The data set used in this study was obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database¹. The ADNI was launched in 2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies, and non-profit organizations, as a $60 million 5-year public–private partnership. The primary goal of ADNI is to test whether serial MRI, positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of MCI and early AD. Determination of sensitive and specific markers of very early AD progression is intended to aid researchers and clinicians in developing new treatments and monitor their effectiveness, as well as lessen the time and cost of clinical trials.

The Institutional Review Board approved the study of all the participating institutions. Informed written consent was obtained from all participants at each site. Based on the ADNI three databases, we identified 378 right-handed CN subjects and 182 MCI patients with complete neuropsychological evaluations, structural MRI, and rsfMRI scans. We downloaded data from the ADNI database before March 2021. The criteria for MCI in ADNI were: (1) subjective memory complaints, either self-reported, reported by a study partner, or reported by a clinician; (2) objective memory loss defined as scoring below education adjusted cutoff score on delayed recall of the Wechsler Memory Scale-Logical Memory (WMS-LM) test; (3) a global Clinical Dementia Rating (CDR) score of 0.5; (4) a Mini-Mental State Examination (MMSE) score of equal to, or higher than, 24 out of 30; (5) general cognitive and functional performance sufficiently preserved such that a diagnosis of dementia could not be made by the site physician at the time of screening; and (6) no signs of depression (Geriatric Depression Scale, GDS < 6). Criteria for CN were defined as an MMSE score equal to or higher than 24 out of 30, a CDR score of 0, having no report of any cognition complaint, and no signs of depression. Then, we classified MCI and CN subjects into non-smoking and smoking subgroups separately based on self-reported smoking history. Non-smoking subgroup included participants who reported that they never smoked cigarettes during their lifetime, and smoking subgroup included participants who reported any history of smoking (Supplementary Table 1). After screening, 49 participants (including 29 non-smoking CN, 1 smoking CN, and 19 non-smoking MCI) were excluded for excessive head motion (details later). Finally, 304 non-smoking CN, 44 smoking CN, 130 non-smoking MCI, and 33 smoking MCI subjects entered subsequent analyses.

Demographics, including age, sex, education level, hypertension, diabetes mellitus, and hypercholesterolemia, were assessed. Hypertension was defined as systolic blood pressure >140 mm Hg, diastolic blood pressure >90 mm Hg, medical history of hypertension, or treatment with antihypertensive medication. Diabetes mellitus was defined as random blood glucose ≥11 mmol/dL, medical history of diabetes mellitus, or treatment with glucose-lowering medication. Hypercholesterolemia was defined as random blood cholesterol ≥11 mmol/L, medical history of hypercholesterolemia, or treatment with lipid-lowering medication. Neuropsychological tests in different cognitive domains, such as memory (WMS-LM, immediate and delayed recall), attention [trail-making test, Part A (TMT-A)], execution [trail-making test, Part B (TMT-B)], and language [semantic verbal fluency (SVF)], were included. The demographics and neuropsychological performance of all subjects are listed in Supplementary Table 2.

All subjects were scanned using a 3.0-Tesla Philips MRI scanner. The structural images were acquired using a 3D magnetization prepared rapid gradient echo (MPRAGE) T1-weighted sequence with the following parameters: repetition time (TR) = 2,300 ms; echo time (TE) = 2.98 ms; inversion time (TI) = 900 ms; 170 sagittal slices; within plane FOV = 256 mm × 240 mm; voxel size = 1 mm × 1 mm × 1 mm; flip angle = 9°; bandwidth = 240 Hz/pix. The rsfMRI images were obtained using an echo-planar imaging sequence with the following parameters: 197 time points; TR = 3,000 ms; TE = 30 ms; slice thickness = 3.39 mm; spatial resolution = 3.39 mm × 3.39 mm × 3.39 mm; flip angle = 90°; and matrix = 64 × 64.

We preprocessed rsfMRI data using the Data Processing & Analysis for Brain Imaging (DPABI²) (Yan et al., 2016) with Statistical Parametric Mapping software (SPM12³) on the MATLAB platform (MathWorks, Natick, MA, United States). The first 10 time points of rsfMRI data were discarded due to the instability of the initial MRI signal and the adaptation of subjects to the scanning noise. The remaining 187 images were corrected for timing differences between each slice and head motion (six-parameter rigid body). Subjects with more than 3.0 mm maximum displacement in any of the x, y, or z directions or 3.0° of any angular motion were discarded (including 29 non-smoking CN, 1 smoking CN, and 19 non-smoking MCI). Subsequently, rsfMRI images were spatially normalized to the standard EPI template and resampled into 3 mm × 3 mm × 3 mm. The rsfMRI images were spatially smoothed with a Gaussian kernel of 6 mm × 6 mm × 6 mm full width at half maximum. Finally, linear trends and temporally filter (0.01 Hz < f < 0.08 Hz) were performed. Nuisance covariate regression was performed to minimize physiological noise using the Friston-24 head motion parameters (6 head motion parameters, 6 head motion parameters from the previous time point, and the 12 corresponding squared items), as well as white matter (WM) signal and corticospinal fluid (CSF) signal. In addition, the framewise displacement (FD) Jenkinson value of each subject was calculated to correct for the head motion artifacts.

The T1-weighted images were preprocessed and analyzed using the Computational Anatomy Toolbox (CAT12⁴) and SPM12. The images were bias-corrected, tissue-classified [gray matter (GM), WM, and CSF], and registered using linear (12 parameter affine) and non-linear transformations (warping) within the CAT12 default preprocessing pipeline.

The seed NBM was identified using a probabilistic anatomical map from the SPM Anatomy Toolbox, which was microscopically delineated, 3D reconstructed, and warped to the reference space of the Montreal Neurological Institute (MNI) brain from 10 postmortem human brains (Zaborszky et al., 2008). The NBM, which is the largest of the cholinergic cell clusters constituting the basal forebrain, provides the primary source of cholinergic innervation to the cortex (Gratwicke et al., 2013; Liu et al., 2015). Dynamic brain connectome (DynamicBC) analysis toolbox⁵ (Liao et al., 2014) was used to create individual subject seed-to-voxel connectivity maps. First, each mask was resampled to the dimension of our normalized functional image with 3 × 3 × 3 voxel size for seed-based resting-state functional connectivity (rsFC) analyses. Next, the rsFC maps were generated by calculating the Pearson’s correlation between the time course of the NBM and whole brain areas. Finally, the resulting rsFC maps were transformed to Z maps using Fisher’s Z transformation.

Propensity score matching (PSM) implanted in Statistical Product and Service Solutions (SPSS version 26) was performed to balance the differences in demographic features between non-smoking and smoking subgroups in CN and MCI and to reduce the bias due to confounding factors. A propensity score was estimated using multiple logistic regression analysis based on the following covariates: age, sex, and education level. A 1:2 matching was used to pair participants with smoking CN and smoking MCI, respectively. Significant testing and standardized difference (d) were applied to assess the balance of covariates before and after PSM. Through PSM, 86 non-smoking CN, 44 smoking CN, 62 non-smoking MCI, and 32 smoking MCI were selected from the initial population. The distribution of propensity scores was relatively balanced after matching (Supplementary Figure 1). There was no significant difference of the key covariates between the smoking and non-smoking groups in both CN and MCI, and the d values were acceptable (Supplementary Table 3), suggesting balanced demographic features in CN and MCI subgroups.

We rated the total MRI burden of small vessel disease (SVD) on an ordinal scale from 0 to 4, by counting the presence of each of the four MRI features of SVD, including white matter hyperintensity (WMH), lacunes, cerebral microbleeds (CMBs), and perivascular spaces (PVS) (Staals et al., 2014). The presence of WMH was defined as either confluent deep WMH (Fazekas score 2 or 3) or irregular periventricular WMH extending into the deep WM (Fazekas score 3) on T2 fluid-attenuated inversion recovery (FLAIR) (1 point if present). The presence of lacunes and CMBs was defined as the presence of one or more lacunes on T2 FLAIR (1 point if present) or any CMB on T2^∗-weighted gradient echo images (1 point if present). The presence of PVS was counted if there were moderate to severe (grades 2–4) PVS in the basal ganglia on T1-weighted axial images (1 point if present) (Zhu et al., 2010). The detailed SVD scores are listed in Supplementary Table 4.

The statistical analyses of demographics and neuropsychological data were performed using IBM SPSS 26.0 statistical software. We performed a group-level analysis using a one-way ANOVA for continuous variables. If group-level test results were significant, post hoc pairwise comparisons were performed (Bonferroni’s multiple comparison correction in parametric tests and Dunn’s multiple comparison tests in non-parametric tests). Binary data, such as sex, hypertension, diabetes mellitus, and hypercholesterolemia, were compared between groups using a chi-square test.

The statistical analyses of the rsFC of the NBM were performed using the DPABI toolbox (Yan et al., 2016). Specifically, we performed a 2 × 2 mixed effect analysis to explore the main effects of condition (non-smoking vs. smoking) and diagnosis (CN vs. MCI), as well as the interaction effects of condition × diagnosis. Age, sex, education level, and head motion (FD value) were used as covariates. To control the effect of cortical atrophy on the functional analysis, normalized modulated (with the volumetric information encoded) GM maps were used as covariate images. The threshold was set to the voxel level at p < 0.005 and the cluster level at p < 0.05 after Gaussian random field (GRF) correction. The interaction effects between condition and diagnosis before PSM are listed in Supplementary Table 5 and Figure 2. In addition, we also evaluated the interaction effects after controlling for SVD scores (Supplementary Figure 3). To further understand how condition and diagnosis interacted on the rsFC of the NBM, we extracted the mean rsFC values from the interaction regions and further performed post hoc pairwise comparisons (p < 0.05, Bonferroni’s correction). Finally, partial correlation analysis was performed to investigate the correlation between the mean rsFC values of interaction regions and neuropsychological scores with age, sex, and education level as covariates (p < 0.05).

After PSM, a total of 86 non-smoking CN, 44 smoking CN, 62 non-smoking MCI, and 32 smoking MCI were included. The demographics and neuropsychological performance are summarized in Table 1. There were no significant differences between subgroups in age, sex, education level, hypertension, diabetes mellitus, and hypercholesterolemia (p > 0.05). The MCI subgroups (non-smoking and smoking) showed significantly poor neuropsychological performance on memory, attention, execution, and language compared with CN subgroups (p < 0.05). The smoking subgroups (CN and MCI) showed no significant differences in neuropsychological performance compared with corresponding non-smoking subgroups (p > 0.05).

The interaction effects of condition (non-smoking vs. smoking) × diagnosis (CN vs. MCI) on rsFC of the NBM after PSM are displayed in Figure 1 and Table 2. Based on the seed of left NBM, the regions with interaction effects were primarily located in the bilateral prefrontal cortex (PFC) and bilateral supplementary motor area (SMA). Based on the seed of the right NBM, the interaction region was located in the right precuneus/middle occipital gyrus (MOG) (p < 0.005). The main effects of the condition or diagnosis were not detected.

The post hoc test revealed the group differences of rsFC values of interaction regions in four subgroups after PSM (Figure 2). Specifically, the smoking CN showed decreased rsFC between left NBM and PFC, and increased rsFC between left NBM and SMA compared with non-smoking CN and smoking MCI. The smoking MCI showed reduced rsFC between right NBM and precuneus/MOG compared with non-smoking MCI.

We studied the relationship between rsFC of the NBM and neuropsychological performance in different domains (Table 3). We only found that rsFC of the NBM to the SMA showed a significant negative correlation with memory (WMS-LM immediate recall) in the smoking CN group (r = −0.321, p = 0.041; Figure 3). After Bonferroni’s correction (p < 0.05/15) for multiple comparisons, there was no significant correlation between rsFC of the NBM and neuropsychological scores.

Several limitations need to be addressed for this study. First, due to the limited sample size of the ADNI database, the sample sizes of the smoking CN and smoking MCI groups are relatively small compared with the non-smoking groups. In addition, the biomarkers were not included in this analysis because only a few smoking MCI patients had positive biomarker evidence (i.e., Aβ+). Future studies with a larger sample size and biomarker evidence are needed to verify our work. Second, this cross-sectional study is lacking clinical follow-up to make any possible inference between smoking and AD. Thus, longitudinal studies are needed to determine whether rsFC of the NBM alterations in smokers is related to disease progression. Third, our results of the correlation analysis did not survive after multiple comparison corrections. However, this is an explorative study, and our results could partly reflect the effects of cigarette smoking on different cognitions. Finally, smoking history in the ADNI database is defined by subjective self-report from the medical record, including former and current smokers. Many participants lack detailed records such as the number, duration, and status (former or current) of smoking. Future studies may further explore the effects of different smoking degrees or statuses on cognitive functions in MCI because these factors could affect cerebral functional activity.