Ventromedial frontoinsular connectivity is associated with long-term smoking behavior change in aging

The global prevalence of Alzheimer’s disease (AD) and other dementias is projected to exceed 139 million by 2050 (Nichols et al., 2022), highlighting the urgent need to enhance disease prevention amid an aging population. Targeting older individuals at risk for AD presents an opportunity for neuroprotection and disease risk reduction (McDade, 2022). In addition to age and APOE ε4 carriership, modifiable vascular risk factors have emerged as significant contributors to cognitive decline and dementia (Gottesman & Seshadri, 2022; Livingston et al., 2020), with cigarette smoking accounting for the third highest population-attributable fraction of dementia cases (5.2%) after low educational attainment and hearing loss (Livingston et al., 2020).

Cigarette smoking is associated with diminished brain health in aging, including increased white matter hyperintensities, acceleration of cerebral atrophy, and poor cognitive functioning (e.g., memory and executive dysfunction) (Anstey et al., 2007; Gottesman & Seshadri, 2022; Livingston et al., 2020; M. C. Power et al., 2015; Rusanen et al., 2011). Smokers are at a higher risk of dementia compared to non-smokers, and dementia risk is reduced for individuals who quit or reduce smoking (Livingston et al., 2020). Quitting smoking, even later in life, has been linked to a reduced risk of cognitive decline in older adults (Choi et al., 2018). For example, an epidemiological study with over 50,000 individuals aged 60 and above found that quitting smoking for more than 4 years significantly lowered the risk of developing dementia in the subsequent 8 years compared to those who continued smoking (Choi et al., 2018). Additionally, another study found that ex-smokers who maintained abstinence for at least 3 years had a similar risk of incident dementia as individuals who had never smoked (Lu et al., 2020).

There is evidence suggesting that reducing smoking, without necessarily quitting entirely, may also mitigate the risk of AD. Tyas et al. (2003) demonstrated a dose-dependent connection between smoking intensity and a heightened risk of AD and AD-related neuropathology, up to heavy smoking levels (>40.5–55.5 packs/year) (Tyas et al., 2003). Smoking reduction is a promising behavioral intervention strategy as a reduction in smoking may be a more attainable goal compared to complete cessation. Additionally, once achieved, it may further encourage attempts to quit without relapse. Currently, however, the neurobehavioral mechanisms underlying smoking behavior change and the mechanisms of AD risk reduction itself in at-risk aging are completely unknown.

Despite the well-established benefits of smoking cessation, sustaining behavior change can be incredibly challenging. Long-term maintenance of behavior change, especially in the context of addictive behaviors, has proven to be a particular challenge due to the risk of relapse. Although some public health messaging and behavioral strategies have helped individuals quit smoking, and nicotine-dependent treatments are available, approximately 40 to 60 percent of individuals return to addictive substances within the first year after cessation (Etter & Stapleton, 2006; Janes et al., 2010; Stead et al., 2008). Understanding the mechanistic underpinnings of behavior change maintenance has relied upon neuropsychological frameworks and cognitive-behavioral theories. Automatic processes, habits, self-efficacy expectations, and positive outcome expectations have been suggested to moderate smoking cessation maintenance (Baldwin et al., 2006; Dijkstra et al., 2007; Hertel et al., 2008). From a cognitive perspective, cognitive control processes, including response inhibition, working memory, and sustained attention, have been associated with smoking behavior change (Evans et al., 2018). Furthermore, to successfully maintain smoking cessation behavior, executive functions, such as inhibitory control, goal-directed decision-making, and self-regulation, are required (Hofmann et al., 2011; Kaplan & Berman, 2010). However, our understanding of the neural mechanisms underlying long-term maintenance of smoking reduction remains unknown, especially in an at-risk aging population. This investigation is especially relevant considering that the main barriers to successful smoking cessation and maintenance are withdrawal effects and cravings (Evans et al., 2018). A reduction in smoking could disrupt the neuroadaptive processes triggered by smoking cues, thereby reducing the likelihood of these cues inducing a desire to smoke in the future.

Human lesion and neuroimaging studies underscore the importance of the insula in smoking cessation and its role in conscious urges. These urges, which arise in response to specific cues, can be considered an emotional response involving a range of bodily responses, conscious feelings, and motivated behaviors triggered by objects of significant value to the individual (Naqvi & Bechara, 2010). Lesion studies have demonstrated a causal relationship between insular integrity, craving, and smoking cessation. Individuals with insular damage from a stroke showed abrupt decreased subjective cue-induced drug urges, less intense withdrawal symptoms, and less nicotine-seeking behavior compared to individuals with strokes in other brain regions (Naqvi et al., 2007). The likely involvement of diaschisis, leading to the loss of afferents from injured cortex, in this case insula, to other network nodes, is also likely at play (Carrera & Tononi, 2014). Lesion network mapping found that focal brain lesions resulting in addiction remission showed differential resting-state functional connectivity (RSFC) fMRI in a network involving the insula (Joutsa et al., 2022). These findings are consistent with task-based fMRI studies that have shown anterior insula (aINS) activation in response to subjective cue-induced drug urges (Naqvi & Bechara, 2010). Heightened activation of the right anterior insula during a decision-making task is associated with an increased likelihood of relapse to drug use (Claus et al., 2013; Janes et al., 2017; Luijten et al., 2011; Moeller et al., 2022; Namkung et al., 2017). Additionally, the aINS is crucial in predicting future reward and comparing present feelings with those from the past and the future, forming the basis of the involvement of this region in craving and drug dependence (Preuschoff et al., 2008). The anatomy and flexible RSFC profile of the aINS makes it a key hub within the salience network, allowing it to coordinate activity with the other major functional networks including the default mode network (DMN) and the central executive network (CEN) (Molnar-Szakacs & Uddin, 2022). Given its large-scale RSFC, the aINS assumes a critical role in higher-order cognitive processes and goal-directed behaviors (Molnar-Szakacs & Uddin, 2022). Emerging evidence from resting-state network models propose that aINS is involved in directing attention towards either internal or external stimuli by mediating the dynamic activity between the other two large scale networks: DMN and the CEN (Sutherland et al., 2012).

In the present study, we aimed to assess whether altered functional connectivity of the aINS at baseline predicted successful change in smoking behavior (i.e., decreased or stopped smoking) longitudinally in older adults at risk for AD. We applied seed-to-voxel RSFC to address this guiding question in two separate studies. RSFC allows for the exploration of large-scale networks, and their interactions, thus moving towards a systems-level understanding of brain function. First, we examined whether aINS RSFC predicted long-term smoking behavior change (mean = 2.7 years) in cognitively normal current smokers followed longitudinally in the PRE-symptomatic Evaluation of Experimental or Novel Treatments for Alzheimer’s Disease (PREVENT-AD) cohort. We focused on this high-risk group as they most stand to benefit from smoking behavior change for disease prevention. We sought to ensure the robustness and generalizability of our results in an independent and larger sample. Next, we conducted the independent replication study utilizing the UK Biobank longitudinal cohort. In the replication study, our sample consisted of cognitively normal older adult (≥60 years) current smokers who were longitudinally followed for an average duration of 5.2 years. We employed a rigorous and comprehensive approach to replication, aiming to strengthen the validity of our conclusions.

Based on the results of the discovery sample, we preregistered the replication study design and hypothesis on the Open Science Framework platform (https://doi.org/10.17605/OSF.IO/BJV63). Crucially, the entire exploratory analysis of the discovery dataset was finalized before the preregistration was written and uploaded. Our preregistered research question, “Does frontoinsular connectivity successfully predict long-term smoking behavior change in cognitively unimpaired, but at risk, older adults?” was based on the finding of diminished aINS to vmPFC functional connectivity in the discovery cohort. Our primary hypothesis was “Increased smoking behavior is associated with increased functional connectivity between aINS and vmPFC” in the UK Biobank replication cohort. The preregistered analytic plan was described as follows: “we will use a seed-to-voxel connectivity approach to study the relationship between bilateral anterior insula and smoking behavior change. We will be using the CONN toolbox to do these analyses. We will run a series of linear regression models to assess the effect of behavioral predictors on the smoking behavior change.” This analytic approach was identical to that used in the finalized results obtained from the PREVENT-AD discovery sample.

Twenty-three (18 women) cognitively unimpaired older individuals who currently smoked cigarettes (mean age = 61.5 years, SD = 3.7) with a first-degree family history of AD (at least one parent or multiple siblings affected) were included from a prospective longitudinal cohort, PREVENT-AD. Out of 425 participants, our initial sample size consisted of 25 participants; 2 participants were removed from the final analyses for having poor-quality brain imaging data, resulting in a final sample size of 23 participants. Protocols and study procedures were approved by the McGill Institutional Review Board and/or Douglas Mental Health University Institute Research Ethics Board in accordance with the Declaration of Helsinki. Briefly, 399 cognitively healthy older individuals were enrolled between September 2011 and November 2017. Data from the present study are from Data Release 5.0. Inclusion criteria for the current cohort subsample were 1) age 60 years or older (or 55–59 years for individuals who were less than 15 years from the age of their relative at symptom onset); 2) no history of major neurological or psychiatric disease; 3) cognitively unimpaired at the time of enrollment; and 4) reported being a current cigarette smoker at the baseline timepoint. Normal cognition was defined as a Clinical Dementia Rating (CDR) global of 0 and a Montreal Cognitive Assessment (MoCA) score of ≥26/30. Borderline individuals who did not fit the above criteria of normal cognition underwent a comprehensive diagnostic assessment by a trained neuropsychologist. All participants underwent longitudinal neuropsychological, AD genetic biomarker and serial MRI brain imaging assessments. All 23 participants who reported being current smokers at baseline, and for whom baseline structural and resting-state functional MRI (rs-fMRI) were available, were included in the present study. Brain imaging was obtained at baseline, and self-reported smoking outcome data were obtained for two time-points: Baseline and follow-up (mean difference = 2.7 years, SD = 1.56). Figure 1 shows the flowchart of participant inclusion and exclusion criteria for the PREVENT-AD sample. Tables 1 and 2 describe the demographic and neuropsychological information for this cohort, respectively.

One hundred and eighteen cognitively unimpaired older individuals who currently smoked cigarettes (mean age = 60.32, SD = 2.70) were included from the UK Biobank cohort, a large population longitudinal cohort. This final sample size of 118 was a result after removing six participants for having poor brain-imaging data. Single nucleotide polymorphism (SNP) data for rs-429358 and rs-7412 were used to determine APOE genotypes. Inclusion criteria were 1) age 60 years or older; 2) no history of major neurological or psychiatric disease; 3) cognitively normal at the time of enrollment; and 4) reported being current cigarette smokers at the baseline timepoint. This was based on previous recommendations (Lyall et al., 2016). Normal cognition was defined as follows: performance scores on each of the cognitive test were converted into percentile rank, and the raw score corresponding to the 5^th percentile (or 95^th, on tests where higher scores represented worse performance) was identified as the cut-off for impairment (Lyall et al., 2016). Imaging visit (Instance 2 of the UK-Biobank) was considered baseline timepoint, and the first repeat imaging visit (Instance 3 of the UK-Biobank) was considered the follow-up timepoint. Participants who reported being current smokers at baseline and for whom baseline structural and resting-state functional MRI data were available were included in the present study. Brain imaging was obtained at baseline, and self-reported smoking outcome data were obtained for two timepoints: Baseline and follow-up (mean duration = 5.2 years, SD = 1.07). Tables 1 and 3 describe the demographic and cognitive function information for the UK Biobank sample, respectively. Figure 2 shows the flowchart of participant inclusion and exclusion criteria for this cohort.

Smoking habits were assessed using the seven items from the Educoeur questionnaire smoking sub-scale (number of cigarettes/day, age started smoking) (Goyer et al., 2013) at two timepoints, baseline and follow up (mean = 2.7 years, ranging from 8 months to 5 years). The behavioral outcome of interest (i.e., successful smoking behavior change) was defined as the difference between the number of cigarettes per day measured at follow-up compared to the number of cigarettes per day measured at baseline.

Questions about smoking habits were taken from the touchscreen questionnaire on smoking habits (duration of smoking, number of cigarettes/day, age started smoking). Individuals who report their current smoking status as “current smokers” at the first brain imaging timepoint were included in the study. The behavioral outcome of interest (i.e., successful smoking behavior change) is defined as the difference between the number of cigarettes per day measured at follow-up compared to the number of cigarettes per day measured at baseline (mean = 5.2 years, ranging from 3 years to 7 years).

Cognitive function was assessed using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS). This neuropsychological battery is a widely used and validated inventory. It includes five cognitive domains: Immediate Memory (i.e., list learning, story remembering), Visuospatial Ability (i.e., figure copy, line orientation), Language (i.e., picture naming, semantic fluency), Attention (i.e., digit span, coding), and Delayed Memory (i.e., list recognition, story recall, and figure recall) (Randolph et al., 1998). To reduce test-retest effects, four different versions of the RBANS were used in follow-up. The battery was administered annually to all participants enrolled in PREVENT-AD. Index scores are age-adjusted and have a mean of 100 with a standard deviation of 15.

Brief cognitive tests were administered via touchscreen. The tests were designed specifically for UK Biobank but share some characteristics with other established tests of cognition. The following tests were included: Reaction Time, Numeric Memory, Fluid intelligence, Matrix pattern completion, Tower rearranging, and Trail making.

Participants were instructed to quickly press a button with their dominant hand whenever a matching pair of symbols appeared on the screen. The test encompassed five practice trials followed by seven test trials. The metric analyzed was the average time (measured in milliseconds) taken to press the button. This value was derived from the four trials where a matching pair emerged (recorded in UK Biobank data field 20023). Higher scores indicated slower (i.e., worse) performance.

Participants were shown a sequence of numbers on the screen. After a brief pause, they were asked to recall and input the sequence in reverse order using a numeric keypad. Each number sequence was displayed for 2000 ms, with an additional 500 ms added for each digit in the sequence. A delay of 3000 ms occurred between clearing the screen and activating the response keypad. The test began with a sequence length of two digits and progressed to longer sequences, increasing by one digit each time, up to a maximum length of 12 digits. The test ended either after five consecutive incorrect responses for two-digit sequences or after two consecutive incorrect responses for sequences of three digits or more. The analysis score was based on the longest correctly recalled sequence length (recorded in UK Biobank data field 4282), with higher scores indicating better performance.

Thirteen questions were presented in succession on a touchscreen for the Fluid Intelligence test. Participants could respond at their own pace within a 2-minute time frame. Responses were chosen from a set of multiple-choice options. Any questions not attempted within the 2-minute period received a score of zero. The analysis score represented the total number of correct answers, ranging from 0 to 13 (UK Biobank data field 20016), with higher scores indicating better performance.

Non-verbal fluid reasoning was assessed through an adapted version of the COGNITO Matrices test (De Roquefeuil Guilhem, 2014). Participants were presented with matrix pattern blocks, each with a missing element. Their task was to select the element that completed the pattern best from a range of choices. The items varied in difficulty across a total of 15 items, and the analysis score was the count of correctly answered items within a 3-minute interval.

To gauge planning abilities, an adapted form of the One-touch Tower of London test (Hampshire et al., 2013) was employed. Participants were shown illustrations featuring three differently colored hoops positioned on three pegs. Their task was to determine the number of moves needed to rearrange the hoops to a specified configuration. The analysis score was the tally of correctly answered items within a 3-minute period.

The UKB Trail Making Test (TMT) is a digital adaptation of the widely recognized Halstead-Reitan Trail Making Test (Reitan & Wolfson, 1985), often regarded as an assessment of executive function. The test comprised two parts. Part A displayed numbers 1–25 arranged pseudo-randomly on the screen, requiring participants to touch them in numerical order. Part B presented numbers 1–13 and letters A–L arranged pseudo-randomly, necessitating participants to alternate between touching numbers in numeric order and letters in alphabetical order (e.g., 1-A-2-B-3-C). Participants were instructed to complete each part as swiftly and accurately as possible. Incorrect responses prompted a red screen flash, requiring participants to correct their answer before proceeding. The analysis score represented the time taken, in seconds, to complete each part.

All participants underwent an MRI scanning session in a 3T Magnetom Tim Trio (Siemens) scanner using a Siemens standard 12-channel head coil. T1-weighted structural images were obtained using a GRE sequence with the following parameters: TR = 2300 ms; TE = 2.98 ms; flip angle = 9°; FOV = 256 × 240 × 176 mm; voxel size = 1 × 1 × 1 mm. For resting-state functional MRI scans, two consecutive functional T2*-weighted runs were collected with eyes closed using a blood oxygenation level-dependent (BOLD) sensitive, single-shot echo planar sequence with the following parameters: TR = 2000 ms; TE = 30 ms; flip angle = 90°; FOV = 256 × 256 mm; voxel size = 4 × 4 × 4 mm; 32 slices; 150 volumes; and acquisition time = 5 minutes 4 seconds per run.

Details of the image acquisition and processing are available on the UK Biobank Protocol (http://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=2367), and Brain Imaging Documentation (http://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=1977). Briefly, all brain MRI data were acquired on a single standard Siemens Skyra 3 T scanner with a standard Siemens 32-channel RF receiver head coil, with the following parameters: TR = 2000 ms; TI = 800 ms; R = 2; FOV = 208 × 256 × 256 mm; voxel size = 1 × 1 × 1 mm. For resting-state fMRI scans, two consecutive functional T2*-weighted runs were collected with eyes closed using a BOLD sensitive, single-shot echo planar sequence with the following parameters: TR = 735 ms; TE = 39 ms; flip angle = 52°; FOV 88 x 88 x 64 matrix; resolution = 2.4 × 2.4 × 2.4 mm; 490 volumes; and acquisition time = 6 minutes per run.

Preprocessing of raw functional images from both cohorts was completed using an identical approach. This included application of fMRIprep (version 20.2.4) (Esteban et al., 2019). For each of the BOLD runs per subject, the following preprocessing was performed: First, the T1w reference was skull-stripped using a Nipype implementation of the antsBrainExtraction.sh tool (ANTs). A B0-nonuniformity map (or fieldmap) was estimated based on a phase-difference map calculated with a dual-echo gradient-recall echo (GRE) sequence, which was then co-registered to the target echo-planar imaging (EPI) reference run and converted to a displacements field map. A distortion-corrected BOLD EPI reference image was constructed and registered to the T1-weighted reference using a boundary-based approach (bbregister, Freesurfer). Rigid-body head-motion parameters with respect to the BOLD EPI reference were estimated (mcflirt, FSL 5.0.9) (Jenkinson et al., 2002) before spatiotemporal filtering was performed. BOLD runs belonging to the single band acquisition sessions were slice-time corrected (using 3dTshift, AFNI 20160207). The BOLD time series were resampled into their original, native space by applying a single, composite transform to correct for scan-to-scan head motion and susceptibility distortions. Functional scans were spatially smoothed using a 6 mm full width at half maximum (FWHM) Gaussian smoothing kernel.

Additional preprocessing steps were undertaken to remove physiological, subject-motion, and outlier-related artifacts, which were implemented in the CONN toolbox (version 20.c, https://www.nitrc.org/projects/conn/) (Whitfield-Gabrieli & Nieto-Castanon, 2012). The anatomical image for each participant was segmented into white matter, gray matter, and cerebrospinal fluid (CSF) masks using SPM12. White matter and CSF masks were eroded by one voxel to minimize partial volume averaging. Non-neuronal sources of noise from white matter and CSF were estimated and removed using the anatomical CompCor method (aCompCor) (Behzadi et al., 2007) to allow for valid identification of correlated and anticorrelated networks (Chai et al., 2012; Murphy et al., 2009). During the outlier detection step, images with framewise displacement (FD) above 0.9 mm or global mean intensity above 5 standard deviations were flagged as outliers using Artifact Detection Tools (ART, www.nitrc.org/projects/artifact_detect). This cutoff was determined based on preserving at least 5 minutes of scanning time (Van Dijk et al., 2010). Temporal band-pass filtering (0.008–0.09 Hz) was also applied.

Additionally, scan-to-scan mean head motion (framewise displacement) was used as a covariate of non-interest in all second-level analyses (mean head motion = 0.2 mm, SD = 0.1 mm). Head motion is a known important potential confound as it produces systematic and spurious patterns in connectivity, and is accentuated in AD and cognitively typical aging populations (Power et al., 2012). Critically, we did not identify a relationship between the mean head motion parameter and smoking behavioral change (all p > 0.05). To ensure that results were not influenced by the head motion threshold, we repeated our preprocessing with a more conservative 0.5 mm scan-to-scan framewise displacement to ensure our results effectivity replicate (see Supplementary Fig. 1). The framewise displacement timeseries was determined by calculating the maximum shift in the position of six control points situated at the center of a bounding box around the brain, computed independently for each scan. Two participants were removed from the PREVENT-AD sample final analysis for having >30 scan volumes flagged, leading to the final sample size of 23 participants. Similarly, six participants were removed from the UK-Biobank cohort for the same reason, resulting in the final sample size of 118 participants. This cut off was determined based on preserving at least 5 minutes of scanning time (Van Dijk et al., 2010).

To investigate the relationship between brain connectivity and long-term smoking behavior change, seed-to-voxel analyses were implemented identically in the PREVENT-AD sample and the UK Biobank as a replication and extension sample. Resting-state functional brain connectivity analyses were carried out using the CONN toolbox (version 20.c, https://www.nitrc.org/projects/conn/) (Whitfield-Gabrieli & Nieto-Castanon, 2012). The average time series in the anatomically defined region of interest (ROI), bilateral anterior insula (aINS) was extracted (Fig. 3a). The ROI was defined based on the Harvard-Oxford atlas. Product-moment correlation coefficients were computed between the average time series within the ROI and the time series within all other voxels in the whole brain and converted to normally distributed Fisher transformed z-scores prior to entering them in the second-level general linear model. Individual change in smoking was entered as a covariate of interest in the second-level analysis, controlling for nuisance variables including age, sex, scan-to-scan mean head motion, APOE4 carrier status, and baseline smoking amount (number of cigarettes/day) in a general linear model for the bilateral aINS.

To determine whether the aINS showed an expected pattern of whole-brain connectivity in our sample, we first conducted a seed-to-voxel analysis for the whole sample without the behavioral covariate of interest. Figure 3b illustrates the functional connectivity pattern from the aINS to the entire brain for all participants. The expected insular-to-whole-brain connectivity pattern is evident, including positive functional connectivity between the aINS and regions comprising the salience network (e.g., anterior cingulate cortex (ACC) and frontal operculum), as well as negative correlations between the aINS and nodes of the DMN (e.g., medial prefrontal cortex, inferior parietal lobule, and posterior cingulate cortex). Next, we performed Generalized Linear Model (GLM) analysis to examine the association between smoking behavior change and the functional connectivity between bilateral aINS and the whole brain. The results showed that decreased smoking at follow-up compared to baseline was associated with diminished functional connectivity between bilateral aINS and a cluster within the vmPFC that encompasses the following regions: frontal pole, frontal medial cortex, subcallosal cortex, paracingulate gyrus, anterior cingulate cortex, and frontal orbital cortex (Bhanji et al., 2019). Notably, the peak voxel was localized in subcallosal cortex (F(2,17) = 19.48; voxel height p < 0.001 uncorrected; cluster p-FWE < 0.05; k = 112 voxels; Peak MNI voxel = 0, 30, -8 [x, y, z coordinates]) (Fig. 4a). To ensure that our results were not driven by outliers, we performed a post-hoc robust regression analysis to visualize and understand functional connectivity at the level of the individual. We extracted the mean Z-value from the vmPFC cluster for each participant and plotted this against change in smoking across the two timepoints. Findings of the post-hoc analysis showed that our results were not driven by outliers, r² = 0.77, p < 0.001 (Fig. 4b).

To replicate and extend the findings in our discovery sample, we examined whether aINS connectivity was associated with long-term smoking behavior change in an independent and larger sample from the UK Biobank (Fig. 5). Seed-to-voxel analyses showed that decreased smoking in follow-up compared to baseline was associated with diminished RSFC between bilateral aINS and vmPFC, with the peak voxel located in ACC, (F(2, 17) = 17.48; voxel height p < 0.001 uncorrected; cluster p-FWE < 0.05; k = 60 voxels; Peak MNI voxel = 0, +18, +24 [x, y, z coordinates]) (Fig. 5a). We then extracted the vmPFC cluster voxels from the discovery dataset (PREVENT-AD) and used this as an a-priori region of interest in the replication analysis. An ROI-to-ROI analysis was performed that showed a significant positive association between the bilateral aINS and the extracted vmPFC ROI.

To again ensure that our results were not driven by outliers, we performed a post-hoc robust regression analysis to visualize RSFC at the level of the individual. Excluding outliers over 2.5 SD from mean resulted in the exclusion of six participants. We extracted the mean Z-value from the vmPFC ROI as defined in the PREVENT-AD sample, for each participant and plotted this against change in smoking across the two timepoints. Findings of the post-hoc analysis showed that our results were not driven by outliers, r² = 0.42, p < 0.001 (Fig. 5b).

To render this line of inquiry clinically meaningful, it is important to delineate the shift in smoking behaviors toward a health-promoting trajectory, marked by a reduction or cessation of smoking. Subsequently, we repeated the seed-to-voxel analysis after excluding participants who increased their smoking in both cohorts, ensuring the robustness of our results. We found that the relationship between bilateral aINS and the vmPFC cluster remained significant in the PREVENT-AD sample, with the peak voxel in the subcallosal cortex, (F(2, 17) = 5.60; voxel height p < 0.001 uncorrected; cluster p-FWE < 0.05; k = 178 voxels; Peak MNI voxel = +40, +30, -10 [x, y, z coordinates]). Other significant regions within this vmPFC cluster included the paracingulate gyrus, frontal pole, and anterior cingulate gyrus. It is crucial to exercise caution in interpreting these results, given the reduced sample size in this subsample, consisting of only 11 participants (refer to Supplementary Fig. 2a for details). Similarly, in the UK Biobank replication sample, the association between bilateral aINS and the vmPFC cluster persisted when participants who increased smoking in follow up were excluded, with the peak voxel in ACC (F(2, 13) = 10.58; voxel height p < 0.001 uncorrected; cluster p-FWE < 0.05; k = 75 voxels; Peak MNI voxel = +06, +34, +12 [x, y, z coordinates]) (Supplementary Fig. 2b).

In order to evaluate the precision of functional connectivity measurement in predicting changes in smoking behavior, we assessed the test-retest reliability using the Intra Correlation Coefficient (ICC) at the individual edge level, within the UK Biobank Replication Sample. The choice of ICC was based on its widespread use for assessing edge-level reliability, ensuring compatibility with other studies (Noble et al., 2019). Typically, ICC values are calculated for fMRI sessions that lasted 10–15 minutes duration, with an inter-session interval one day and 593 over one month (Noble et al., 2019). We compared examining functional connectivity matrices between baseline and follow-up timepoints. The average reliability of edge-level functional connectivity was found to be fair (ICC µ = 0.42, 95% CI = 0.39 to 0.46). However, this analysis is critically confounded by age-related effects given the long period between the two imaging timepoints (~5 years). Longer intervals are associated with decreasing reliability, which may partly be attributed to age-related brain changes over time (Noble et al., 2019). Furthermore, many lifestyle variables that were not controlled for in this analysis, including smoking, are intricately linked to brain connectivity, and conversely, smoking itself can alter brain connectivity (Livingston et al., 2020; Naqvi & Bechara, 2010; Sutherland et al., 2016). Consequently, in this analysis, we expect the connectivity patterns to differ between the two timepoints. Unfortunately, the UK Biobank only provides one scan session per timepoint. Therefore, considering these factors collectively, reliability scores in this cohort may not provide an accurate reflection of the precision of the connectivity measurements.

Traditional significance testing, while informative, may lead to false negatives when sample size and power are constrained (Gerchen et al., 2021). Conversely, in large datasets, the risk of observing spurious correlations is increased. To assess for replication effects of the findings from our PREVENT-AD discovery sample in the UK Biobank replication sample, we employed equivalence testing, calculating Hedges’ g (the bias-corrected version of Cohen’s d) and its confidence interval. This approach enables a formal comparison of effect sizes within and between datasets. Replication was assessed by determining whether the effect sizes from the reference sample (PREVENT-AD) fell within the 90% confidence interval of the voxel’s effect size in the replication sample (UK Biobank). T maps were thresholded as per Gerchen et al. (2021) at p < 0 .05 whole‐brain FDR corrected for both the PREVENT-AD and UK-Biobank samples. Effect Sizes, and their confidence intervals (CI) were estimated from SPM t maps (https://github.com/Fungisai/g_ci_spm). The effect size map for the PREVENT-AD sample is shown in Figure 6a. Notably, the effect size of the vmPFC region in the PREVENT-AD sample was replicated in the UK Biobank (as shown by the blue circles) (Fig. 6b). The voxel‐wise test reveals additional regions involved beyond the overlap of significant effects. Large effect sizes were replicated in areas highlighted by black boxes where no significant effects were detected in the discovery sample.

In an exploratory analysis, we investigated the relationship between aINS and smoking behavior change in older non-smoking individuals by adopting the opposite set of inclusion criteria to understand the generalizability of aINS-vmPFC functional connectivity as a marker of future smoking behavior. Specifically, we focused on participants from a new sample who were non-smokers at baseline but initiated cigarette smoking in follow-up after an average of 6 years (ranging from 3 years to 7 years). This subsample consisted of 13 (7 women) cognitively unimpaired participants from the UK Biobank (mean age = 58±3.4). Of these, 11 participants reported their smoking status at baseline as previous smokers, while 2 reported never having smoked. The outcome variable of interest (i.e., change in smoking behavior) was measured as a continuous variable. Again utilizing seed-to-voxel analysis with the aINS as the ROI, we found that greater daily cigarette number smoked at follow-up was linked to enhanced functional connectivity between aINS and the left frontal pole at baseline (F(2, 8) = 8.64; voxel height p < 0.001 uncorrected; cluster p-FWE < 0.05; k = 16 voxels; Peak MNI voxel = -36, +46, +28 [x, y, z coordinates]) (Fig. 7a). To ensure that our results were not driven by outliers, we performed a post-hoc robust regression analysis to visualize and understand functional connectivity at the level of the individual. Excluding outliers over 2.5 SD from mean resulted in the exclusion of one participant. We extracted the mean Z-value from the left frontal pole for each participant and plotted this against change in smoking across the two timepoints. Findings of the post-hoc analysis showed that our results were not driven by outliers, r² = 0.86, p < 0.001 (Fig. 7b). Given the small sample size, interpretation of these results must be approached with caution.

The primary goal of the current study is to characterize the neural substrates associated with long-term smoking reduction in typical and at-risk aging. In a discovery and replication cohort, we observed that higher antagonistic aINS-vmPFC RSFC predicted future smoking reduction. We conducted a hypothesis-driven RSFC seed-to-voxel analysis in a primary sample of current smokers who are cognitively unimpaired older adults at an increased risk for AD from the PREVENT-AD cohort. We identified that diminished baseline functional connectivity between aINS and vmPFC involving frontal pole, frontal medial cortex, subcallosal cortex, paracingulate gyrus, ACC, and frontal orbital cortex, with a peak voxel in the subcallosal cortex, was associated with decreased smoking in long-term follow-up after 2.7 years. The robustness of our results was confirmed through a pre-registered replication in a larger sample of cognitively unimpaired older adults who were current smokers from the UK Biobank longitudinal cohort. Both studies applied identical preprocessing and analytic approaches. In the replication study, we found that smoking maintenance behavior was again associated with aINS RSFC: Decreased smoking at long-term follow-up after 5.2 years was associated with decreased aINS-vmPFC RSFC, with a peak voxel in ACC. In addition we assessed replication effects through equivalence testing and again replicated our results within a 90% confidence interval. This supports the validity and generalizability of our results and underscores the importance of aINS-vmPFC RSFC as a brain marker of long-term smoking reduction maintenance in typical and among older adults at risk for AD.

The prior literature on the neural circuitry underlying successful smoking behavior change is controversial, with some studies showing that prefrontal hypoactivation is linked to increased drug use, consistent with the finding that stimulating the prefrontal cortex could serve as a treatment target for addiction (Chen et al., 2013). Conversely, evidence from human lesion studies present a paradoxical scenario wherein nicotine-addicted individuals have achieved abrupt and effortless smoking cessation following damage to the insular cortex (Naqvi et al., 2007). At first glance, these two lines of inquiry (lesion studies on the one hand, and functional neuroimaging on the other) seem contradictory. However, our results help harmonize these seemingly disparate views by demonstrating that the connectivity between nodes in this circuit, rather than activity or function of a single region, is crucial for successful smoking cessation. This convergence of evidence helps refine our understanding of the integrated role of insula and prefrontal cortex in successful smoking cessation behavior and underscores the importance of considering neural circuit connectivity in the context of smoking cessation. Our findings showed an association between aINS connectivity and smoking behavior change across three separate samples: the PREVENT-AD discovery cohort, the UK Biobank replication sample, and in a third sample from the UK Biobank of non-smokers who take up cigarette smoking in follow-up. Future studies combining functional neuroimaging and lesion methods will help strengthen the causal cascade between insular injury and diminished aINS-vmPFC RSFC. Separately, characterizing the potential contribution of component cognitive processes such as emotion regulation, temporal discounting, and cognitive control from the role of cravings and urges will be an important future direction of this work.

Our central findings are in line with findings coordinated activity between the aINS and ACC in goal-directed behavior and decision-making (Craig, 2002; Fedota et al., 2018; Janes et al., 2010, 2015; Menon & Uddin, 2010; Uddin, 2015). In line with these roles, the anterior insula is critical for awareness of bodily urges, while the ACC is important in initiating behaviors and resolving conflict (Craig, 2002). Prior evolutionary research provides a context for the coordinated activity between these regions; ACC evolved first as a motor-control region aligned with the sensory integration of olfactory-guided group behavior in mammals, while the insula evolved later for cortical processing of homeostatic sensory activity in the individual animal (Heimer & Van Hoesen, 2006). Their conjoint activation is therefore thought to constitute a system of self-awareness: conscious interoception arising from the anterior insula is re-represented in the anterior cingulate cortex, forming the basis for selecting and preparing responses to inner or outer events (Allman et al., 2011; Janes et al., 2015; Medford & Critchley, 2010). The integration of interoceptive representations is thought to progress sequentially from the posterior insula to the anterior insula (Craig et al., 2000; Olausson et al., 2002). The primary interoceptive stimuli, encompassing numerous individually mapped and distinct feelings from the body, are established in the PCC (Craig, 2002; Craig et al., 2000). This information is then transmitted to the anterior insula, where re-represented interoceptive signals are integrated with emotional, cognitive, and motivational signals from other cortical and subcortical regions, such as the amygdala, the anterior cingulate cortex, the dorsolateral prefrontal cortex, and the ventral striatum (Brooks et al., 2002; Craig, 2002; Craig et al., 2000).

The role of insula and vmPFC in reward-related decision making and thereby, smoking behavior change is well established (Bechara et al., 2002; Bechara & Damasio, 2002; Mitchell, 2004; Spinella, 2002). VMPFC is a node in the reward and DMN that is relatively spared in aging and AD, integrates value, social and self-referential processing, and is linked to successful behavior change (Alexander et al., 2023; A. Gutchess & Samanez-Larkin, 2019; A. H. Gutchess et al., 2007; Damasio et al., 2000; Kang et al., 2018; Naqvi et al., 2007; Salat et al., 2005; Winecoff et al., 2013). For example, a study that applied graph theory and modularity analyses showed that higher vmPFC flexibility (i.e., the degree to which the vmPFC changes its community assignment over time) is positively correlated with smoking behavior change a month later (Cooper et al., 2018). Our current findings support the existing body of work suggesting aINS-vmPFC regions work in concert to maintain smoking change behavior.

So far, our focus has been on highlighting the roles played by aINS and vmPFC in supporting successful behavior change. In a third exploratory sample, we also showed that the degree to which cognitively unimpaired older non-smoking participants who relapsed or took up smoking at follow-up exhibited increased RSFC between the aINS and the left frontal pole, aligns coherently with our primary findings. This observation provides preliminary support for the idea that diminished fronto-insular connectivity is important for long-term successful and sustained smoking behavior change (i.e., without relapse). In fact, the frontal pole has continued to emerge as a peak therapeutic target for transcranial magnetic stimulation (TMS) in the context of substance use disorders (Cho et al., 2015; Hanlon et al., 2015, 2018; Joutsa et al., 2022; Kearney-Ramos et al., 2018). This finding, together with other classic findings indicating that damage to the FPC results in planning impairments (Okuda et al., 2007; Shallice & Burgess, 1991), strengthens the hypothesized role for the frontopolar cortex (FPC). The FPC is thought to be crucial for metacognitive decision-making and goal-directed persistence and is implicated in functions such as anticipating future events and outcomes, motivating cognitive and physical effort, and exercising metacognitive control (Boorman et al., 2009; Mansouri et al., 2017). The ability to plan an uncertain but potentially controllable future is essential for complex planning (Hosoda et al., 2020). Impaired control over impulsive choices can lead to suboptimal planning, especially considering that complex plans require persistence for execution and testing of their feasibility (Hosoda et al., 2020).

The current findings are in line with the framework of Incentive Sensitization Theory (IST). IST posits that drug addiction is characterized by a hyper-reactive mesolimbic dopaminergic system that responds to rewarding cues, resulting in heightened “wanting” to consume more drugs (Berridge & Robinson, 2016; Olney et al., 2018). This phenomenon, known as incentive salience, refers to the attention-grabbing and motivational features of rewards and their associated cues (Bindra, 1978). Reward cues have the ability to trigger bursts of reward-seeking motivation and can become a “motivational magnet” over time (Olney et al., 2018; Peciña & Berridge, 2013; Robinson et al., 2015). The incentive salience circuitry is mediated by larger brain systems involving mesocorticolimbic dopamine, the central amygdala (CeA), lateral hypothalamus (LH), orbitofrontal cortex (OFC), ventromedial prefrontal cortex, and the insular cortex (Castro et al., 2015; DiFeliceantonio & Berridge, 2016; Naqvi & Bechara, 2010; Olney et al., 2018; Warlow et al., 2017). In line with IST, disruption of frontoinsular circuitry would be expected to reduce incentive salience or ‘wanting’ with subsequent reduction in smoking behavior.

Indeed, Naqvi et al. (2014) proposed a model for insula involvement in drug addiction-related decision making: The insula plays a key role in weighing the positive and negative consequences of addictive substances. According to the model, the insula encodes the positive hedonic effects and the physical sensations experienced during drug use, which are recalled during drug cravings through coordinated activity with the amygdala, orbitofrontal cortex (OFC), vmPFC, and dorsolateral prefrontal cortex (DLPFC). This can manifest as an urge to use the drug. Simultaneously, the insula represents the negative consequences of drug use in terms of its impact on bodily integrity, survival, and homeostasis. The anterior cingulate cortex is responsible for detecting conflicts between the goal of drug use and other goals and shifting attention accordingly. Therefore, the decision to abstain depends upon the ability to suppress the representation of the positive hedonic effects of drug use and to enhance the representation of the negative consequences of drug use. This suggests that successful and sustained smoking cessation occurs when the representation of the negative consequences of drug taking outweighs the representation of the positive hedonic effects of drug taking. Insular lesions might support successful behavior change as the hedonic impact would be significantly diminished. Evaluation and support for the model, however, has largely been limited to lesion studies and acute smoking abstinence behavior, over only a few days to months.

Smoking might exert different effects on insular connectivity depending on the AD risk level—be it high or low. Past MRI investigations have showcased loss of insular gray matter (Guo et al., 2012) and disrupted insular networks in the early stages of AD (Xie et al., 2012). These studies have underscored the significance of the insula, which emerges as a potentially vulnerable region and a pivotal hub in AD patients. A recent study explored the potential effects of smoking on insula functional connectivity among patients with mild cognitive impairment (MCI) and found that anterior insular connectivity decreased among smokers with MCI (Zhang et al., 2023).

Our study provides support for expanding the investigation of neurocognitive processes underlying sustained behavior change to include smoking reduction, in addition to cessation. Typically, smoking outcomes are dichotomized in studies as quitters versus non-quitters. However, any reduction in smoking behavior, even if not leading to complete cessation, can still have positive effects on cognitive functioning and lower the risk of dementia (Begh et al., 2015). Moreover, individuals who successfully reduce their smoking are more likely to attempt and achieve complete cessation (Balfour, 2009). Thus, examining the neural substrates of smoking reduction provides valuable mechanistic insights with translational potential in individuals at risk for AD.

It is important to acknowledge that a key limitation of the current study is the correlational nature of functional connectivity analyses which do not allow for determination of causality. Additionally, our replication cohort (UK-Biobank) significantly differs in its composition when compared to our primary discovery cohort, PREVENT-AD. Notably, PREVENT-AD consists mainly of French-Canadian participants, a demographic distinction that does not hold true for the UK-Biobank dataset. Furthermore, the UK-Biobank is not specifically designed to study individuals at risk for AD. Similarly, the PREVENT-AD dataset was not explicitly designed to explore the effects of smoking, and as such detailed smoking data, beyond that of self-reported smoking status, is lacking. Taken together, further replication is needed in diverse cohorts to validate our findings.

Despite these limitations, our study has important strengths, including the use of a longitudinal sample of older adults at risk for AD and a rigorous and hypothesis-driven approach in a pre-registered larger replication and extension study. The translational potential of our findings include the identification of the aINS-vmPFC circuit as a potential therapeutic target for smoking cessation strategies. Neurostimulation approaches to behavior change are in their infancy with some early evidence in physical activity engagement in aging (Lo et al., 2021) as well as in smoking cessation (Dinur-Klein et al., 2014). Furthermore, our study highlights the importance of a brain-based mechanistic approach to understanding behavior change and resilience in typical and at-risk aging.

Publicly available software used for all the analyses is CONN toolbox (https://web.conn-toolbox.org/). The individual-level UK Biobank data can be obtained from https://www.ukbiobank.ac.uk/. Requests to access the PREVENT-AD dataset used in this study can be made to the StoP-Alzheimer team: https://prevent-alzheimer.net/?page_id=1760&lang=en.

N.T.: Conceptualization, Methodology, Formal analysis, Investigation, Data curation, Writing—original draft, Writing—review & editing, and Visualization. M.A.: Investigation, Writing—review & editing. A.N.C. and C.W.: Data curation, Writing—review & editing. G.B.: Methodology, Data curation. J.T.-M.: Methodology, Data curation, Resources, and Writing—review & editing. S.V: Methodology, Resources, Writing—review & editing. N.S: Methodology, Resources, and Data curation. M.R.G: Conceptualization, Methodology, Resources, Writing—review & editing, Funding Acquisition, and Supervision.

This Project has been made possible by a National Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant (DGECR-2022-00299), an NSERC Early Career Researcher Supplement (RGPIN-2022-04496), a Québec Brain Imaging Network Grant, a Fonds de Recherche Santé Quebec (FRSQ) Salary Award, the Canada Brain Research Fund (CBRF), an innovative arrangement between the Government of Canada (through Health Canada) and Brain Canada Foundation and the Alzheimer’s Society of Canada New Investigator Grant to M.R.G. This research was undertaken thanks also to funding from the Canada First Research Excellence Fund, awarded to the Healthy Brains Healthy Lives initiative at McGill University to N.T.

The authors declare no competing interests.

This study utilized data from the UK-Biobank study and the PREVENT-AD study. The UK-Biobank has obtained ethics approval from the North West Multi-Centre Research Ethics Committee (MREC, approval number: 11/NW/0382), and obtained written informed consent from all participants prior to the study. All protocols and study procedures in the PREVENT-AD study were approved by the McGill Institutional Review Board and/or Douglas Mental Health University Institute Research Ethics Board in accordance with the Declaration of Helsinki.

We are extremely grateful to the research participants who dedicated their time to participating in this study. We would also like to thank lab members who participated in data collection for the PREVENT-AD study, including Maggie Nguyen, Sarah Elbaz, Andréanne Powers, Linda Li, Sarah Elbaz, Helen Pallett-Wiesel, Nicolas Lavoie and Shania Fock Ka Bao. We would like to thank members of the PREVENT-AD and UK Biobank Research Groups for providing access to the open datasets. We would also like to extend our gratitude to all the co-investigators, consultants, and staff members from the PREVENT-AD Research Group for contributing to the work presented in this manuscript. A complete list of the PREVENT-AD Research Group can be found at: https://preventad.loris.ca/acknowledgements/acknowledgements.php?date=[2024-03-26].

Paul Aisen^8,9, Elena Anthal^4,5, Melissa Appleby^4,5, Pierre Bellec^4,8,10, Fatiha Benbouhoud⁵, Véronique Bohbot^4,5,6, Jason Brandt¹¹, John C. S. Breitner^4,5,6, Céline Brunelle^4,5, Mallar Chakravarty^4,5,6, Laksanun Cheewakriengkrai^4,5,12, Louis Collins^4,8,13, Doris Couture⁵, Suzanne Craft¹⁴, Mahsa Dadar^6,8, Leslie-Ann Daoust⁴, Samir Das^8,15, Marina Dauar-Tedeschi^4,5,12, Doris Dea^4,5, Nicole Desrochers^4,5, Sylvie Dubuc⁵, Guerda Duclair^4,5, Marianne Dufour^4,5, Mark Eisenberg¹⁶, Rana El-Khoury^4,5, Pierre Etienne^4,5,6, Alan Evans^4,5,13, Anne-Marie Faubert⁵, Fabiola Ferdinand^4,5, Vladimir Fonov^13,15, David Fontaine^4,5, Renaud Francoeur^4,5, Joanne Frenette^4,5, Guylaine Gagné^4,5, Serge Gauthier^3,4,5,6,12, Maiya R. Geddes^3,4,8, Valérie Gervais^4,5, Renuka Giles^4,5, Julie Gonneaud^4,5, Renee Gordon^4,5, Claudia Greco^4,5, Rick Hoge^3,4,13, Louise Hudon⁴, Yasser Ituria-Medina^3,4,13,15, Justin Kat^4,5,13, Christina Kazazian^4,5, Stephanie Kligman^4,5, Penelope Kostopoulos^13,17, Anne Labonté^4,5, Marie-Elyse Lafaille-Magnan^4,5,18, Tanya Lee^4,5, Jeannie-Marie Leoutsakos¹¹, Illana Leppert^4,5,8, Cécile Madjar^4,5,8, Laura Mahar^4,5, Jean-Robert Maltais^4,5,12, Axel Mathieu⁵, Sulantha Mathotaarachchi^5,12, Ginette Mayrand^4,5, Melissa McSweeney¹⁸, Pierre-François Meyer^4,5,18, Diane Michaud⁵, Justin Miron^4,5,18, John C. Morris¹⁹, Gerhard Multhaup²⁰, Lisa-Marie Münter²⁰, Vasavan Nair^5,6,12, Jamie Near^5,6, Holly Newbold-Fox⁵, Nathalie Nilsson^4,5,18, Véronique Pagé⁵, Tharick A. Pascoal^3,4,5,12, Mirela Petkova^4,5,8, Cynthia Picard^4,5, Alexa Pichet Binette^4,5, Galina Pogossova^4,5, Judes Poirier^4,5,6, Natasha Rajah^4,5,17, Jordana Remz⁵, Pierre Rioux⁸, Pedro Rosa-Neto^4,5,6,12, Mark A. Sager²¹, Eunice Farah Saint-Fort⁵, Mélissa Savard^4,5, Jean-Paul Soucy^8,15, Reisa A. Sperling²², Nathan Spreng⁸, Frederic St-Onge^4,5,18, Christine Tardif^4,5, Louise Théroux^4,5, Ronald G. Thomas²³, Paule-Joanne Toussaint^8,15, Jennifer Tremblay-Mercier^4,5, Miranda Tuwaig^4,5, Etienne Vachon-Presseau^5,24, Isabelle Vallée^4,5, Vinod Venugopalan^4,5, Sylvia Villeneuve^4,5,6, Simon Ducharme⁴,⁵,⁶, Karen Wan⁴,⁵ & Seqian Wang⁵,¹²,¹⁸ 

⁴STOP-AD Centre, Centre for Studies on Prevention of Alzheimer’s Disease, Montreal, QC, Canada. ⁵Douglas Mental Health University Institute Research Centre, Affiliated with, McGill University, Montreal, QC, Canada. ⁶Department of Psychiatry, McGill University, Montreal, QC, Canada. ⁸Montreal Neurological Institute, Montreal, QC, Canada. ⁹Alzheimer’s Therapeutic Research Institute at University of Southern California, San Diego, CA, USA. ¹⁰Université de Montréal, Montreal, QC, Canada. ¹¹John Hopkins University, Baltimore, MD, USA. ¹²Research Centre for Studies in Aging, McGill University, Montreal, QC, Canada. ¹³Department of Biomedical Engineering, McGill University, Montreal, QC, Canada. ¹⁴Wake Forest School of Medicine, Winston-Salem, NC, USA. ¹⁵McConnell Brain Imaging Center, McGill University, Montreal, QC, Canada. ¹⁶School of Population and Global Health, McGill University, Montreal, QC, Canada. ¹⁷Department of Psychology, McGill University, Montreal, QC, Canada. ¹⁸Neuroscience Department, McGill University, Montreal, QC, Canada. ¹⁹Washington University School of Medicine in St-Louis, St. Louis, MO, USA. ²⁰Department of Pharmacology, McGill University, Montreal, QC, Canada. ²¹Wisconsin Alzheimer’s Institute, UW School of Medicine and Public Health, Milwaukee, WI, USA. ²²Center for Alzheimer’s Research and Treatment Harvard Medical School, Boston, MA, USA. ²³School of Medicine, University California, San Diego, La Jolla, CA, USA. ²⁴Northwestern University, Chicago, IL, USA.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00142.