Integration of Multilocus Genetic Risk into the Default Mode Network Longitudinal Trajectory during the Alzheimer’s Disease Process

Participants

The Affiliated ZhongDa Hospital of Southeast University Research Ethics Committee approved the investigation, and each participant provided written informed consent. All procedures performed on subjects were in accordance with the Helsinki Declaration of 1975. At baseline, 87 MCIs and 131 matched HCs were enrolled using media advertisements and community health screening events. For the same group of participants, on average, 35-month (35±5.1, range from 24.4 to 44.3 months) longitudinal studies were performed, and 57 MCIs and 64 HCs returned (we paused the follow-ups of HCs after comparative numbers of MCIs and HCs returned). The remaining MCI subjects (n = 30) were lost due to the development of neurological or psychiatric diseases, moves to other cities, death, and unwillingness to participate. Four HCs developed into MCI and seven MCIs reverted to normal cognition; these subjects were not included in the analysis. Moreover, each participant had complete MRI images with acceptable motion artifacts at baseline; however, four HCs and seven MCIs were excluded for excessive motion artifacts and incomplete image scans at follow-up. Eleven of the remaining 44 MCIs developed into AD. Finally, 56 HCs, 32 nc-MCIs, and 11 c-MCIs entered the longitudinal analysis. Each subject underwent the same neuropsychological tests at baseline and follow-up (details of the neuropsychological battery are provided in Table 2).

Inclusion and exclusion criteria

All participants met the following criteria: (1) Han Chinese; (2) age between 54 and 80 years; (3) ≥8-year education; (4) right-handed; and (5) adequate visual and auditory abilities that enabled cognitive testing (hearing prostheses and glasses allowed).

MCI subjects were included according to the criteria of Petersen et al. [43] and others’ recommendations [44, 45] as follows: (1) subjective memory impairment corroborated by the subject and an informant; (2) objective impairment of memory with an Auditory Verbal Learning Test-20-minute delayed recall (AVLT-20-min DR) score less than or equal to 1.5 standard deviations of age- and education-adjusted norms; (3) a Mini-Mental State Examination (MMSE) score ≥24 or Mattis Dementia Rating Scale-2 (MDRS-2) score ≥120; (4) a Clinical Dementia Rating of 0.5, with at least a 0.5 in the memory domain; (5) no or minimal impairment in daily activities: Activities of Daily Living scores less than 26; and (6) absence of dementia or insufficient dementia to meet the Diagnostic and Statistical Manual of Mental Disorders-IV (DSM-IV) and the National Institute on Aging-Alzheimer’s Association workgroups on the diagnostic guidelines for Alzheimer’s disease (NIA-AA).

Furthermore, the HCs were required to have a clinical dementia rating of 0, MMSE scores ≥26, MDRS-2 scores >120, and neuropsychological battery performances in the normal range. The DSM-IV and NIA-AA criteria were used for the clinical diagnosis of AD.

The exclusion criteria were as follows: (1) a history of neurological or psychiatric diseases and major medical illness; (2) contraindications for MRI; (3) gross structural abnormalities of the brain; or (4) major white matter hyperintensity, cerebral infarction, or other lesions of the brain.

SNP selection and construction of genetic risk scores

We selected the top SNPs and calculated the GRSs similar to previous investigations [38–42]. Only the SNPs with solid evidence to support their associations with AD in Eastern Asians (i.e., Chinese, Japanese, and Korean) were selected, and the results from previous GWAS are our principal references. In detail, no large-sampled GWAS have been performed in Eastern Asians; thus, we selected the SNPs that have been demonstrated by GWAS performed in other populations (i.e., Caucasians). Furthermore, for the SNPs that reached the threshold of p < 5×10^–8 in previous GWAS, only some SNPs have been replicated to modify AD risk in Eastern Asians in subsequent case/control or meta-analyses, which were included in the present analyses. We also included rs6691117 despite the absence of evidence from GWAS. A relatively large-sampled case/control study in Chinese demonstrated that this missense variation was associated with AD. Based on a recent study demonstrating rs6691117 caused atrophy of the middle temporal lobe (MTL) [46], we assumed that rs6691117 was likely an AD-related variation. In total, 11 polymorphisms were selected; the details are listed in Table 1.

To estimate the joint effects of these polymorphisms, GRS1s were calculated by multiplying the number (0/1/2) of risk (sometimes protective) alleles by weight and subsequently obtaining the sum across the 11 variants. The weight was the natural log of the odds ratio (OR) obtained from the original case/control studies or meta-analysis (listed in Table 1). Missing data were imputed with the averaged numbers of specific alleles. To exclude the possibility that DMN changes were a result of the large effect size of APOE ɛ4, GRS2s were also calculated without considering this genotype [41].

Functional connectivity analysis

Detailed information regarding the MRI data acquisition and image preprocessing are provided in the Supplementary Material. The T2-weighted images were obtained to rule out subjects with major white matter changes, cerebral infarction, or other lesions. The high-resolution, T1-weighted axial images covering the whole brain were acquired to perform the voxel-wise-based gray matter volume correction and rule out the subjects with gross structural abnormalities. Regarding the preprocessed functional data, a spherical region of interest (ROI; radius = 6 mm) within the PCC was centered at the Montreal Neurological Institute (MNI) coordinate –2, –45, 34 (22, 33), followed by coregistration to the functional data. Then, the averaged time course of the PCC region was correlated with the time courses of all brain voxels using the Pearson cross-correlation. Fisher’s Z-transformations were applied to improve the normality of the correlation coefficients. Additionally, the mean time series of global, white matter, and cerebrospinal fluid signals and head motion parameters were introduced as covariates of no interest [47, 48]. The processes were performed using Data Processing Assistant for Resting-State fMRI (DPARSF) (http://www.restfmri.net), which is based on Statistical Parametric Mapping (SPM8) (http://www.fil.ion.ucl.ac.uk/spm) and Resting-State fMRI Data Analysis Toolkit (REST, http://www.restfmri.net).

Statistical analysis

Demographic and neuropsychological data

For all the 11 variants, no significant deviations from Hardye-Weinberg equilibrium analysis were detected in MCIs or HCs (all p > 0.05, data not shown). Further, the GRSs did not significantly vary among different groups (p > 0.05). Regarding the neuropsychological data, the MCIs (n = 80) showed deficits in all cognitive domains compared with the HCs at baseline (n = 127, all p < 0.05). At follow-up, the c-MCIs (n = 11) showed worse performances in all cognitive domains than HCs (n = 56) and nc-MCIs (n = 32) (all p < 0.05), whereas significant differences were primarily identified in episodic memory and visuospatial and executive functions between nc-MCIs and HCs (all p < 0.05, details in Table 2).

Altered DMN patterns between MCIs and HCs at baseline

The DMN patterns for each group at both baseline (Supplementary Fig. 1) and follow-up (Supplementary Fig. 2) were obtained using PCC seed-based FC analysis. At baseline, two-sample t-tests revealed five regions with significantly decreased FCs in MCIs compared with HCs. These regions were primarily within the RPHG/Hip and bilateral middle and superior temporal gyrus (BMTG and BSTG). Increased FCs in MCIs were detected in three regions within the LmPFC, LIPL and left middle frontal gyrus (LMFG) (details in Fig. 1 and Table 3).

Genetic basis for the altered DMN pattern at baseline

Cognitive significance of the altered DMN pattern

The 8 previously described regions were extracted as ROIs. In HCs, no significant correlations were observed between the FC within the 8 ROIs and the cognitive performance (all p > 0.05).

For the MCIs, at the threshold of corrected p < 0.05, the FC within the LIPL was negatively correlated with episodic memory (AVLT-20 min DR scores, rho = –0.33, raw p = 0.003), and similar association was detected between the FC within LMFG and the general cognition (MDRS-2 scores, rho = –0.43, raw p = 0.001). However, an opposite correlation was detected with FCs within the RHip/PHG and AVLT-20 min DR scores (rho = 0.44, raw p = 0.00007) and general cognition (MMSE scores, rho = 0.44, raw p = 0.00006). Furthermore, the increased FCs within the RMTG were significantly associated with increased scores of AVLT-20 min DR and MMSE (rho = 0.33 and 0.39, raw p = 0.003 and 0.0003, respectively, details provided in Fig. 2B).

Longitudinal DMN changes in HCs, nc-MCIs, and c-MCI

Genetic factors that drove the differential longitudinal trajectory

Cognitive relevance of the differential DMN changes

As shown in Fig. 5B, for HCs, the FC changes within LIPL were significantly related to the declines of MMSE (rho = –0.47, raw p < 0.001) and MDRS-2 (rho = –0.40, raw p = 0.003). Furthermore, for nc-MCIs, less declines of FC within LIPL were associated with better maintenances of general cognition (rho = 0.41, p = 0.02). No associations were identified for c-MCIs.

Trajectory of Longitudinal DMN Changes in Aging and AD Development

As expected [28, 50], the decreased FCs of DMN were detected in MCIs primarily within the temporal cortex. Greater decreases in FC were associated with greater impairments in episodic memory and general cognition. Similar correlations have been reported [28, 49]. Here, these results were replicated with a substantially larger sample; the positive correlations detected in MCIs but not HCs specifically supported the hypothesis that DMN disconnections correlated with AD severity. Furthermore, interactions between Aβ, FC, and hypometabolism have been identified within the PCC/Pcu and temporal cortex [51, 52], which suggests the reduced FCs manifest as a consequence of molecular AD-related pathology. The spatial overlap between hypometabolism and FC reduction within DMN reflects synaptic dysfunction and functional disconnection in these hubs [52] and represents key mechanisms that underlie cognitive damage in AD. Increased FC was detected within the prefrontal-parietal cortices in MCIs, which was also consistent with previous investigations [53, 54]. Moreover, the negative correlations between FC and cognition supported the compensatory nature of the increased recruitment of prefrontal-parietal regions [55, 56]. The DMN has been reported to exhibit age-related increases in activity within prefrontal-parietal regions [51], which thereby indicates the reorganization of brain sources to cope with brain injury in other hubs [57]. For MCIs, an accelerated compensatory process was identified at baseline, which suggests more serious injuries to temporal cortex caused by AD-related pathology. Moreover, this idea may also be supported by the findings that the c-MCIs trended to exhibit more disease-related alterations than the nc-MCIs, i.e., more disconnections within temporal cortex and more compensations within frontal regions prior to the conversion to AD (for details, see Supplementary Fig. 3; the neuropsychological data for these subjects are provided in Supplementary Table 3).

Differential longitudinal changes were primarily detected within regions in which compensation occurred at baseline among the HCs, nc-MCIs, and c-MCIs, and the compensation within prefrontal-parietal regions collapsed for c-MCIs. The behavioral significances of the DMN trajectory were also highlighted. During the follow-up, the greater FC increases within LIPL for HCs may suggest more severe injury for other hubs, such as the MTG, thus more cognitive declines were detected. For MCIs, the maintenance of the compensation was necessary to uphold relatively intact cognition; thus, the less the FC decreased, the better the cognition was preserved over time. Therefore, it is not difficult to interpret findings with the opposite cognitive significances of DMN trajectory for HCs and nc-MCIs, and the breakdown of compensation within prefrontal-parietal brains may represent one mechanism that underlies AD development.

We also concluded that compensation in AD first appeared in PCC [58] and MTL and was subsequently present in frontal-parietal cortex. This finding is consistent with the posterior-anterior brain activity shift [59] and scaffolding theory of aging and cognition [60], which suggested aging is associated with improved recruitment of the anterior and frontal-parietal brain. This theory also fits the order of neuropathological marker spread in AD brains [61] and the brain abnormalities revealed by structural MRI and PET [62]. Therefore, aging and AD-related pathology represent impetuses for the accelerated compensation. During follow-up, compensation occurred in response to normal aging and attempted to preserve in nc-MCIs. In the c-MCIs, the ability to provide effective compensation diminished and the patients showed severe cognitive deficits. Investigation of the genetic factors associated with the unfavored trajectory will facilitate the identification of AD pathophysiology.

Genetic factors that cause DMN deterioration

An additional new finding included the partial identification of the SNPs associated with DMN deterioration. Using task-dependent fMRIs, Rao et al. found that the FC was increased to compensate for accelerated AD-related pathology, most likely including Aβ- and tau-induced neurotoxicity and neuroinflammation in healthy participants who carried APOE ɛ4; the FC subsequently declined, which reflects a diminished compensation associated with cognitive impairments [63]. Our results provided similar insights to explain the functional activities of ɛ4, which restrained the compensation of parietal lobes. Neural imaging also expanded our understanding of the functional activities of other SNPs. For example, associations have been identified between rs7561528 [17] and rs6691117 [46] and atrophy of the temporal lobe. Our results suggested that these SNPs also exacerbate the brain functional network. Moreover, rs11136000 was related to task-induced neural hyperactivation and longitudinal increases in the cerebral blood flow in the Hip and PCC during normal aging [64, 65]. We found that this SNP promoted DMN disturbances and attenuated functional compensation within the LIPL during the early stage of AD. To conclude, with one exception (rs3865444) these risk alleles exacerbated brain disconnection and diminished DMN compensation over the entire AD spectrum.

Paradoxically, the protective rs3865444^C predicted FC decreases. However, it should be noted that rs3865444^C represents a risk variation in other races [15]. Moreover, its functional activities implicated in AD have been partially identified. Caucasians who carry rs3865444^C have weakened abilities to remove Aβ, deteriorate neuroinflammation [66], and have smaller intracranial volumes [18]. Taken together, the inconsistencies may suggest that rs3865444 influences the AD risk via other mechanisms in Chinese individuals. More likely, these discrepancies may imply a shortcoming of the present investigation (i.e., the OR for each SNP should be replicated in larger samples; details are provided in the limitation paragraph).

An essential issue of GWAS is that most detected SNPs have small predictive values. This issue is the reason that no loci survived the Bonferroni correction. The GRS combining the effect sizes of multiple loci has been identified to better predict the risk for individuals of developing complex diseases, including type 2 diabetes [40], coronary heart disease [38], stroke [42], and multiple sclerosis [39]. For AD, the MCIs with increased GRS exhibited increasing disease deterioration [41]. Previously, the GRS have been shown to be associated with brain atrophy [17, 18]. Our present data suggested that the combined actions of the multiple loci led to functional dysfunction within the MTL, and it may also suppress the functional compensation within the IPL. As previously discussed, DMN disintegrations within the MTL were associated with cognitive impairment in MCIs, and compensatory breakdown within the IPL may be imaging indicators for the c-MCIs. Therefore, these important hubs represent the target regions of the gene-regulatory changes in the AD brain network. However, it was difficult to interpret that GRS1 was not related to DMN changes in the converters. Furthermore, in the single SNP-based analysis, APOE ɛ4 failed to modify DMN longitudinal changes in MCIs. Our group has previously suggested that young carriers of ɛ4 may exhibit hyper-connectivity between the PCC and ACC compared with ɛ3ɛ3 and ɛ2ɛ3 carriers during the normal aging process. For middle-aged subjects (55 to 70 years), the FC for ɛ4 carriers falls between ɛ3ɛ3 and ɛ2ɛ3; after the age of 70 years, ɛ4 carriers exhibited the lowest FC [33]. These findings suggested the APOE genotype interacted with age to modify the DMN pattern. Accordingly, we assumed that ɛ4 also had a differential influence on DMN during the various AD stages (i.e., ɛ4 primarily functions in the very early stage of AD and plays a role via uniting the influences of other variations in the later stages). During the AD phase, ɛ4 deteriorates the disease process via othermechanisms.

There are several limitations. First, the selections and weightings of the SNPs were based on previous studies in East Asians with smaller sample sizes than studies in Caucasians. Further large-sampled case/control studies are necessary to better elucidate the genetic basis for AD in various races. Second, we employed MCIs based on clinical evaluations and not the assessment of in vivo AD-related biomarkers. Our laboratory has previously reported MTL atrophy in these MCIs [67], which contributes additional biomarker information to support the diagnosis. However, substantial difficulties exist in performing lumbar punctures and PET scans in China. Third, seven MCI subjects reverted to normal cognition during follow-up, which was inconsistent with the progressive course of AD. These subjects were excluded in the analysis, and we could not determine the reasons for the unexpected cognitive improvement. Moreover, we excluded the subjects with major white matter hyperintensity, cerebral infarction, or other lesions of the brain. However, to better illustrate the genetic basis for DMN changes in AD process, the influences of the vascular changes and white matter hyperintensity should be calculated. Finally, and most importantly, the sample sizes of MCIs in the follow-up analysis were small, particularly the c-MCIs group, which exerted a strong influence on the statistical power. We expect further investigations with larger cohorts and better designs to validate our results and better reveal the gene-braininteractions.

In summary, the DMN initiates compensatory scaffolding during normal aging. MCIs exhibit accelerated processes, and precipitous declines in compensation will occur once AD develops. The trajectory is attributed to the AD-related loci identified by previous GWAS. The combined effects of multi-SNPs predicted the longitudinal trajectory of the DMN. Finally, the GRS may be used as a clinical biomarker to track AD development.