Acute particulate matter exposure diminishes executive cognitive functioning after four hours regardless of inhalation pathway

This project received ethical approval from the University of Birmingham Science, Technology, Engineering and Mathematics (STEM) ethics committee, number ERN_21-1188. Written informed consent was obtained from each participant, and we confirm that the research complied with all relevant ethical regulations.

Briefly, the methodology of the study was a single-blind within-participants design. Participants took part in four cognitive tasks (total time 60 min) prior to, and 4 h following, a 1-h exposure session. Air quality was manipulated using candle burning as previously described in Shehab and Pope (2019)¹⁸. This has been shown to be a low-cost, controlled method to increase PM_2.5 concentrations to levels that might be experienced in an urban area⁷⁵. PM_2.5, PM₁₀, Carbon Dioxide (CO₂), and Carbon Monoxide (CO) were monitored throughout exposure to all air quality conditions. Inhalation pathway was manipulated by using a swimming nose clip, blocking the olfactory pathway, thereby decreasing the concentration of particulate pollution entering the body via this route. Participants took part in both PM pollution and clean air conditions whilst wearing the nose clip (allowing for ‘restricted’ oral-only inhalation) or without nose clip (’normal’ inhalation through both nose and mouth).

Participants

Thirty-nine staff and students at the University of Birmingham, Birmingham, UK were recruited through on-campus advertisements and offered cash on completion of each study visit. Individuals who reported current neurological, psychiatric, inflammatory, or respiratory disorders (e.g., multiple sclerosis, depression, rheumatoid arthritis, asthma), or current smoking (including e-cigarettes) were excluded. See Supplementary Methods for full study exclusion criteria. A full dataset was collected from 30 participants. All data was removed from four participants due to scores on the Depression, Anxiety, and Stress Scale (DASS) indicative of undisclosed or diagnosed mental health conditions; a pattern of random responses during cognitive tasks, suggesting potential interference or disruption (referred to as Sabotage); and self-reported non-compliance or adherence to study instructions.

The resultant data set for analysis contained 26 participants. Mean age  = 27.7 years, s.d.,  = 10.6, range 19–67; 57.7% female (reported gender and biological sex).

Power analyses were conducted based on unpublished data from previous experiments, using conservative estimates to ensure adequate sensitivity. For the Expression Recognition Task, using effect size and variance from prior data on approach bias before and after exposure to low-quality air (via candle burning) or ambient air (ΔApproach bias = 0.27, s.d., = 0.45), a sample size of 30 participants was conservatively estimated to provide 90% power to detect a significant change in socio-emotional processing at p  < 0.05. Similarly, for the Face Identification Task, based on previous data on cognitive load and attention after diesel exhaust exposure (ΔRT  = 22 ms, s.d.,  = 25 ms), 30 participants were estimated to provide 90% power to detect a 15 ms change in response time (p < 0.05). Although the final sample size was 26 participants, the conservative nature of the power analyses ensures that the study retains sufficient sensitivity despite the slightly smaller sample.

Design

This single-blind study used a repeated measures experimental design across 4 days. The within-subject factors were inhalation pathway, (normal inhalation, restricted inhalation); pollution exposure, (clean air, PM pollution); and session time (pre-exposure, post-exposure). The order of inhalation and exposure conditions were counterbalanced across the four sessions. Condition orders were randomly assigned to participant ID numbers (1–32) before data collection began; note that inhalation pathway condition was always the same for two subsequent exposures. Participants were subsequently assigned unique ID numbers sequentially based on successful recruitment. To address unequal distribution of conditions following participant attrition, a new participant ID was generated for the same condition. This aimed to control for potential psychological biases such as learning and fatigue effects in the experimental design. See Supplementary Table 1 for the condition order details of the 26 participants included in the data analysis.

Materials

A Windows 10 computer running Matrix Laboratory (MATLAB) version R2022a (9.4; MATLAB, 2022)⁷⁶ was used to run the cognitive tasks. All tasks were in the format of a MATLAB script utilising the Psychophysics Toolbox version 3.0.14⁷⁷. Task scripts and instructions are available on GitHub. We opted not to use assessments of general cognition, such as the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and Wechsler Adult Intelligence Scale (WAIS) as these were primarily derived to index severe damage or failure of cognition, as occurs in stroke or brain damage. As such they are insensitive to more subtle but nonetheless important cognitive degradation. TSI Optical Particle Sizer (OPS) 3330 was used to measure concentrations of PM_2.5 and PM₁₀ throughout participant exposure. LI-COR LI-820 CO₂ Gas Analyzer was used to measure concentrations of Carbon Dioxide. LASCAR EL-USB-CO Data Logger was used to measure concentrations of Carbon Monoxide. To provide PM (particulate matter) pollution, 100% stearin candles made from animal fat were burned. These unscented candles, measuring 190 mm in height and 22 mm in radius, were purchased from a major hypermarket. A swimming nose clip was used to reduce pollution concentration entering the olfactory pathway in the restricted inhalation condition.

Air quality monitoring

TSI OPS 3330: This instrument uses single particle counting technology to measure particles from 0.3 to 10 micrometres (μm) across 16 size channels. These channels were set to the default values of sample weighting factors defined by European Standard EN481 for PM_2.5 and PM₁₀. The instrument was set to record particle concentration and size distribution in 10 second intervals. The authors note there was no means of measuring the relative number of ultrafine particles. We use PM_2.5 mass concentration as our metric of PM pollution, following standard public health protocols.

LI-COR LI-820 CO₂ Gas Analyzer: This sensor is a non-dispersive infrared gas analyser measuring CO₂ concentrations. The instrument was set to record in 1 second intervals, with resultant values measured as particles per million (ppm).

LASCAR EL-USB-CO Data Logger: This standalone instrument was set to record in 10 second intervals, with CO values measured as particles per million (ppm).

Subjective measures

The Depression, Anxiety, and Stress Scale (DASS)⁷⁸ was used to identify recent participant depression, anxiety, and stress. Participants rated to what extent 42 statements applied to them over the past week on a scale of 0 to 4 (Did not apply at all—Applied to me very much, or most of the time). Higher scores indicate higher levels of psychological distress, with a maximum of 42 available for each metric (depression, anxiety, and stress).

Participants responded to a question about their awareness of the pollution exposure condition: “This morning, you spent 60 min in a room. Before entering, the room either had a burning candle or not. Please indicate which air condition you believe that you experienced today” responding to one of two options: Candle or No candle. Participants were also asked of their confidence in their answer: “How confident are you in your answer?” responding to one of the following five options: Not confident at all (0); Slightly confident (1); Somewhat confident (2); Fairly confident (3); or Completely confident (4).

Cognitive tasks

These tasks, originally developed for this study, have subsequently been detailed in a published protocol for a different experiment, HIPTox⁷⁹.

Spatial n-back task

The n-back task is a continuous performance assessment employed to evaluate working memory capacity⁸⁰. In this widely used cognitive task, participants are presented with sequences of stimuli, in this case, spatial locations, and must determine whether the currently displayed stimulus matches the one presented presented ‘n’ trials earlier. As ‘n’ increases, the volume of information retained in working memory also grows. It is broadly recognised that working memory has a limited capacity, with Miller (1956)⁸¹ proposing that the number of items that can be held is ~7 ± 2. Consequently, as ‘n’ increases, both the task’s difficulty and the proportion of errors tend to rise. Poor performance on the task suggests deficits in encoding, maintaining, and/or retrieving information.

The stimuli comprised a centrally positioned 3 × 3 grid with white grid lines (red-green-blue coordinate, RGB [255, 255, 255]) displayed on a grey background (RGB [128, 128, 128]). The grid measured 16.5 cm in both height and width, subtending a visual angle of ~15.78^∘ in both dimensions when viewed from a distance of 60 cm. The grid was situated centrally on the screen to ensure focused stimulus presentation.

During each trial, a blank 3 × 3 grid was displayed for 600 ms, followed by a single white square appearing in one of the nine possible grid locations for 1000 ms. Subsequently, the blank grid was shown once more. Participants were required to remember the sequence of grid positions and indicate for each trial whether the square occupied the same location as it had ‘n’ trials earlier, or different location. This was a two-alternative forced choice task, where participants used the ‘m’ and ‘z’ keyboard keys to respond ‘same’ and ‘different’ respectively; these keys were reversed for left-handed participants. Responses were expected within 10 seconds of the prompt. See Figure 4a for an illustration of the sequence of displays in each trial. The task consisted of four blocks of 45 trials, each containing eight matches (i.e., ‘same’ locations). The value of ‘n’ increased with each block: block 1, n = 1; block 4, n = 4.

a Spatial n-back Task; (b) Face Identification Task, which used face images sourced from Set A of the Karolinska Directed Emotional Faces dataset⁸⁶ (placeholder images are displayed in the figure rather than the actual stimuli); (c) Expression Recognition Task, which used face images sourced from the RADIATE database^91,92 (placeholder images are displayed in the figure rather than the actual stimuli); (d) Psychomotor Vigilance Task—Psychomotor speed trials; and (e) Psychomotor Vigilance Task—sustained attention trials. The screens shown are for illustrative purposes only; refer to the text for the actual stimuli sizes and on-screen locations.

Face identification task

This task evaluates executive function by assessing selective attention, akin to established cognitive assessments such as the Stroop⁸² and Flanker⁸³ tasks. Participants must disregard distractions and focus on the primary task objective.

Given the limited capacity of high-level cognitive systems, the brain employs two complementary mechanisms for attentional control. Proactive control is used to plan strategically and engage selectively with anticipated relevant information while avoiding foreseeable distractions. Simultaneously, reactive control is activated to adjust behaviour in response to unexpected events. For example, preparing coffee involves proactive control to locate and approach the kettle while ignoring the biscuit tin (particularly if on a diet). Conversely, reactive control is required to avoid a colleague who unexpectedly steps into your path. These mechanisms compete for attentional resources; successful task execution depends on sustained proactive control^84,85, while distraction caused by unforeseen, irrelevant events reflect reactive control.

The stimuli included a white (RGB [255, 255, 255]) spatial cue arrow pointing left or right and a centrally presented white fixation cross, all displayed on a black background (RGB [0, 0, 0]). Faces were sourced from the A set of the Karolinska Directed Emotional Faces⁸⁶, encompassing all available emotional expressions: fearful, angry, disgusted, happy, neutral, and surprised. Scrambled images were generated by dividing face images into 13,984 squares and randomising their positions. Each image subtended 8.6^∘ × 11.0^∘, with two images displayed simultaneously, each laterally offset by approximately  ± 8.1^∘ of visual angle for symmetrical presentation.

Participants were instructed to respond as quickly as possible to identify the gender presentation of the target face using either the ‘A’ and ‘Z’ keyboard keys (for left-handed participants) or the ‘L’ and ‘P’ keys (for right-handed participants), using the index and middle fingers of their dominant hand. Key assignment to ‘male’ and ‘female’ was counterbalanced across participants. The task consisted of four blocks of 60 trials each, totalling 240 trials.

Each trial began with a fixation cross presented for 500 milliseconds (ms), followed by a spatial cue for 400 ms that reliably indicated the location of the upcoming target. This was followed by another fixation cross (350–850 ms, determined by a random integer between 20 and 50 sampled at a 17 Hz frame rate). The face array then appeared for 75 ms, followed by a final fixation cross presented for 1500 ms or until participants responded. The face array comprised a central fixation cross, a distractor image (either scrambled or a face), and a target image (face). Gender and emotional expression congruency were balanced across trials, ensuring all stimulus combinations were equally likely. Participants were instructed to identify the gender presentation of the target face as quickly and accurately as possible following the short (75 ms) presentation of the face array. See Fig. 4b for the sequence of displays in each trial.

In this task, participants were required to identify the gender presentation of a target face while either another face (in two-face trials) or a non-face (in one-face trials) distractor was present. The presence of a face distractor typically increases response time⁸⁷, reflecting reactive selective attention capture. Studies on cognitive control consistently show that response slowing caused by distractors is more pronounced when the preceding trial lacked a compelling distractor (one-face trials) compared to trials with distractors (two-face trials)⁸⁸. This phenomenon is attributed to prior distractor suppression (in trial n-1), which enhances proactive processing of the target in the current trial n. In contrast, the absence of prior distractor suppression weakens proactive control, increasing vulnerability to distractor capture⁸⁹. Differences in response times (RTs) between two-face trials following two-face trials (repeat sequences) and those following one-face trials (change sequences) inversely reflect proactive control (ΔRT). Thus, greater response times in change sequences compared to repeat sequences indicate reduced proactive control.

Expression recognition task

This Go/No-go task utilised happy and fearful facial expressions as target stimuli to assess decision-making performance between positive-affective and negative-affective expressions. The task examined both approach bias and the ability to discriminate between emotional expressions.

Approach bias

This aspect of the task explores the tendency to engage with positive-affective stimuli (e.g., happy faces) more readily than with negative-affective stimuli (e.g., fearful faces). This preference is thought to reflect underlying emotion regulation processes⁹⁰. A natural inclination to approach positive stimuli and avoid negative stimuli is expected to produce a bias in accuracy, with better performance for happy targets and poorer performance for fearful targets.

Expression discrimination

This aspect constitutes the perceptual component of the task. Participant accuracy, regardless of the target expression, indicates the ability to distinguish among different emotion expressions. A higher overall accuracy demonstrates an improved capacity to promptly perceive and categorise happy and fearful facial expressions, which is a pivotal concept in theory of mind⁵⁶, a critical social-cognitive skill.

The expressive faces used in this task were sourced from the RADIATE database^91,92. Each face image was displayed centrally against a white background (RGB [255, 255, 255]), with explanatory text and fixation crosses presented in black (RGB [0, 0, 0]). Stimuli consisted of a centrally positioned face image measuring 8 cm in both height and width, subtending a visual angle of ~7.68^∘ in both dimensions when viewed from a distance of 60 cm.

Participants were instructed to respond to faces displaying the target expression (happy or fearful) and to withhold responses to non-target expressions. Participant responses were made as quickly as possible by pressing the keyboard spacebar with their dominant hand. The target expression alternated between blocks, beginning with ‘happy’. Each trial began with a fixation cross presented for 550–950 ms, followed by an expressive face image displayed for 100 ms, and then a blank screen lasting 700 ms or until a response was made. See Fig. 4c for the sequence of displays in each trial. The task consisted of six blocks of 44 trials each, with 28 target (Go) images and 16 non-target (No-go) images per block. All face stimuli featured either open mouths (more expressive) or closed mouths (less expressive), with these conditions evenly distributed between Go and No-go trials within each block.

Psychomotor vigilance task

This basic response time task assessed simple reaction time, providing a measure of global processing (psychomotor speed) and the ability to maintain concentration over extended periods (sustained attention)⁹³.

The target stimulus was a centrally presented red circle with a diameter of 0.5 cm (RGB [255, 60, 0]), displayed against a black background (RGB [0, 0, 0]). The circle subtended a visual angle of ~0.48^∘ when viewed from a distance of 60 cm. Explanatory text and fixation crosses were shown in white (RGB [255, 255, 255]).

Participants were instructed to focus on a centrally positioned fixation cross and respond as quickly as possible when the target, a small red circle, appeared. Each trial began with a fixation cross displayed for 400–800 ms in psychomotor speed trials, or for 25–35 s in sustained attention trials. This was followed by the appearance of the red circle target for 400 ms or until the participant responded. If the keyboard spacebar was pressed during the target presentation, a white circle was shown for 400 ms as feedback. See Fig. 4d and e for a visual representation of both trial sequence types. The task consisted of one block of 85 trials, including 10 sustained attention trials with long fixation intervals and 75 psychomotor speed trials with shorter fixation intervals.

Procedure

Upon arrival, participant suitability was checked against exclusion criteria and after informed consent participants completed the four cognitive tasks in the cognitive testing room: Spatial n-back Task; Face Identification Task; Expression Recognition Task; and Psychomotor Vigilance Task. Twenty minutes prior to participants completing the tasks, two candles were lit (PM pollution condition only) in an adjacent ‘exposure’ room (4.15  × 2.95  × 3.40 m  = 41.6 m³) and windows closed (all conditions). One minute prior to task completion, the candles were extinguished and the air quality sensors were activated to record Particulate Matter (PM), Carbon Dioxide (CO₂), and Carbon Monoxide (CO) concentrations. Participants, wearing a nose clip in the restricted inhalation condition, were taken into the exposure room and seated close to the air quality sensors. A fan circulated air in the room. Participants remained in the exposure room for 60 min, during which they completed the DASS questionnaire. After 60 min of exposure to either elevated PM concentrations (PM pollution condition) or room air (clean air condition), participants left the lab.

Four hours later, participants returned to the cognitive testing room. The cognitive tasks were repeated in the same order as previously described. Participants were then asked to indicate whether they believed they were exposed to the clean air or PM pollution condition earlier that day, and they received monetary compensation for their time. This procedure was repeated, after a minimum two-week washout period, until all four conditions were completed. See Fig. 5.

As described in text, candles were burned 20 min prior to the exposure (2) in the PM pollution condition.

Data processing

Spatial n-back task

Each trial has four possible outcomes⁹⁴: a hit (correctly identifying a ‘same’ location), a correct rejection (correctly identifying a ‘different’ location), a miss (failing to identify a ‘same’ location), or a false alarm (incorrectly responding ‘same’ to a ‘different’ location).

Working memory ability was quantified using \({{{{\rm{d}}}}}^{{\prime} }\), which represents the standardised difference between the signal-present (same location) and signal-absent (different location) distributions. It was calculated as the Z-score of the hit rate [#Hits / (#Hits + #Misses)] minus false alarm rate [#False Alarms / (#False Alarms + #Correct Rejections)]. This calculation was performed separately for each n-back value, with lower \({{{{\rm{d}}}}}^{{\prime} }\) scores indicating poorer working memory performance. As ‘n’ increases, performance is expected to decline, consistent with the increased working memory load. Response time (RT) was not considered in this task, as participants were not instructed to respond quickly and were given a relatively long response window of 10 seconds.

Face identification task

The primary metric of interest in this task is cognitive control, measured using ΔRT (the difference in response time between repeat sequences and change sequences). Accuracy is not considered in the analyses, as it was only necessary to establish a task goal for participants, but is not relevant for calculating cognitive control.

Trials were excluded from statistical analyses if there was no response on the current or previous trial, or if the response on the current trial (n) was too fast (RT < 200 ms). Additionally, individual RTs were trimmed if they deviated by more than ± 2 standard deviations from the mean RT for repeat and change sequences, respectively.

Expression recognition task

As with the Spatial n-back Task, there are four possible outcomes for each trial⁹⁴: a hit (correct ‘Go’ response to the target expression), a correct rejection (correct ‘No-go’ response to the non-target expression), a miss (incorrect ‘No-go’ response to the target expression), and a false alarm (incorrect ‘Go’ response to the non-target expression).

Trials were excluded if response times (RT) were below 200 ms, indicating an anticipation error. The metric \({{{{\rm{d}}}}}^{{\prime} }\), which measures expression sensitivity, was calculated as the Z-score of the hit rate [#Hits / (#Hits + #Misses)] minus false alarm rate [#False Alarms / (#False Alarms + #Correct Rejections)]. This was computed separately for each emotion expression type. A lower \({{{{\rm{d}}}}}^{{\prime} }\) score indicates reduced sensitivity to the stimulus signal, reflecting greater difficulty in distinguishing between expressions.

Psychomotor vigilance task

The primary metric of interest in this task was Response Time (RT), irrespective of whether participants respond within the 400 ms window required for ‘correct’ response feedback. RT is defined as the time elapsed between perceiving the target visual stimulus and pressing the response button.

Sustained attention was measured by RT following long fixation pauses (ten trials), with a shorter RT indicating better sustained attention. Psychomotor speed was assessed by RT following short fixation pauses (75 trials), with a shorter RT reflecting faster psychomotor speed.

It is expected that average RTs will not differ significantly between the pre-exposure and post-exposure test sessions. This would suggest that exposure to low-quality air does not affect sustained attention or basic psychomotor functioning. If significant RT differences are observed, they may reflect fatigue rather than more complex mechanisms linking air pollution exposure to diminished higher-order cognitive functioning.

Data analysis

Cognitive Tasks: Approach bias (Expression Recognition Task) was analysed using a four-way (2 × 2 × 2 × 2) repeated measures analysis of variance (ANOVA), with inhalation pathway (normal, restricted); pollution exposure (clean air, PM pollution); session time (pre-exposure, post-exposure); and emotion expression (happy, fearful) as the factors. All other cognitive metrics of interest were analysed using a three-way (2 × 2 × 2) repeated measures ANOVA, with inhalation pathway; pollution exposure; and session time as the factors. Bonferroni corrections for multiple comparisons were applied to each analysis. To further investigate interaction effects, one-tailed paired samples T-tests were conducted as necessary. The authors note that the statistical significance of all reported T-tests would be the same if conducted as two-tailed tests.

Air Quality: Pollutant concentrations were averaged across the 1-h exposure period. PM_2.5, PM₁₀, and CO₂ were analysed separately using a two-way (2 × 2) repeated measures analysis of variance (ANOVA), with inhalation pathway (normal, restricted) and pollution exposure (clean air, PM pollution) as the factors. Bonferroni corrections for multiple comparisons were applied to each analysis.

Exposure Awareness: The frequencies of the actual air pollution exposure condition and the self-reported exposure condition were compared using a χ² test.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.