Impaired Parahippocampal Gyrus–Orbitofrontal Cortex Circuit Associated with Visuospatial Memory Deficit as a Potential Biomarker and Interventional Approach for Alzheimer Disease

Abstract

The parahippocampal gyrus–orbitofrontal cortex (PHG–OFC) circuit in humans is homologous to the postrhinal cortex (POR)–ventral lateral orbitofrontal cortex (vlOFC) circuit in rodents. Both are associated with visuospatial malfunctions in Alzheimer’s disease (AD). However, the underlying mechanisms remain to be elucidated. In this study, we explored the relationship between an impaired POR–vlOFC circuit and visuospatial memory deficits through retrograde tracing and in vivo local field potential recordings in 5XFAD mice, and investigated alterations of the PHG–OFC circuit by multi-domain magnetic resonance imaging (MRI) in patients on the AD spectrum. We demonstrated that an impaired glutamatergic POR–vlOFC circuit resulted in deficient visuospatial memory in 5XFAD mice. Moreover, MRI measurements of the PHG–OFC circuit had an accuracy of 77.33% for the classification of amnestic mild cognitive impairment converters versus non-converters. Thus, the PHG–OFC circuit explains the neuroanatomical basis of visuospatial memory deficits in AD, thereby providing a potential predictor for AD progression and a promising interventional approach for AD.

Electronic supplementary material

The online version of this article (10.1007/s12264-020-00498-3) contains supplementary material, which is available to authorized users.

Background

Alzheimer’s disease (AD) is a highly prevalent neurodegenerative disease that is characterized by a progressive decline in memory and other cognitive functions [1]. Amnestic mild cognitive impairment (aMCI) is a well-characterized high-risk factor for AD, and is defined as an early stage in the AD spectrum [2]. A core clinical feature of AD is episodic memory deficits, which can predict disease progression [3]. Recently, a deficit in visuospatial function has also been considered as a potential early predictive indicator for AD [4–6]. In 2018, the A/T/N (amyloid, tau, and neurodegeneration) biomarker classification scheme was proposed; this is based on the nature of the pathology, thereby emphasizing the importance of cerebrospinal fluid (CSF) and molecular positron emission tomography (PET) in the diagnosis of AD [7–9]. However, in China, obtaining CSF is challenging, and molecular PET is rarely implemented, particularly during the screening of patients with MCI. Thus, identifying novel methods for the early diagnosis of AD is of utmost importance.

The postrhinal cortex (POR) in rodents is homologous to the parahippocampal gyrus (PHG) in humans [10]. In previous studies, advances in visuospatial processing have shown that an array of networks contributes to visuospatial memory [11–13]. Recently, a new neural framework for primates proposed by Mishkin and coworkers indicated that the mesial temporal lobe (including the hippocampus, PHG, and entorhinal cortex) and the frontal lobes are critical regions for visuospatial memory [14]. Lesions in the PHG in humans [15, 16] or in the POR in rats [17] cause severe visuospatial learning and memory deficits, with profound effects on memory retrieval, but only mild effects on memory encoding [17]. We previously demonstrated that inhibition of the glutamatergic input from the POR to the ventral lateral orbitofrontal cortex (vlOFC) specifically impairs visuospatial memory retrieval in wild-type mice [18], suggesting that impairment of the POR–vlOFC circuit contributes to visuospatial memory deficits. However, changes in the anatomy and function of this circuit in a mouse model of AD have not yet been reported.

In humans, the uncinate fasciculus (UF) is a hook-shaped long-range association pathway that connects the anterior temporal lobes (including the anterior PHG) to the orbitofrontal cortex (OFC) [19]. Disruption of the UF as measured by magnetic resonance imaging (MRI) has been reported in patients with autism spectrum disorder [20], MCI [21], temporal lobe epilepsy [22], Parkinson disease [23], and schizophrenia [19]. In addition, impaired diffuse tensor imaging (DTI) indexes of the UF have been reported in patients with subjective cognitive decline [24], aMCI [25] and AD [26], and have been correlated with general cognitive impairment [27, 28], visuospatial malfunction [29], and episodic memory deficits [25, 30, 31]. Moreover, improvement in visuospatial working memory during childhood is significantly associated with changes in DTI indexes in the right UF [32], suggesting that the UF contributes to visuospatial functions. During visuospatial tasks, PHG activation is enhanced in participants with normal cognition [21, 33, 34] and in prodromal AD patients [35], but the activation declines in the PHG [36] and OFC [37] of AD patients. However, it remains unclear how the anatomy and function of the PHG–OFC circuit are changed in the AD spectrum.

In the present study, we investigated the role of the POR–vlOFC circuit during visuospatial memory through exploring neuron-specific alterations in 5XFAD mice. Furthermore, we assessed the changes in structural and functional connectivity (FC) of the PHG–OFC circuit in the AD spectrum. Together, our findings further illustrate the neuroanatomical basis of visuospatial memory and identify novel predictors of AD conversion.

Materials and Methods

Mice

Adult (4–6 months) male 5XFAD mice and their wild-type (WT) littermates were purchased from the Jackson Laboratory (Bar Harbor, ME, USA) and were housed at the Animal Center of the Medical School of Southeast University and Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences. Mice were maintained at 22 °C–24 °C under a 12-h light/dark cycle with food and water provided ad libitum.

Behavioral Tests

To assess visuospatial memory, we used the novel object place recognition (NOPRT) and Y-maze tests. Moreover, the open field test (OFT) and elevated plus maze (EPM) tests were used to assess anxiety-like behaviors. All behavioral tests were carried out as previously described [32, 38–40] with minor modifications (Supplementary Material). Behavior was monitored and analyzed using ANY-maze software (Stoelting, Co., Wood Dale, IL, USA) by two experienced investigators who were blinded to the grouping of the experimental mice.

Immunofluorescence

To determine and compare differences in activation of the POR and vlOFC regions during the NOPRT in WT and 5XFAD mice, the expression profiles of c-Fos, CamKIIα, and GAD67 proteins in the two regions were quantified by immunofluorescence analysis as previously described [39, 41] (Supplementary Material).

Retrograde Tracing

To analyze structural integrity, 200 nL of cholera toxin B (CTB)-Alexa Fluor 488 was injected unilaterally into the vlOFC (AP, 2.30 mm; ML, −1.20 mm; DV, 2.60 mm). Four weeks later, the mice were deeply anaesthetized and perfused transcardially, and the expression of CTB and CamKIIα was assessed in the POR and vlOFC regions using immunofluorescence analysis (Fig. 2A). Brain sections (40 μm) were imaged on an Olympus VS120 digital slice scanner (Tokyo, Japan).

Fig. 2.

CamKIIα+ POR inputs to the vlOFC region. A Retrograde tracing strategy using CTB and its experimental timeline. B–C Representative images showing retrograde labeling of CamKIIα+ POR input to the vlOFC region. Blue: DAPI; green: CTB - Alexa 488; red: CamKIIα; scale bars: 500 μm and 50 μm (B), 200 μm and 50 μm (C); arrows indicate co-localization; dashed outline indicate the vlOFC (B) and POR (C) regions; three mice each from 5XFAD and WT groups; n =27 sections of the POR and 48 sections of the vlOFC in each group. D Compared to WT mice, 5XFAD mice have significantly fewer CTB neurons in the POR [ 5XFAD (n = 3) 9.33 ± 2.96 vs. WT (n = 3) 37.33 ± 3.33; independent sample t-tests, t = 6.278, P = 0.003]. E 5XFAD mice have a significantly lower ratio of CamKIIα-positive CTB neurons to total CTB neurons in the POR compared to WT mice [5XFAD (n = 3) 26.28% ± 13.54% vs WT (n = 3) 58.28% ± 6.35%; independent sample t-tests, t = 3.208, P = 0.033]. F Compared to WT mice, 5XFAD mice have a slightly lower ratio of the CamKIIα-positive CTB neurons in the POR than in the vlOFC [5XFAD (n = 3) 7.85% ± 2.27% vs WT (n = 3) 32.90% ± 6.89%; non-parametric Mann-Whitney U test, U = 9, P = 0.05]. Data are presented as the mean ± SEM. WT, wild type; POR, postrhinal cortex; vlOFC, ventral lateral orbitofrontal cortex; CTB, cholera toxin B.

In vivo Optogenetic Manipulation and Microelectrode Array Implantation

For in vivo optogenetic manipulations, 250 nL of virus was injected into the POR. Three weeks later, an optical fiber was implanted into the ipsilateral vlOFC for the stimulation of terminals. All mice received the same intensity of blue laser stimulation.

For in vivo electrophysiological recordings, a 16-channel microelectrode array containing 2 hand-crafted tetrodes was implanted into WT and 5XFAD mice, separately in the POR and vlOFC. To quantify the FC between the POR and vlOFC during visuospatial memory, the coherence of local field potentials (LFPs) between the two regions during exploration in the NOPRT was assessed (Fig. 3A). Detailed procedures are provided in Supplementary Materials.

Fig. 3.

Alterations of LFP functional connectivity in 5XFAD and WT mice. A Locations of recording electrodes in the POR and vlOFC regions. B, D Average power spectra of LFP recordings in the POR and vlOFC during exploration of a displaced object (red), a familiar object (blue), and a wall (green) (n_WT = 8, n_5XFAD = 6). C No significant differences are found between the two groups in the power of the beta (13–40 Hz), theta (4–7 Hz), and delta (<4 Hz) bands of the POR region (n_WT = 8, n_5XFAD = 6; two-way ANOVA, LSD post-hoc test, P >0.05 for all groups). E 5XFAD mice show reduced power of LFP signals in the beta band of the vlOFC region during exploration of the displaced object compared to WT mice (n_WT = 8, n_5XFAD = 6; two-way ANOVA, LSD post-hoc test, P = 0.013). F, G Coherence spectral analysis shows a significantly stronger functional connectivity between the two regions at 15 Hz in 5XFAD mice than in WT mice (n_WT = 8, n_5XFAD = 6, non-parametric Mann-Whitney U test, U = 6, P = 0.02). Data are presented as the mean ± SEM. WT, wild type; ANOVA, analysis of covariance; POR, postrhinal cortex; vlOFC, ventral lateral orbitofrontal cortex; LFP, local field potential.

Recruitment of Participants and Neuroimaging Data Analysis

Patients with aMCI and healthy controls (HCs) were recruited to establish a registry at the Affiliated Zhongda Hospital of Southeast University and Xuanwu Hospital Affiliated to Capital Medical University. Written informed consent was given by all participants prior to the start of the study. Based on the outcome of an average of 3 years of follow-up (i.e., whether patients with aMCI converted to AD), 37 aMCI converters (aMCI-Cs), 38 aMCI non-converters (aMCI-NCs) were identified. Also, 45 HCs, who were all right-handed Han Chinese, were recruited through routine community health screening and advertisements. All participants underwent a standardized clinical interview and a neuropsychological assessment battery. The inclusion and exclusion criteria are described in Supplementary Materials.

All participants [29/37 aMCI-Cs, 31/38 aMCI-NCs, and 45/45 HCs (the number in the follow-up/that at baseline)] underwent baseline and follow-up multi-modal MRI examination (high-resolution T1-weighted and resting-state functional imaging). Notably, none of the selected aMCI patients showed excessive motion artifacts (≥3 mm translational or ≥3° rotational movements) during MRI scans or incomplete image coverage. Details of imaging acquisition and preprocessing are presented in Supplementary Materials.

To investigate changes in the volume of grey matter (GM) and brain neuronal activity [indicated by the amplitude of low-frequency fluctuations (ALFFs)] in the bilateral PHG and OFC, regions of interest (ROIs) were defined based on the Brainnetome atlas (Fig. 5A), a fine-gained, cross-validated atlas that correlates brain anatomy with psychological and cognitive functions [42]. Accordingly, for all patients, the mean GM volume and ALFF values for each OFC and PHG subregion were extracted. Furthermore, to measure inter-regional resting-state functional connectivity (RSFC), Pearson correlation coefficients between pairs of ROIs were calculated, generating a 24 × 24 correlation matrix for each patient. These coefficients were subjected to Fisher’s z-transformation to yield variants from the normal distribution. The RSFC was investigated between all OFC and PHG sub-regions, and the pattern of connectivity between 144 (12 × 12) pairs of ROIs was explored.

Fig. 5.

Group differences in the OFC/PHG gray matter volume and OFC–PHG functional connectivity at baseline. A Bilateral OFC and PHG sub-regions according to the Brainnetome atlas. B Volumes of OFC and PHG sub-regions are significantly smaller in aMCI-C patients than in aMCI-NC patients and HCs at baseline (ANCOVAs; FDR-corrected, P < 0.05). C Compared to HCs, aMCI-C and aMCI-NC patients show increased ALFF values in the PHG ROIs (ANCOVAs; FDR-corrected, P < 0.05). aMCI-NC patients have higher ALFF values in the PHG ROI (i.e., right TH) than aMCI-C patients. D Compared to aMCI-NC patients and HCs, aMCI-C patients show decreased OFC-PHG functional connectivity (ANCOVAs; FDR-corrected, P < 0.05). Compared to HCs, aMCI-NC patients show higher OFC-PHG functional connectivity. E Merged SVM classifiers for GM volume, ALFF, and connectivity distinguish aMCI-C from aMCI-NC patients with an accuracy of 77.33% (sensitivity 72.97%, specificity 81.58%) and an area under the curve of 0.8378. Data are presented as the mean ± SEM. aMCI, amnestic mild cognitive impairment; aMCI-C, aMCI-converters; aMCI-NC, aMCI-non-converters; HC, healthy control; OFC, orbitofrontal cortex; PHG, para-hippocampal gyrus; GM, grey matter; ANCOVAs, analyses of covariance; AUC, area under the curve; ALFF, low-frequency fluctuations; ROIs, the regions of interest; FDR, false discovery rate; SVM, support vector machine; GM, grey matter. For the abbreviations of the OFC and PHG sub-regions, see Supplemental Table S1.

Statistical Analysis

Data are presented as the mean ± standard error of the mean (SEM). For animal behavioral tests and immunofluorescence analysis, the Shapiro-Wilkes test was used to assess the normality of the distributions. Normally distributed data were compared using paired and unpaired independent t-tests (two groups), and one-way or two-way analysis of covariance (ANCOVA) (multiple groups). In addition, non-normally distributed data were analyzed using the Mann-Whitney U test. ANCOVA was used to investigate aMCI-C-related changes of the mean GM volume and ALFF values in the OFC and PHG ROIs, and to determine the pattern of connectivity between the 144 pairs of ROIs at baseline, with age, gender, and years of education as covariate variables. Comparisons were adjusted using the false discovery rate (FDR) at a significance threshold of P < 0.05. The support vector machine (SVM) [43] was used to classify GM volume, ALFF, and connectivity, which were combined to distinguish aMCI-Cs, aMCI-NCs, and HCs. The merged classifier was evaluated through 10-fold cross-validations.

For each patient, annual change (AC) estimates were calculated using the mean GM and ALFF values from the OFC and PHG ROIs, as well as the connectivity between the pairs of ROIs from baseline and follow-up scans. The following equation was used:$AC=\frac{MRImeasure\left(t1\right)-MRImeasure\left(t0\right)}{t1-t0}$, where MRI measure (t0) and MRI measure (t1) represent MRI measurements (GM volume, ALFF, and RSFC) at the time of entry and at an average of 3 years of follow-up, and t1 − t0 represents the individual delay between evaluations. ANOVA assessments were further used to evaluate group differences in AC estimates with an FDR-corrected P < 0.05 considered as a significant difference. Furthermore, using relevance vector regression as a multivariate analytical approach, we further examined the relationship between aMCI-C-related changes in AC estimates (GM and RSFC) and the AC of mini-mental state examination (MMSE) scores.

Results

Retrieval Deficit of Visuospatial Memory in 5XFAD Mice is Associated with Anatomical and Functional Disruption of the POR–vlOFC Circuit

The visuospatial memory of mice was initially assessed by NOPRT (Fig. 1A). Both 5XFAD and WT mice at 6 months of age showed no preference for either object in the study phase (Fig. 1B), however, 5XFAD mice showed a prominent reduction in the discrimination index (D2) (Fig. 1C) (for the measurement in 4-month-old mice see Fig. S1). In addition, 6-month-old 5XFAD mice showed decreased spontaneous alternations in the Y-maze test as compared to WT mice (Fig. S2A). Again, no changes were found in the time and distance spent in the center of the OFT, or the time and number of entries into the closed arms in the EPM between 5XFAD and WT mice (Fig. S2B–C).

Fig. 1.

CamKIIα activation in the POR and vlOFC regions of 6-month-old 5XFAD and WT mice during visuospatial memory retrieval. A Schematic of the novel object place recognition test. D1-D5, Day1-Day5; red, blue and orange dots in the boxes represent different objects that the mice explored. B There are no differences in T(f) and T(n) between WT and 5XFAD mice. C Compared to WT mice, 5XFAD mice show a reduced D2 during the recognition phase [5XFAD (n = 7) 14.41% ± 3.8% vs WT (n = 8) 33.20% ± 5.29%; independent sample t-tests, t = –2.807, P = 0.015]. D–G Representative images and statistics showing no significant differences between WT and 5XFAD mice in CamKIIα neurons, c-Fos+ cells, and c-Fos-activated CamKIIα neurons in the POR region (n_5XFAD = 4, n_WT = 5; independent sample t-tests, t = 0.237, P = 0.820; t = 2.099, P = 0.074; t = 1.039, P = 0.334). H–K Representative images and statistics showing that, compared to WT mice, 5XFAD mice show decreased c-Fos+ cells and c-Fos-activated CamKIIα neurons in the vlOFC region (n_5XFAD = 4, n_WT = 5; independent sample t-tests, t = 2.928, P = 0.022; t = 2.585, P = 0.042). However, no significant differences between the two groups were shown in the numbers of CamKIIα-positive neurons (independent sample t-tests, t = 1.023, P = 0.346). Data are presented as the mean ± SEM. Blue, DAPI; green, CamKIIα; red, c-Fos; scale bars, 50 μm; n = 64 sections of vlOFC and n = 36 sections of POR from four 5XFAD mice and n = 80 sections of vlOFC and n = 45 sections of POR from five WT mice; arrows indicate co-localization. WT, wild type; POR, postrhinal cortex; vlOFC, ventral lateral orbitofrontal cortex; T(f), time for the familiar object; T(n), time for the novel object; D2, discrimination index.

After NOPRT, the differences in activation of the POR and vlOFC regions were quantified by c-Fos, CamKIIα, GAD67, c-Fos/CamKIIα, and c-Fos/GAD67 protein staining in both WT and 5XFAD mice. When compared to WT mice, 5XFAD mice showed a significant decrease in the numbers of both c-Fos- and c-Fos/CamKIIα-positive neurons in the vlOFC region (Fig. 1H, J, K,), but not in the POR region (Fig. 1D, F, G). However, no significant differences between the two groups were shown in the numbers of CamKIIα-positive neurons in the POR and vlOFC regions (Fig. 1E, I). Notably, no significant differences between the two groups in the numbers of both GAD67- and c-Fos/GAD67-positive neurons were found in either region (Fig. S3).

Glutamatergic neurons in the vlOFC were deactivated in visuospatial memory-impaired 5XFAD mice as measured by the integrity of the anatomical connection in the glutamatergic POR–vlOFC circuit through retrograde anatomical tracing using CTB labeling (Fig. 2A). Compared to WT mice, 5XFAD mice had significantly fewer CTB neurons (Fig. 2C, D), as well as a lower ratio of CamKIIα-positive CTB neurons to total CTB neurons in the POR (Fig. 2C, E). Furthermore, compared to WT mice, 5XFAD mice had a slightly lower ratio of CamKIIα-positive CTB neurons in the POR to that in the vlOFC (Fig. 2B, C, F).

Subsequently, we determined the integrity of the functional connection between the two regions in both WT and 5XFAD mice by LFP recording during NOPRT. Our data showed that, compared to WT mice, 5XFAD mice showed reduced power of LFP signals in the beta band of the vlOFC region during exploration of the displaced object (Fig. 3B, D, E). In contrast, no significant differences were found between groups in the power of the beta, theta, and delta bands of the POR region (Fig. 3B–D). Importantly, coherence spectral analysis showed a significant increase in FC between the two regions at 15 Hz in 5XFAD mice during exploration of the displaced object as compared to WT mice (Fig. 3F, G).

Opto-stimulation of CamKIIα:: POR-vlOFC Terminals in 5XFAD Mice Rescues the Visuospatial Memory Deficit

After determining that the impaired POR-vlOFC glutamatergic pathway in 5XFAD mice is involved in the visuospatial memory deficit, we further explored whether optical activation of the CamKIIα:: POR-vlOFC circuit could rescue this deficit in 5XFAD mice. After surgery (Fig. 4A), similar times and distances in the center of the OFT were recorded among the four experimental groups (Fig. S4), and no significant difference was found for locomotor activity between groups with and without optogenetic manipulation among the four groups (Fig. S5). Furthermore, mice did not show a preference for either object during the NOPRT study phase (Fig. 4D). However, during the recognition phase, significant effects on D2 were found for “group” [F (1, 34) = 6.242, P = 0.018] or for the interaction of “group” and “optical stimulation” [F (1, 34) = 17.50, P < 0.001]. Moreover, post-hoc analysis showed that during the recognition phase, the D2 decreased in the 5XFAD-CamKIIα-mCherry group that was exposed to opto-stimulation compared to the WT-CamKIIα-mCherry, WT-CamKIIα-ChR2-mCherry and 5XFAD-CamKIIα-ChR2-mCherry groups (Fig. 4E; P = 0.041, P = 0.021, P < 0.001). Furthermore, no significant difference in D2 was found between the 5XFAD-CamKIIα-ChR2-mCherry and WT-CamKIIα-mCherry or WT-CamKIIα-ChR2-mCherry groups (P = 0.162, P = 0.389).

Fig. 4.

Rescue of visuospatial memory deficit by opto-stimulation of CamKIIα:: POR-vlOFC terminals in 5XFAD mice. A Optogenetic strategy and experimental timeline for unilateral POR injections of AAV-CamKII-ChR2-mcherry or AAV-CamKIIα-mcherry virus and unilateral vlOFC optical activation. B Representative immunofluorescence image showing the position of the fiber track in the vlOFC (blue, DAPI; solid lines, fiber track; scale bar, 200 μm). C Representative images showing selective targeting of ChR2-mCherry to CamKIIα+ neurons in the POR and co-labeling of CamKIIα with mCherry-labeled POR terminals from POR CamKIIα+ neurons in the vlOFC (blue, DAPI; green, CamKIIα; red, mCherry; scale bars, 200 μm and 20 μm; arrowheads indicate co-localization). D There is no significant difference between T(f) and T(n) in each group (n_{WT-CamKIIα-mCherry} = 9, n_{WT-CamKIIα-ChR2-mCherry} = 11, n_{5XFAD-CamKIIα-mCherry} = 9, n_{5XFAD-CamKIIα-ChR2-mCherry} = 9; paired t-test, P >0.05 for all groups). E D2 decreases in the 5XFAD-CamKIIα-mCherry group (n = 9) during the recognition phase with opto-stimulation compared to the WT-CamKIIα-mCherry (n = 9), WT-CamKIIα-ChR2-mCherry (n = 11), and 5XFAD-CamKIIα-ChR2-mCherry groups (two-way ANOVA by post-hoc Bonferroni multiple comparison tests, *P < 0.05, ***P < 0.001). Data are presented as the mean ± SEM. WT, wild type; ANOVA, analysis of covariance; POR, postrhinal cortex; vlOFC, ventral lateral orbitofrontal cortex.

Disruption of the OFC–PHG Circuit in aMCI-C Patients

As previously noted, the OFC–PHG circuit in humans is homologous to the POR–vlOFC circuit in rodents, so we further explored whether the OFC–PHG circuit is damaged in aMCI patients and whether this impairment can predict the conversion of aMCI to AD. At baseline, the analysis of MRI scans (GM volume and RSFC) showed significant “group” effects (FDR-corrected P < 0.05), and post-hoc comparisons showed that the volumes of the OFC and PHG ROIs were significantly smaller in aMCI-C patients than in aMCI-NC patients and HCs (Fig. 5B). No significant differences were found between aMCI-NC patients and HCs. Furthermore, compared to HCs, both aMCI-C and aMCI-NC patients showed significantly increased ALFF values in the PHG ROIs (Fig. 5C) with slightly higher ALFF values in the PHG ROI (right TH) in aMCI-NC than in aMCI-C patients. In addition, aMCI-C patients showed decreased OFC-PHG connectivity compared to the HC and aMCI-NC groups (Fig. 5D) with slightly increased OFC–PHG connectivity in aMCI-NC patients compared to HCs. Finally, SVM classifiers for GM volume, ALFF, and connectivity were merged to distinguish between aMCI-C and aMCI-NC patients. Importantly, the merged classifier had an accuracy of 77.33% (sensitivity 72.97%, specificity 81.58%) and an area under the curve of 0.8378 (Fig. 5E).

Our data indicated that the AC estimates (GM volume and RSFC) showed significant “group” effects (FDR-corrected, P < 0.05). The most important finding involved the annual change in GM atrophy, which was statistically significant in the OFC and PHG of only aMCI-C patients (Fig. 6A). Furthermore, a significant annual loss of OFC-PHG connectivity only occurred in the aMCI-C group (Fig. 6B). Interestingly, aMCI-NC patients conversely showed slightly increased ACs of OFC–PHG connectivity when compared to HCs. In addition, for aMCI-C-related ACs in GM volume and connectivity, the use of relevance vector regression permitted quantitative predictions of ACs in MMSE scores with statistically significant accuracy (r = 0.671, P < 0.001, Fig. 6C).

Fig. 6.

Comparison of changes in annual GM volumes and RSFC in the aMCI-C, aMCI-NC, and HC groups. ANOVA analysis shows significant “group” effects on their AC estimates (FDR-corrected P < 0.05). A, B Significant annual GM atrophy in the OFC and PHG (A), and annual OFC-PHG connectivity loss (B) (A14m.L – A35/36r.L, A11m.L – A35/36r.R) of aMCI-C patients. C For aMCI-C-related ACs in GM volume and connectivity, the use of relevance vector regression permits prediction of ACs in MMSE scores with statistically significant accuracy (r = 0.671, P < 0.001). Data are presented as the mean ± SEM. aMCI, amnestic mild cognitive impairment; aMCI-C, aMCI-converters; aMCI-NC, aMCI-non-converters; OFC, orbitofrontal cortex; PHG, parahippocampal gyrus; GM, grey matter; MMSE, mini-mental state examination; RSFC, resting-state functional connectivity; ANOVA, analyses of variance; RVR, relevance vector regression; AC, annual change. For abbreviations of the OFC and PHG sub-regions, see Table S1.

Discussion

In the present study, we revealed visuospatial memory deficits early in 6-month-old 5XFAD mice with selectively impaired integrity of the glutamatergic POR–vlOFC pathway. Furthermore, opto-stimulating terminals in that specific neural circuit significantly improved the impaired visuospatial memory in 5XFAD mice. In addition, the progressively damaged PHG-OFC circuit measured by MRI distinguished aMCI-C from aMCI-NC patients and HCs, and AC estimates further confirmed the capacity of prediction for the progression of aMCI and AD conversion. Taken together, disruption of the glutamatergic POR–vlOFC circuit indicated an anatomical basis of visuospatial memory deficits, and the aggravating impairment of the PHG–OFC pathway in the aMCI group predicted the disease progression and AD conversion.

In our previous study [44, 45], we demonstrated that 1-month-old 5XFAD mice show memory impairment as measured by the Morris water maze test. This test is widely used to evaluate cognitive functions, including visuospatial memory, episodic memory, and working memory [46, 47]. Here, we mainly focused on visuospatial memory as measured by NOPRT, thereby taking advantage of the natural preference of rodents for new over familiar objects. However, visuospatial memory deficits appear in 5XFAD mice at the age of 6 months, when neuronal loss starts to occur in multiple regions of the brain [48]. In this study, a combination of NOPRT, cell-type-specific retrograde tracing, in vivo LFP recording, and optogenetic manipulation were used to determine the possible underlying mechanism of the glutamatergic POR–vlOFC circuit in visuospatial memory impairment in 6-month-old 5XFAD mice.

To our knowledge, this is the first example of disruption of the glutamatergic POR–vlOFC circuit in 5XFAD mice, which was directly related to deficient visuospatial memory retrieval. In both humans and rats, the OFC makes tight connections with the visual cortex [49] and temporal lobe structures [50–52], while the vlOFC is an efferent target of the POR in rats [53]. In a previous study, we have demonstrated that in wild-type mice, the vlOFC is one of the efferent targets of glutamatergic neurons in the POR [39], suggesting an anatomical foundation for the POR–vlOFC circuit in visuospatial memory. In addition, neurons in the POR of rats have been shown to encode representations that link locations and objects through in vivo recordings from postrhinal neurons and LFPs during visual discrimination tasks [54]. Neurons in the OFC are not only responsive to specific visual stimuli [55], but are also required for cognitive maps and task space in both humans [56] and rodents [57]. The data indicate that both the POR and OFC respond to visual stimuli and encode spatial information. Rats with injuries to the POR [17] and vlOFC [58] show impaired orientation to visual stimuli, while inhibition of the glutamatergic POR–vlOFC pathway reduces the retrieval deficit of visuospatial memory [39], indicating that lesions of the POR–vlOFC circuit result in visuospatial memory deficits. Using in vivo LFP recording, we found abnormal beta oscillation in the POR–vlOFC circuit in 5XFAD mice during NOPRT. These findings suggest that malfunction of the above circuit may contribute to visuospatial memory impairment in 5XFAD mice. Previous studies have shown that beta oscillations occur during learning, cognition, and cortical reorganization [59, 60]. A study by Quinn et al. showed that beta oscillations in the basal forebrain of rats are learning-dependent [59]. During an associative learning task, it has been reported that peak amplitudes are significantly greater around the novel object [59]. In a spatial exploration and memory-guided behavior task, beta oscillations in the hippocampus increase during memory retrieval [61]. In this study, 5XFAD mice showed decreased beta oscillation in the vlOFC during exploration of the displaced object, indicating that reduced beta oscillations contribute to visuospatial memory deficits. Moreover, increased FC between the POR and vlOFC regions in 5XFAD mice may compensate for visuospatial memory deficits. Taken together, in this study we demonstrated that anatomical and functional disruptions in the POR–vlOFC circuit in 5XFAD mice are associated with visuospatial memory retrieval deficits, which are significantly improved upon optogenetic activation.

Neuroimaging studies have shown that the PHG is a region with greater cortical thickness in individuals with normal cognition and is vulnerable to atrophy in AD, independent of amyloid load and the apolipoprotein E genotype [62]. In addition, reduced GM and FC of the PHG has been reported in aMCI [25, 63–65] and AD patients [26, 64, 66, 67]. Hypoperfusion of the PHG [68] and OFC [69] in AD patients is correlated with the MMSE scores, which indicates that they may be vulnerable hubs to maintain general cognition in AD patients, particularly for memory. In the current study, we focused on the PHG–OFC circuit in aMCI patients. In the AD spectrum, both decreased PHG–OFC activity and the ACs of PHG–OFC connectivity were only found in aMCI-C patients during follow-up, whereas an increased capacity of the neural circuit was found in aMCI-NC patients, suggesting that hyperactivation of the PHG–OFC circuit in aMCI-NC patients plays a role in memory preservation. Importantly, no significant differences were found in clinical cognitive performance between the two aMCI groups, which further indicated that the damaged neural circuit may be an early biomarker for the prediction of AD conversion. In addition, Mitolo and colleagues showed that MCI-C patients have significantly reduced GM in the PHG compared to MCI-NC patients [70]. Mattsson and coworkers demonstrated that ¹⁸F-AV-1451 in the PHG has 93% diagnostic accuracy for AD (prodromal or dementia) [71]. Initial metabolic impairment of AD converters has been shown in the OFC during the early stages of AD [72]. In this study, we demonstrated that anatomical and functional impairment of the PHG–OFC circuit had an accuracy of 77.33% to distinguish aMCI-C from aMCI-NC patients at baseline, and that aMCI-C-related ACs in GM volume and connectivity accurately predicted the ACs in MMSE scores. Taken together, our findings consistently showed that the impaired PHG–OFC circuit could serve as a potential biomarker for AD conversion.

This study had several limitations that should be addressed. The classical pathological biomarkers for AD and amyloid imaging should be applied for precise aMCI diagnosis. A task-functional MRI should be designed to track visuospatial memory-related changes in neural activity, and its complex underlying circuits should be elucidated through c-Fos mapping in the whole brain of mice [73]. Different subtypes of neurons in the POR–vlOFC circuit resulting in visuospatial memory impairment in 5XFAD mice should be clarified. Furthermore, the specific molecular regulatory mechanisms of the POR–vlOFC circuit require elucidation for application with PET in the clinic.

In conclusion, our findings indicated that impairment of the glutamatergic POR–vlOFC circuit contributes to visuospatial memory deficits in AD mice, which were significantly rescued by opto-stimulation of this specific neural circuit. Together, these findings suggested a potential target for intervention. The impaired PHG–OFC circuit as measured by MRI in aMCI patients is highlighted as a promising predictor of AD conversion.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Compliance with Ethical Standards

Conflict of interest

The authors declare that they have no conflict of interest.