and controlled for confounding variables such as education level. Five studies did not control for confounding variables, and two studies were unclear. These studies were not excluded because they reported confounding characteristics in a table format, which were not significantly different between groups. Therefore, it is unlikely that they contributed to sources of bias between studies.

Previous reviews have outlined the effects of chronic exercise on cognitive performance (Li et al., 2017; Rathore and Lom, 2017), but few have assessed the impact of CRF on cerebral oxygenation. In a meta-analysis, Rooks et al. (2010) examined the impact of training status on cerebral oxygenation during incremental tests without considering how CRF and training status were measured (i.e., V̇O_2max or physical activity level). More recently, Agbangla et al. (2021) reviewed the impact of CRF on cerebral oxygenation during cognitive tasks. This review examined the effects of CRF level, as determined by direct or estimated measures, on cerebral oxygenation during exercise and cognitive task performance. Exercise-related effects on cognition have been explored using the cardiorespiratory hypothesis, which maintains that improvements in cognitive performance are moderated by factors such as increased cerebral perfusion in individuals with greater CRF (Agbangla et al., 2019b).

Studies examining the PFC revealed increased [HbO₂] in higher compared to lower fit groups during cognitive tasks. In fact, most studies in this review used tasks that draw on executive functions since they are susceptible to age-related declines and can be improved with exercise interventions (Colcombe and Kramer, 2003). There is evidence that the frontal lobe is disproportionately affected by aging but is activated during executive function tasks (West, 1996). In older adults, this results in bilateral PFC activation compared to younger adults who demonstrate unilateral activation to support task performance (Reuter-Lorenz and Cappell, 2008). This is outlined in the Compensation-Related Utilization of Neural Circuits Hypothesis (CRUNCH) which accounts for the effect of task demands and processing capacity that when exceeded, results in performance declines (Reuter-Lorenz and Cappell, 2008). Therefore, PFC activation is more pronounced with increasing executive function demands and greater CRF can facilitate these processes (Albinet et al., 2014; Agbangla et al., 2019a; Goenarjo et al., 2020a). In the case of M1, activation is observed in the contralateral hemisphere to the movement being performed (Shibuya and Kuboyama, 2010). Previous studies have reported that right-handed individuals activate the left M1 but recruit M1 bilaterally during exhaustive exercise to compensate for decreased muscle force (Shibuya et al., 2008). Therefore, both the PFC and M1 can demonstrate bilateral activation as a mechanism of compensation.

In studies comparing younger and older adults, higher V̇O_2max only contributed to increased cerebral oxygenation in older adults (Dupuy et al., 2015; Agbangla et al., 2019a). In other words, high and low fit younger adults (Dupuy et al., 2015) as well as a pooled group (Agbangla et al., 2019a) did not exhibit increased cerebral oxygenation or differences in performance. These findings have been attributed to the high functioning status of young adults and intact brain structures that help maintain cognitive performance regardless of CRF level. In addition, older adults aged 65 and over have shown the greatest capacity to demonstrate improvements in executive function compared to younger adults (Colcombe and Kramer, 2003). Therefore, differences in CRF may only lead to small improvements in cerebral oxygenation, which may be less evident in younger compared to older adults.

[HHb] findings prove to be more variable with studies reporting both decreased and insignificant changes based on CRF. Two studies found decreased [HHb] during executive function and dual-tasks in high compared to low fit younger and older adults (Agbangla et al., 2019a; Goenarjo et al., 2020b). In contrast, one study indicated that the lack of significant findings was due to their fNIRS device, which was not configured to measure [HHb] (Hyodo et al., 2016). Insignificant differences in [HHb] can also be attributed to its low signal to noise ratio making it hard to identify the hemodynamic component of the signal. Nonetheless, [HHb] is less likely to be influenced by systemic and motion artifacts making it a reliable measure during tasks that require unrestricted movements (Menant et al., 2020). To better interpret these results, it is important to remember that during cognitive stimulation, the fNIRS responses are manifested by an increase in [HbO₂] and a decrease in [HHb] (Villringer, 1997). [HHb] is also strongly correlated with brain activity and inversely correlated with the BOLD signal (Villringer, 1997). The greater the decrease in [HHb], the more the BOLD signal increases. Based on this information, a greater decrease in [HHb] during a cognitive task in high fit subjects could represent greater brain activity as shown by Colcombe et al. (2004).

Studies examining mixed samples of male and female participants demonstrated that increased CRF was associated with increased cerebral oxygenation and better performance (Agbangla et al., 2019a; Mekari et al., 2019). There are, however, known differences in CRF between males and females that contribute to changes in cerebral oxygenation (Colcombe and Kramer, 2003; Erickson et al., 2007; Dimech et al., 2019). For example, findings differed across male and female groups in the present review such that high fit females demonstrated increased bilateral activation in the PFC (Albinet et al., 2014) and increased right activation as task demands increased (Dupuy et al., 2015). Older males, however, only demonstrated left lateralized PFC activation in the higher fit group, which is a pattern typically observed in younger adults (Hyodo et al., 2016). This is outlined in the hemispheric asymmetry reduction in older adults (HAROLD) model whereby bilateral activation is expected due to decreased white matter integrity, vascularization, and less efficient adaptations to task-related metabolic demands (Cabeza, 2002; Hyodo et al., 2016; Agbangla et al., 2019b). Upon re-examining the studies investigating mixed samples, participants were predominantly female. Therefore, data from female participants may be contributing to bilateral activation in mixed samples more so than males. More evidence is needed that directly compares activation in males and females across different CRF levels.

Incremental and maximal exercise elicit an increased metabolic demand for oxygen. In the present review, cerebral oxygenation was measured during acute cycling exercise and in one case during a static handgrip task (Shibuya and Kuboyama, 2010). During submaximal exercise, [HbO₂] was greater in the PFC of trained compared to untrained young males (Oussaidene et al., 2015). In addition, moderately active young males demonstrated increased PFC oxygenation until 80% V̇O_2max compared to the sedentary group, which displayed constant but lower cerebral oxygenation (Brugniaux et al., 2014). These findings build on a previous meta-analysis where healthy participants demonstrated steady increases in cerebral oxygenation during incremental exercise compared to their baseline resting levels (Rooks et al., 2010).

In terms of [HHb], two studies measured middle-aged females and males, respectively (Buzza et al., 2016, 2020). Only the male long-term training group (higher fit) demonstrated greater PFC [HHb] compared to the short-term training group (lower fit) at 90% intensity on a cycle ergometer (Buzza et al., 2020). A similar interaction was observed in younger adults whereby left PFC [HHb] increased in a trained vs. untrained group (Oussaidene et al., 2015). From low to submaximal intensities, PFC [HbO₂] is expected to increase until it reaches a plateau near maximum intensity (Jung et al., 2015). During the handgrip exercise test, however, activation in M1, which is responsible for sending efferent information to the contracting hand muscles, continued to rise in the non-athlete group at voluntary exhaustion whereas it dropped below baseline values in the athlete group (Shibuya and Kuboyama, 2010). More specifically, participants were right-handed and cerebral oxygenation was measured in the contralateral M1. Continued activation may be a mechanism used to compensate for decreased muscle force when exhausted (Shibuya et al., 2008). A second mechanism is the bilateral activation of M1, which has been observed during low intensity static handgrip tasks (Shibuya et al., 2008). For the same task, compensation and greater oxygenation in the lower compared to higher fit group reflects how [HbO₂] and [HHb] are regulated in M1 during exhaustive motor tasks (Shibuya and Kuboyama, 2010).

Numerous studies have demonstrated that aerobic training is beneficial for cognitive performance, but few interventional studies have evaluated the effect of aerobic training on brain oxygenation during physical exercise or cognitive tasks. Only one study in this review measured a 6-week aerobic exercise training intervention and demonstrated that [HbO₂] and [HHb] amplitude and slope increased post-intervention in younger adults during maximal incremental exercise test (Caen et al., 2019). It is unclear if this is the case for older adults, but previous reports have demonstrated improvements in executive functions following exercise interventions (Voss et al., 2012; Northey et al., 2018; Domingos et al., 2021). In addition, fitness training selectively improved executive control processes compared to speed, controlled, and spatial ability in older adults (Colcombe and Kramer, 2003). Fewer studies have examined cerebral oxygenation following exercise interventions but changes in cerebral blood flow have been observed in cortical and subcortical regions (Brown et al., 2010; Chapman, 2013). To the best of our knowledge, only Coetsee and Terblanche (2017) reported lower cerebral oxygenation in older adults after aerobic training, but the measure of V̇O_2max was not explicitly stated. In addition, the meta-analysis by Rooks et al. (2010) reported greater cerebral oxygenation during incremental exercise tests between trained and untrained participants but lacked defined criteria for training status making the results difficult to compare. Previous reports have used direct measures of CRF in older adults (Dupuy et al., 2015) while others suggest that estimated measures are more feasible to obtain in older adults (Agbangla et al., 2019a). Nonetheless, more training interventions are needed to assess the effects of CRF on cerebral oxygenation.

Exercise has known benefits on cerebrovascular health such that the cardiovascular system is involved in delivering oxygen and regulating cerebral metabolism to sustain cognitive processing. From a cognitive perspective, exercise may increase angiogenesis, neurovascular plasticity, and oxygen saturation in brain regions related to cognitive performance including prefrontal and motor cortices and the hippocampus (Stimpson et al., 2018). Exercise also upregulates growth factors like brain-derived neurotrophic factor (BDNF), insulin-like growth factor (IGF), and vascular endothelial growth factor (VEGF), which are involved in synaptic plasticity, neurogenesis, promote angiogenesis, and support memory (Cotman and Berchtold, 2002; Ploughman, 2008; Davenport et al., 2012; Hayes et al., 2013). These neurotrophins can also offset the effects of age-related cerebral atrophy that interfere with adequate oxygen delivery to the brain (Ainslie et al., 2008; Erickson et al., 2014). For example, 3 months of aerobic exercise have been reported to increase neurotrophins and cerebral blood volume (Pereira et al., 2007). Therefore, higher fit individuals demonstrate increased BDNF compared to lower fit individuals because of improved cerebral blood flow and better vascularization, which foster neurotrophic and growth factors in the brain (Brown et al., 2010; Stimpson et al., 2018). In addition, older adults who are more physically active have been found to display a higher number of small cerebral vessels than less physically active older adults (Bullitt et al., 2009). Angiogenesis is upregulated by VEGF which also promotes endothelial cell proliferation (Cotman and Berchtold, 2002). Similarly, exercise increases the production of plasmatic VEGF which is the main growth factor associated with capillary formation in the brain (Cotman and Berchtold, 2002; Duman, 2005). Stimpson et al. (2018) and Dupuy et al. (2019) summarized these factors in a simplified model that describes the relationship between physical activity and cognition. In combination with these molecular mechanisms, it has also been observed that higher fit individuals have higher cerebral blood flow at rest, and during the tilt test and exercise than lower fit people (Murrell et al., 2011a,b). This explains the reports of greater cerebral oxygenation in higher fit individuals. Given that higher CRF is associated with increased cerebral oxygenation, fNIRS measures can complement the existing literature on the mechanisms involved in exercise and improved cognition.

Given the growing popularity of fNIRS in cognitive and exercise physiology research, guidelines have been established to increase reporting transparency and the reproducibility of different study designs (Menant et al., 2020; Yücel et al., 2021). In terms of fNIRS devices, most commercially available continuous wave fNIRS systems have the capability of measuring [HbO₂] and [HHb], which is an advantage over fMRI. Therefore, studies should report on both measures to broaden our understanding of potential shifts between increases and decreases in these measures related to CRF. Amongst the studies in this review, the most common justification for only reporting [HbO₂] is that it better reflects cortical activation, but it should be noted that it contains a larger signal-to-noise ratio (Strangman et al., 2002). In terms of [HHb], it is less contaminated by systemic artifacts, but the signal is attenuated compared to [HbO₂], and certain devices are not configured to assess both measures (Kirilina et al., 2012; Hyodo et al., 2016).

As stated in the guidelines and consensus measures by Menant et al. (2020) and Yücel et al. (2021), signal processing methods should be clearly described. Many studies in this review were unclear or did not specify the methods used to process motion artifacts. For example, cycling tasks are expected to produce greater motion artifacts than seated cognitive tasks due to increased movement (Yücel et al., 2021). Removing irrelevant artifacts from the signal can be achieved through filtering, visual inspection, and is sometimes overcome with specific instructions for participants to minimize sudden head movements (Menant et al., 2020). In order to adequately compare and contrast different research studies and to fully understand cerebral oxygenation changes associated with CRF level, detailed reporting is imperative.

The duration of fNIRS measurements also differed between cognitive and exercise tasks. Exercise measures were significantly longer than cognitive tasks given the nature of incremental cycling. In both cases, the segment analyzed was of sufficient duration to capture the hemodynamic response. Longer measures, however, could increase participant discomfort and affect brain activation due to drifts in the signal (Menant et al., 2020).

This systematic review included cross-sectional studies with relatively small sample sizes. In addition, only one study examined the effects of exercise training on cerebral oxygenation. Future studies should consider implementing randomized controlled trials to ensure a comprehensive examination and minimal bias is introduced when examining the effects of CRF on cerebral oxygenation. The self-reported questionnaires to estimate V̇O_2max may have introduced biased responses if participants overestimated their physical activity levels. However, the questionnaires were reliable and a more feasible method to measure CRF in older adults. V̇O_2max also significantly differed between high and low fit groups in these studies.

Continuous wave fNIRS devices are most frequently reported in the literature, but only provide relative changes in cerebral oxygenation rather than absolute values. This limitation can be overcome by frequency and time domain fNIRS devices, which come at a greater cost, but allow for more in-depth measures. Although it is also limited to surface cortex measurements, changes in PFC and motor cortex activation were identified between studies. The interaction between these two regions should be further examined as connectivity between the PFC and motor regions is important for exercise load management, coordination, and preparation of motor movements, which may differ between high and low fit groups (Voss et al., 2016). In addition, similar deoxygenation between PFC and motor cortices has been reported during maximum intensity exercise (Subudhi et al., 2009). Lastly, preprocessing is essential for removing physiological (e.g., systemic) and motion artifacts (e.g., head movements) in the fNIRS signal. Due to the nature of physical exercise, these variables may be especially prominent. Future studies should report all preprocessing steps to increase signal quality and methodological reproducibility.