Alzheimer’s disease (AD) is the most frequent cause of dementia (Prince et al., 2013). Clinically, AD develops gradually and presents with progressive decline in multiple cognitive domains, particularly affecting episodic memory (Small et al., 1997; Elias et al., 2000). As the Chinese population is aging, more than seven million Chinese people live with AD currently, and the costs are predicted to reach US $507.49 billion in 2030 (Jia et al., 2018). It is, therefore, important to gain a better understanding of the factors associated with dementia, including brain and genetic markers. Understanding how genetic risk for AD affects the brain might shed light on mechanisms leading to AD cognitive decline later in life.

Studies have shown that genetic risk factors play a critical role in AD development (Gatz et al., 2006). Various susceptibility loci have been identified. For example, early-onset AD is considered to be caused by mutations in the amyloid precursor protein (APP), presenilin-1 (PRES-1), and presenilin-2 (PRES-2) genes (Goate et al., 1991; Levy-Lahad et al., 1995; Rogaev et al., 1995). The apolipoprotein E4 allele (APOE) is the most well-known genetic risk factor linked to late-onset AD (Strittmatter et al., 1993; Gatz et al., 2006; Bettens et al., 2013). But the majority of studies have focused on the risk associated with a single-candidate gene (Strittmatter et al., 1993; Wang et al., 2019). Based on a combination of APOE and 31 other genetic variants, a polygenic hazard score (PHS) has been developed and validated for quantifying AD dementia age of onset (Desikan et al., 2017). The PHS showed substantial improvement over APOE in predicting the age of AD onset and was associated with biomarkers of AD, including MRI-based hippocampal volume loss and cognitive impairment (Desikan et al., 2017; Kauppi et al., 2018).

The genetic effects on AD cognitive decline can be revealed in vivo in the human brain by structural and functional magnetic resonance imaging (MRI) methods. Structural MRI studies have provided evidence of hippocampal atrophy as a key factor in memory impairment in AD (Zarow et al., 2011; Weiner et al., 2013; Harper et al., 2014) and hippocampal volume is considered an index of the degree of cognitive decline (Kilpatrick et al., 1997; Ystad et al., 2009). However, the hippocampal formation is not homogeneous but is composed of several interconnected subregions, namely hippocampal subfields that are believed to have different functions (Duvernoy, 2008; Mueller et al., 2011). Their specialization makes the hippocampal subfields differentially susceptible to AD pathogenic disruptions (Apostolova et al., 2010; Pluta et al., 2012; Zeng et al., 2021). But how the anatomical vulnerability in hippocampal subfields observed in AD is linked to memory impairment remains unknown. This might be partially explained by regional differences in the vulnerability to tau accumulation and neurofibrillary tangles (NFTs), which are core pathological markers of AD (Braak and Braak, 1990; Ball, 1997; Schonheit et al., 2004). Therefore, the volume of hippocampal subfields might be more sensitive imaging biomarkers for understanding memory in AD.

Previous studies suggest that memory impairment in AD may be mediated through APOE-induced changes in the hippocampus (Han and Bondi, 2008; Caselli et al., 2009; Tuminello and Han, 2011; Wang et al., 2019). First, episodic memory is critically dependent on the hippocampus and is impaired in AD. Second, both the decrease of hippocampal volume and memory dysfunction follow a similar genetic trajectory, e.g., both decrease in healthy older people with the possession of the ε4 allele. Finally, in AD, APOE, which asserted different effects on episodic memory, was associated with different compensatory recruitment processes in the hippocampus. A recent study revealed that PHS moderated the relationship between the medial temporal lobe (MTL) volume and episodic memory in older AD adults (Prieto et al., 2020). However, it is unknown whether PHS has similar effects on the association of hippocampal subfield volumes with episodic memory across the AD continuum. Our main goal was to examine whether the PHS modulates the relationship between hippocampal subfield volumes and episodic memory in subjects with normal control (NC), mild cognitive impairment (MCI), and AD from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. We hypothesized that individuals with high polygenic risk for AD and smaller hippocampal subfield volume ratios would show reduced episodic memory performance. And we intended to investigate if hippocampal subfield volume ratios mediated the effect of PHS on memory scores.

The purpose of the current study was to examine neurobiological markers that influence cognitive performance related to AD. In this study, we investigated the effect of the PHS and hippocampal subfield volume ratios on a previously validated episodic memory composite score in the three disease status groups (NC, MCI, and AD). There were three main findings. First, a significant interaction between disease status and PHS status was primarily observed in the right fimbria. Second, patients with high PHS and lower right fimbria/hippocampus volume ratio had lower episodic memory composite scores. After conducting diagnosis-specific analyses, this interaction was only observed in the MCI group. Finally, the right fimbria/hippocampus volume ratio partially mediated the effects of PHS on memory performance in the MCI group. Together, these findings suggest that hippocampal subfield volume ratio and polygenic risk for AD represent important markers of episodic memory performance.

We observed no main effect of PHS status but prominent interaction between PHS and disease status (NC, MCI, and AD) in the right fimbria. This finding, although counter-intuitive, is supported by the previous work suggesting that brain injury in AD results from interactions between disease status and PHS (Prieto et al., 2020). We confirmed that the effect of genetic risks for AD on the hippocampus was driven by disease status (e.g., the clinical symptoms). Longitudinal studies exploring healthy aging and transition to MCI and AD will provide further clarity regarding these genetic effects. Further, our volume ratio analysis revealed a decreased VR in the right fimbria in the AD group, compared to the MCI and NC groups. This is reasonable. The fimbria is a white matter structure that extends from the alveus and eventually forms the fornix. It carries axons that emanate primarily from pyramidal neurons in the CA1 and subiculum (Joseph and Cardozo, 2004). In the previous study, a smaller volume of fimbria showed strong associations with poor cognitive/memory performance controlling for total hippocampal volume (Evans et al., 2018) and was commonly found in patients with MCI and AD. This may be, in part, due to its anatomical connections and location within functional pathways. For instance, the fimbria–fornix (FF) bundle, through which noradrenergic and cholinergic afferents reach the hippocampal formation, profoundly affects memory (Buzsáki et al., 1992; Cassel et al., 1997). Thus, the FF-lesion has been used as a model of age-dependent memory deficits (Mercerón-Martínez et al., 2020). So, its integrity is important in preserving the hippocampus’s key role in memory (Nilsson et al., 1987). A recent QSM study showed that the magnetic susceptibility of the fimbria was greater in patients with AD, implying microstructural changes in the tissue (Au et al., 2021). The authors believed that the susceptibility changes may also account for the downstream reduction in hippocampal/fimbrial volume. Diffusion tensor imaging study also has shown changes in the fimbria which were able to explain some variance in the memory tests (Cahn et al., 2021). Despite these promising findings, little is currently known about the biological underpinnings of a fimbria/hippocampus volume ratio, as previous models of neuropathology have primarily focused on the biological mechanisms involved in hippocampal atrophy.

Alterations in the output of hippocampus information due to alterations in the integrity of the fimbria could explain the patients’ reduced memory performance. This proposal seems to be further reinforced by the observed solid and direct correlations between the right fimbria/hippocampus VR and memory score (Figure 3). Our findings are also supported by previous studies, which have shown the reduced fimbria volume and associated altered memory in rats (Winters and Dunnett, 2004; Addy et al., 2005), asymptomatic adults (Tomaiuolo et al., 2004; Zheng et al., 2018), and patients with multiple sclerosis (González Torre et al., 2017). This observation highlights the importance of fine grain analyses at specific hippocampal subfields to detect correlates of memory deficits. In line with this, the destructive effect of the APOEε4 allele on white matter tracts in the right fornix was revealed in old participants, which was positively associated with memory impairment (Gold et al., 2010; Zhang et al., 2015). These findings propose the effects of fimbria volume and genetic risk for AD on memory preceding the development of dementia.

Results of this study showed the utility of the PHS to assess the current levels of hippocampus subfield atrophy and episodic memory function in old people. It is still not fully known through which mechanisms PHS impact AD risk. Still, hippocampus atrophy level and cognitive function are considered intermediate phenotypes that may mediate genetic effects on AD risk (Desikan et al., 2017; Kauppi et al., 2018). Previous studies using both survival analyses and linear mixed effect models showed an improved prediction of AD progression by combining MRI data with cognitive performance, and genetic risk for AD (McEvoy et al., 2009; Li et al., 2016; Kauppi et al., 2018). The predictive capabilities of PHS may have contributed to our finding of genetics moderating the relationship between the right fimbria/hippocampus VR and episodic memory. Similar findings were observed in a prior study in old adults (Prieto et al., 2020) that the PHS moderated episodic memory through the left hippocampus in the MCI group.

Whereas our moderation analysis results are consistent with those previously found for PHS (Prieto et al., 2020), our meditational analyses reported novel findings. Mediation models in cognitive neurogenetic research provide a useful framework for directional hypotheses, which are generally appropriate for cognitive neurogenetics because effects generally transmit from genes to brain and cognition, rather than vice versa. The strong total, direct effects, and partial indirect effect of the PHS on episodic memory through the right fimbria/hippocampus VR may reflect specific hippocampus subfield atrophy patterns in differential diagnosis and prognosis of AD (Mizutani and Kasahara, 1997; de Flores et al., 2015; Zeng et al., 2021), which in turn alters episodic memory performance. The data indicate a gene–brain–cognition effect pathway (Green et al., 2013), whereby PHS influences episodic memory via downregulation in the right fimbria volume ratio in the MCI group. Investigators reported genetic variants influencing hippocampal subfield volume also modified the risk of developing AD (Hibar et al., 2017). The larger volume of the hippocampus subfields might have translated into behavioral advantages. Moreover, neuronal activity within specific subfields of the hippocampus might serve distinct aspects of memory (Squire, 2004; Mueller et al., 2011). Therefore, subfield-specific effects of PHS might further underline a proposed role for genotype in particular aspects of memory formation. For example, fimbria appears to support episodic detail generation, as well as retrieval of other types of episodic content (St-Laurent et al., 2014, 2016; Memel et al., 2020). Improved volume and better microstructure in fimbria have been associated with better memory performance in MCI subjects (Pereira et al., 2016; Kantarci et al., 2017; Berron et al., 2020). Thus, the observed decreased volume ratio in fimbria could have contributed to poorer memory results with high PHS status.

Interestingly, the moderate and partial mediated effect was most evident among patients with MCI, indicating that the complicated pathological process related to hippocampus atrophy and memory performance changes along the Alzheimer’s disease continuum. Genetic effects known to contribute to AD pathogenesis can modulate adult hippocampal neurogenesis (Berron et al., 2020). Notably, these alterations may occur at the very early stage of AD progression, prior to processes like neuronal loss and amyloid deposition which might lead to memory impairment (Jin et al., 2004; Mu and Gage, 2011). Further, episodic memory was also affected in the earliest stages of AD (Backman et al., 2001; Gold and Budson, 2008). This is consistent with functional imaging studies showing that in the first few years of AD different hippocampal subfields were affected in memory network organization and maintenance (Li et al., 2009; Yassa et al., 2010; Sun et al., 2017). In the AD group, mediation analyses revealed a suppression effect: we found no significant total effect of PHS on memory (effect size = 0.1469, 95%CI: −0.0093 to 0.3031), while the indirect effect of PHS on memory was negative (effect size = −0.0533, 95%CI: −0.1757 to −0.0024) and the direct effect was positive (effect size = 0.2002, 95%CI: 0.0460–0.3545). It is possible that by the time individuals progress to AD, even the combination of genetic risk and hippocampus subfield volumes can no longer explain the variance in episodic memory performance. Because these patients with AD probably suffer from great and multiple brain regions volume loss in structural plasticity and further along a cognitive decline trajectory. A promising avenue for future research on this topic could be the construct of multiple subfields of atrophy patterns as the mediate variable to test the effects of PHS on memory. Considering APOE is strongly associated with both cognition and hippocampal volumes even in very early cases (Haller et al., 2019; Herrmann et al., 2019), we have completed all the analyses including APOE genotype as a covariate. We found that a PHS score without APOE4 contribution keeps the same effect.

In addition, in our executive function analysis for validation, we found that the right fimbria/hippocampus VR had distinct associations with executive function. PHS tended to moderate the relationship between right fimbria volume ratio and executive function, such that patients with high PHS and lower volume ratio might have lower executive function composite scores. Significant total and direct effect but no significant mediation effect was found to test the hypothesis that HP subfield VR contributed to PHS-related executive function effects. The reason why this result was found in the executive function data may be that executive function performance can be indicated by a multitude of imaging cues rather than one specific pattern through the right fimbria/hippocampus volume ratio change. A promising avenue for future research on this topic could be the construct of multiple subfields of atrophy patterns as the mediate variable to test the effects of PHS on executive function. The results highlighted a specific pathway in which PHS affects memory rather than executive function by degrading the right fimbria/hippocampus VR.

Our study has several limitations. First, we used T1-weighted images provided by a 3T MR scanner for hippocampal subfield segmentation. Sometimes it may be challenging to determine the boundaries, especially in several small volume subfields. Future studies using 7T MRI would be more precise to measure subfields. Second, since this was the first study to explain a potential pathway associating PHS-related episodic memory alteration with the hippocampus subfield volume ratio modification by structural MRI in old adults (all the MCIs were amnestic MCIs), studies in different cohorts, or recruiting a larger and more diverse population, may be needed to extend our findings. Third, because of the cross-sectional and retrospective design, we were unable to prove that alterations in hippocampus subfield volume ratios were actually a consequence of the high genetic risk for AD. Lastly, there must be other variables affecting the relationship between the hippocampus and memory performance that need to be further discussed.

The present study demonstrated the mediating effect of hippocampal subfield volume ratio and the moderating effect of polygenetic risk for AD between hippocampus subfield volume ratio and episodic memory performance. Genetic risk and hippocampal subfields should be considered as key variables in models tracking the progression of cognitive decline in healthy and pathological aging.

The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.

The studies involving human participants were reviewed and approved by the Ethics committee of the Second Affiliated Hospital of Zhejiang University, School of Medicine. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

JX designed the study and wrote the first draft of the manuscript. XG analyzed the MRI data. JW, MZ, and XX assisted with the study design and interpretation of findings. All authors have contributed to and approved the final manuscript.