https://orcid.org/0000-0002-3377-5684
Abstract
In early Alzheimer’s disease, which initially presents with progressive loss of short-term memory, neurodegeneration especially affects cholinergic neurons of the basal forebrain. Pharmacotherapy of Alzheimer’s disease therefore often targets the cholinergic system. In contrast, cholinergic pharmacotherapy of mild cognitive impairment is debated since its efficacy to date remains controversial. We here investigated the relationship between cholinergic treatment effects and the integrity of the cholinergic system in mild cognitive impairment due to Alzheimer’s disease. Fourteen patients with high likelihood of mild cognitive impairment due to Alzheimer’s disease and 16 age-matched cognitively normal adults performed an episodic memory task during functional magnetic resonance imaging under three conditions: (i) without pharmacotherapy; (ii) with placebo; and (iii) with a single dose of rivastigmine (3 mg). Cortical acetylcholinesterase activity was measured using PET with the tracer 11C-N-methyl-4-piperidyl acetate (MP4A). Cortical acetylcholinesterase activity was significantly decreased in patients relative to controls, especially in the lateral temporal lobes. Without pharmacotherapy, mild cognitive impairment was associated with less memory-related neural activation in the fusiform gyrus and impaired deactivation in the posterior cingulate cortex, relative to controls. These differences were attenuated under cholinergic stimulation with rivastigmine: patients showed increased neural activation in the right fusiform gyrus but enhanced deactivation of the posterior cingulate cortex under rivastigmine, compared to placebo. Conversely, controls showed reduced activation of the fusiform gyrus and reduced deactivation of the posterior cingulate under rivastigmine, compared to placebo. In both groups, the change in neural activation in response to rivastigmine was negatively associated with local acetylcholinesterase activity. At the behavioural level, an analysis of covariance revealed a significant group × treatment interaction in episodic memory performance when accounting for hippocampal grey matter atrophy and function. Our results indicate that rivastigmine differentially affects memory-related neural activity in patients with mild cognitive impairment and cognitively normal, age-matched adults, depending on acetylcholinesterase activity as a marker for the integrity of the cortical cholinergic system. Furthermore, hippocampal integrity showed an independent association with the response of memory performance to acetylcholinesterase inhibition.
Introduction
In Alzheimer’s disease, which presents with the cardinal symptom of progressive short-term memory loss, neurodegeneration especially affects cholinergic basal forebrain neurons (Mesulam, 2004; Schliebs and Arendt, 2006). Most pharmacological therapies of dementia due to Alzheimer’s disease therefore target the cholinergic system (Lahiri et al., 2004). In the clinical course of Alzheimer’s disease, dementia is preceded by the stage of mild cognitive impairment (MCI). MCI is characterized by measurable neuropsychological deficits, which are not yet severe enough to affect significantly a person’s daily functioning (Albert et al., 2011). Compared to cognitively normal ageing, MCI is associated with signs of altered cholinergic neurotransmission (Herholz, 2008; Haense et al., 2012; Richter et al., 2017). Nevertheless, it remains controversial whether patients with MCI due to Alzheimer’s disease also benefit from cholinergic pharmacotherapy. While some studies suggested that acetylcholinesterase (AChE) inhibitors may lower the rate of progression from MCI to dementia (Petersen et al., 2005; Diniz et al., 2009), meta-analyses did not confirm this finding (Birks and Flicker, 2006; Loy and Schneider, 2006; Raschetti et al., 2007). This negative outcome of meta-analyses is surprising in light of basic research indicating a positive effect of AChE inhibitors on memory-related brain activation in MCI and even cognitively normal older individuals (Goekoop et al., 2006; Grön et al., 2006; Kukolja et al., 2009; Pa et al., 2013).
One possible explanation for this discrepancy is that the response to cholinergic stimulation may depend on the integrity of the cholinergic system in analogy to the Yerkes-Dodson law (Yerkes and Dodson, 1908), stating that task performance first improves and then deteriorates with increasing arousal. It has been observed in cognitively normal ageing that individuals with low baseline memory performance benefit from physostigmine treatment, while those with high baseline performance show no such benefit and, at least in part, even poorer memory (Kukolja et al., 2009). Furthermore, previous studies on the effect of AChE inhibitors in MCI selected patients due to clinical criteria only (Petersen et al., 2005; Grön et al., 2006), resulting in a degree of uncertainty regarding the presence of Alzheimer’s pathology in these patients.
Accordingly, the goal of this study was to investigate whether the effect of cholinergic stimulation on memory-related neural activation differs between cognitively normal older individuals and patients with a high likelihood of MCI due to Alzheimer’s disease. Furthermore, we aimed at linking the response to cholinergic stimulation to the integrity of the cholinergic system. In light of previous research (Haense et al., 2012; Richter et al., 2017), we expected decreased cortical AChE activity in MCI due to Alzheimer’s disease. We hypothesized that a change in neural activation in response to rivastigmine would be related to local intrinsic AChE levels: Specifically, we hypothesized that in MCI with Alzheimer’s pathology, neural activation in task-related areas would normalize under rivastigmine (Rombouts et al., 2002; Kircher et al., 2005; Pa et al., 2013; Risacher et al., 2013), consequently improving memory performance. Conversely, we expected that controls would not benefit from rivastigmine or might even perform worse (Kukolja et al., 2009). In light of ambiguous previous findings regarding the cholinergic system in MCI due to Alzheimer’s disease (Petersen et al., 2005; Birks and Flicker, 2006; Raschetti et al., 2007; Diniz et al., 2009), we expected that the cholinergic deficit is not the only factor determining the behavioural response to AChE inhibition. We hypothesized that an additional predictor of the response to rivastigmine is the integrity of task-related cerebral structures, i.e. in the case of recent episodic memory: the hippocampus. Using functional MRI of an episodic memory task and non-invasive in-vivo imaging of cortical AChE activity with 11C-N-methyl-4-piperidyl acetate (MP4A)-PET, we tested these hypotheses in cognitively normal older subjects and patients with a high likelihood of MCI due to Alzheimer’s disease (Albert et al., 2011).
Materials and methods
Participants
Twenty patients recently diagnosed with MCI in the Memory Clinic Cologne-Jülich (11 male, aged 54 to 80, mean 68.7 ± 6.8 years), without neurological or psychiatric comorbidities, were recruited as part of the MACS study (Memory, Aging and the Cholinergic System, EudraCT No. 2008-008896-32). The study was approved by the ethics committee of the medical faculty of the University of Cologne, as well as local and federal authorities. Written informed consent was obtained from all participants prior to the study.
Five patients had to be excluded from further analysis because of excessive head movement during MRI scanning (three), side-effects (nausea and vomiting) during MRI scanning (two), and one subsequent withdrawal of consent. The remaining 14 patients had positive CSF biomarkers indicative of Alzheimer pathology and signs of neuronal injury, thus meeting the criteria for a high likelihood of MCI due to Alzheimer’s disease described by Albert et al. (2011). Evidence of neuronal injury was operationalized as (i) medial temporal atrophy according to the medial temporal atrophy scale (Scheltens et al., 1992) on high-resolution T1-weighted images; (ii) temporo-parietal and precuneal hypometabolism in fluorodeoxyglucose-PET quantified with 3D-SSP (Minoshima et al., 1995); or (iii) elevated tau protein in CSF (>375 pg/ml). Positive CSF amyloid pathology was defined as amyloid-β1-42 < 550 pg/ml or a tau/amyloid-β1-42 ratio > 0.52 (Duits et al., 2014). CSF was obtained via lumbar puncture (for details on processing see Supplementary material).
As control group, 22 age-matched cognitively normal adults (14 male, aged 53 to 80, mean 66.6 ± 7.1 years) without history of neurological or psychiatric disease and without cholinergic, anticholinergic, or other psychoactive drugs were recruited from the community. Six cognitively normal older subjects were not eligible for further analyses due to an incidental pathological finding (one; severe cerebral atrophy), side-effects of rivastigmine during MRI scanning (two), severe dental metal artefacts interfering with MRI data acquisition (two), and PET acquisition failure due to a technical issue (one).
Thus, the data of 16 controls and 14 MCI patients entered the analysis. All participants underwent a physical and neurological examination by a neurologist and a comprehensive neuropsychological assessment. The latter consisted of the Mini-Mental State Examination (MMSE), Beck’s Depression Inventory V (Schmitt et al., 2006), the Bayer Activities of Daily Living (Hindmarch et al., 1998), the Trail-Making-Test (TMT) A and B (Berres et al., 2000), Brief Test of Attention (Schretlen et al., 1996), the Leistungsprüfungssystem part 4 (logical thinking) (Horn, 1983), the Verbal Learning Memory Test (the German version of the Rey Auditory Learning Test; Helmstaedter et al., 2001), and the Rey-Osterrieth Complex Figure Test (Rey, 1964). MCI was defined as performance >1.5 standard deviations (SD) below the norm in the delayed recall of the VLMT, and a score >24 points in the MMSE, normal scores in tests of attention and logical thinking, as well as normal ratings on the Bayer Activities of Daily Living questionnaire. Furthermore, the patients’ functional independence was corroborated by the history obtained from the family (spouse/children) and behaviour during the course of the study.
Procedure
Clinical, neuropsychological and MRI data were acquired in four sessions over the course of 2 months. In the first MRI session, anatomical and functional scans without pharmacological stimulation were acquired. In the second and third MRI sessions, participants received capsules with either 3 mg of rivastigmine or placebo in a double-blind cross-over design, randomized with respect to order of treatment. This rivastigmine dose has been shown to significantly reduce AChE activity in CSF in healthy adults (Kennedy et al., 1999) and patients with Alzheimer’s disease (Cutler et al., 1998). Furthermore, a single oral application has been demonstrated to modulate neural activation in both populations (Di Lazzaro et al., 2002; Rombouts et al., 2002; Langguth et al., 2007). Medication was administered 90 min prior to magnetic resonance scanning to ensure peak plasma concentrations of rivastigmine (Lefèvre et al., 2008). To account for side-effects of cholinergic stimulation (i.e. nausea, dizziness, headache), subjective well-being was assessed before administration of the trial medication and after magnetic resonance acquisition using a semiquantitative rating scale of complaints (von Zerssen, 1976). Both magnetic resonance sessions with pharmacological stimulation took place within 4 weeks. PET measurement took place within an average of 31 days (SD = 35 days) of the initial MRI session. To exclude effects on AChE activity due to treatment with AChE inhibitors (Bohnen et al., 2005), PET measurement took place before, or at least 7 days after pharmacological stimulation.
During MRI scanning, participants performed a visual item memory task consisting of an encoding and a retrieval session spaced 7 min apart. During each visit, the task was rehearsed outside and inside the magnetic resonance scanner prior to scanning, using a shortened version of the paradigm. For a detailed description of the task, see the Supplementary material.
Behavioural data analysis
Trials with reaction times less <400 ms or >2 SD greater than the mean were considered outliers and classified as invalid trials together with missed trials. The analysis focused on items that were correctly classified during the retrieval session. Performance was assessed using the sensitivity index d′ (for details see Supplementary material). Behavioural data were analysed with repeated measures ANCOVAs with treatment (placebo, rivastigmine) as within subject factor. Parameters entered in ANCOVAs were normally distributed within groups as tested with Shapiro-Wilks tests. If the cholinergic deficit is a central aspect of the group difference and hence essential for the group × treatment interaction, the interaction should not be significant when accounting for the cholinergic deficit. To test this, the cholinergic deficit was operationalized as the k3 averaged across voxels showing a significant group difference in AChE activity (FWE-corrected P < 0.05) (Fig. 1, Table 2 and Supplementary Table 3) and included in the ANCOVA as covariate of no interest.
Cortical cholinergic deficit in MCI due to Alzheimer’s disease. Cortical acetylcholinesterase was significantly decreased in patients with MCI due to Alzheimer’s disease, compared to controls. This is effect was most pronounced in the lateral temporal lobes. P < 0.05, FWE-corrected with cluster free threshold enhancement. R = right. Slice labels indicate MNI coordinates.
To account for mild side-effects of rivastigmine that did not require discontinuation of measurement, changes in subjective well-being were incorporated into the analysis as covariate of no interest. To assess the relevance of hippocampal integrity, average grey matter density within the hippocampus was used to reflect hippocampal structure, and verbal episodic memory performance was used to reflect hippocampal function (Kim, 2011; Wolk et al., 2011). All statistical analyses of non-imaging data were conducted with SPSS (Version 23.0, IBM Corp., Armonk, NY).
MRI and PET acquisition
MP4A was synthesized as previously described (Herholz et al., 2000; Haense et al., 2012), with minor modifications. Between 192 and 586 MBq (mean 469.27 MBq, SD 84.21 MBq) of MP4A were injected intravenously as bolus. The wide range of injected activities is mainly attributable to two control subjects with injected doses of 192.37 MBq and 278.17 MBq. Despite this relatively large range of injected activities, a consequent bias on results is unlikely, since scatter correction was performed during reconstruction of PET images (see below), and results of statistical analyses were essentially unchanged when removing these two participants.
A detailed summary of injected activities (Supplementary Table 1) and a discussion of their variability are provided in the Supplementary material. PET scanning was performed using an ECAT HRRT scanner (CPS Innovations) using the protocol described by Haense et al. (2012). Images were reconstructed using 3D ordered subset expectation maximization, including correction for random coincidences, attenuation and scatter. Moreover, putative effects of scattered activities would have been minimal, since these mainly affect the amplitude of time activity curves, while the shape, which governs the estimation of k3, is barely affected (Häggström et al., 2014). High-resolution T1-weighted images, and fluid attenuated inversion recovery (FLAIR) images were acquired using a 3 T Trio scanner (Siemens). MRI scans were acquired using an echo planar imaging (EPI) sequence The field of view was angulated to parallel the medial tentorium cerebelli to reduce susceptibility artefacts in the medial temporal lobe (Deichmann et al., 2003; Weiskopf et al., 2006). A detailed description of MRI sequences is provided in the Supplementary material.
PET processing
AChE activity was assessed by quantifying the hydrolysis rate of MP4A, i.e. k3, at the voxel level. Images were processed as previously described (Richter et al., 2017); for more details please see Supplementary material. k3-values were extracted from neocortical and hippocampal grey matter, as k3-estimation in other subcortical structures is often unreliable (Herholz et al., 2001, 2004).
Processing of functional MRI data
Preprocessing of functional MRI data was performed with standard procedures using different modules of the FSL-software package (FMRIB’s Software Library, Version 5.0, http://www.fmrib.ox.ac.uk/fsl). A detailed description is provided in the Supplementary material.
Neural activation underlying successful encoding can be operationalized as functional MRI signal that is greater for stimuli that were later remembered than for those later forgotten [the subsequent memory effect (SME)]. To analyse the SME, encoding trials were divided into two types of events: items correctly identified as ‘old’ (OC) during the retrieval phase and invalid trials (INV; i.e. misclassification during retrieval, misses, and outliers). For each event type, a constant epoch regressor consisting of a boxcar function beginning at stimulus onset and lasting 3.5 s (stimulus duration) was defined. Additionally, for each event type, a variable epoch regressor with the identical time course as the respective constant epoch regressor, but variable boxcar width defined by the response time of the respective trial was constructed. The variable epoch regressor was orthogonalized to the constant epoch regressor to account for effects of response time variability, which can significantly influence the amplitude of the haemodynamic response (Grinband et al., 2008; Richter et al., 2013). The resulting four regressors were convolved with a double gamma haemodynamic response function and entered into a first level General Linear Model analysis, together with the covariates generated during the preprocessing steps. Resulting statistical maps were normalized to standard space using the parameters generated in the VBM analysis and entered into a second level analysis using non-parametric testing as implemented in the FSL module randomise (Nichols and Holmes, 2002; Winkler et al., 2014). The non-parametric approach was chosen because of its statistical robustness in small sample sizes and because the FSL module randomise allows the inclusion of voxel-wise covariates, in the present analysis the grey matter density maps generated with VBM.
Second level analyses were constrained to the medial temporal lobe (hippocampus and amygdala), the fusiform gyrus, and the posterior cingulate cortex, based on previous studies of memory-related neural activation in cognitively normal subjects (Garoff et al., 2005; Kim, 2011), patients with Alzheimer’s disease (Golby et al., 2005; Huijbers et al., 2012), and patients with Alzheimer’s disease under cholinergic stimulation (Rombouts et al., 2002). Regions of interest were defined using the Harvard-Oxford cortical and subcortical structural atlases (Desikan et al., 2006).
The SME was computed as the difference between OC and INV. Due to the nature of the non-parametric approach all second level analyses were performed as t-tests. To test for a placebo effect, the data from both groups were pooled and a paired t-test was computed across groups. Since no significant difference in the SME was observed at the region of interest or the whole brain level, data from the baseline and placebo scans were pooled when testing for a group average SME and when computing for group differences in the SME. Furthermore, we tested whether there is a differential response to pharmacological stimulation in the two groups. For this purpose, the parameter estimates from the placebo condition were subtracted from those of the rivastigmine condition and the resulting difference maps were subjected to a two sample t-test to compare between groups. VBM maps were incorporated as covariate of no interest in all second level analyses to correct for effects of grey matter atrophy. To correct for multiple testing, threshold-free cluster enhancement (Smith and Nichols, 2009) and FWE-correction were used. Voxels with P < 0.05 are reported as significant.
Voxel-based morphometry
Structural data were analysed using VBM as implemented in the software FSL-VBM (Good et al., 2001; Smith et al., 2004; Douaud et al., 2007); http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLVBM). Grey matter images modulated for non-linear normalization were smoothed with an isotropic Gaussian kernel of 8.55 mm (full-width at half-maximum) resulting in the same final resolution as the functional images.
Results
Demographics and neuropsychological characteristics
Groups did not differ with respect to age [t(28) = −1.035, P = 0.31], gender distribution [χ2(1) = 0.117, P = 0.732], or depressive symptoms assessed using the BDI [t(28) = −0.261, P = 0.8]. All participants had normal ratings of activities of daily living, but there was a trend for poorer ratings in the patient group. However, this was not significant when accounting for multiple comparisons (Table 1).
Neuropsychological performance
| Cognitively normal participants | MCI patients | P-value | |||
|---|---|---|---|---|---|
| n = 16 | n = 14 | ||||
| Mean | SD | mean | SD | ||
| MMSE | 29.56 | 0.51 | 25.79 | 1.25 | <0.001 |
| VLMTDR (words) | 11.25 | 2.18 | 2.71 | 2.02 | <0.001 |
| ROCFDR | 23.00 | 3.83 | 5.32 | 3.23 | <0.001 |
| TMT-A (s) | 42.25 | 10.67 | 67.29 | 30.52 | 0.010* |
| TMT-B (s) | 97.56 | 43.58 | 248.79 | 65.70 | <0.001 |
| LPS4 | 25.50 | 3.16 | 16.57 | 4.67 | <0.001 |
| BTA | 18.06 | 1.12 | 14.07 | 2.73 | <0.001 |
| BDI | 18.00 | 12.26 | 19.07 | 10.22 | 0.796 |
| B-ADL | 1.50 | 0.58 | 2.24 | 1.10 | 0.036* |
| Cognitively normal participants | MCI patients | P-value | |||
|---|---|---|---|---|---|
| n = 16 | n = 14 | ||||
| Mean | SD | mean | SD | ||
| MMSE | 29.56 | 0.51 | 25.79 | 1.25 | <0.001 |
| VLMTDR (words) | 11.25 | 2.18 | 2.71 | 2.02 | <0.001 |
| ROCFDR | 23.00 | 3.83 | 5.32 | 3.23 | <0.001 |
| TMT-A (s) | 42.25 | 10.67 | 67.29 | 30.52 | 0.010* |
| TMT-B (s) | 97.56 | 43.58 | 248.79 | 65.70 | <0.001 |
| LPS4 | 25.50 | 3.16 | 16.57 | 4.67 | <0.001 |
| BTA | 18.06 | 1.12 | 14.07 | 2.73 | <0.001 |
| BDI | 18.00 | 12.26 | 19.07 | 10.22 | 0.796 |
| B-ADL | 1.50 | 0.58 | 2.24 | 1.10 | 0.036* |
B-ADL = Bayer Activities of Daily Life; BDI = Beck’s Depression Inventory; BTA = Brief Test of Attention; BTA = Brief Test of Attention; LPS4 = Leistungsprüfungssystem 4 (logical thinking); ROCFDR = Rey Osterrieth Complex Figure Test delayed recall; TMT = Trail Making Test; VLMTDR = Verbal Learning Memory Test delayed recall.
*Not significant when accounting for multiple comparisons.
Neuropsychological performance
| Cognitively normal participants | MCI patients | P-value | |||
|---|---|---|---|---|---|
| n = 16 | n = 14 | ||||
| Mean | SD | mean | SD | ||
| MMSE | 29.56 | 0.51 | 25.79 | 1.25 | <0.001 |
| VLMTDR (words) | 11.25 | 2.18 | 2.71 | 2.02 | <0.001 |
| ROCFDR | 23.00 | 3.83 | 5.32 | 3.23 | <0.001 |
| TMT-A (s) | 42.25 | 10.67 | 67.29 | 30.52 | 0.010* |
| TMT-B (s) | 97.56 | 43.58 | 248.79 | 65.70 | <0.001 |
| LPS4 | 25.50 | 3.16 | 16.57 | 4.67 | <0.001 |
| BTA | 18.06 | 1.12 | 14.07 | 2.73 | <0.001 |
| BDI | 18.00 | 12.26 | 19.07 | 10.22 | 0.796 |
| B-ADL | 1.50 | 0.58 | 2.24 | 1.10 | 0.036* |
| Cognitively normal participants | MCI patients | P-value | |||
|---|---|---|---|---|---|
| n = 16 | n = 14 | ||||
| Mean | SD | mean | SD | ||
| MMSE | 29.56 | 0.51 | 25.79 | 1.25 | <0.001 |
| VLMTDR (words) | 11.25 | 2.18 | 2.71 | 2.02 | <0.001 |
| ROCFDR | 23.00 | 3.83 | 5.32 | 3.23 | <0.001 |
| TMT-A (s) | 42.25 | 10.67 | 67.29 | 30.52 | 0.010* |
| TMT-B (s) | 97.56 | 43.58 | 248.79 | 65.70 | <0.001 |
| LPS4 | 25.50 | 3.16 | 16.57 | 4.67 | <0.001 |
| BTA | 18.06 | 1.12 | 14.07 | 2.73 | <0.001 |
| BDI | 18.00 | 12.26 | 19.07 | 10.22 | 0.796 |
| B-ADL | 1.50 | 0.58 | 2.24 | 1.10 | 0.036* |
B-ADL = Bayer Activities of Daily Life; BDI = Beck’s Depression Inventory; BTA = Brief Test of Attention; BTA = Brief Test of Attention; LPS4 = Leistungsprüfungssystem 4 (logical thinking); ROCFDR = Rey Osterrieth Complex Figure Test delayed recall; TMT = Trail Making Test; VLMTDR = Verbal Learning Memory Test delayed recall.
*Not significant when accounting for multiple comparisons.
Patients performed significantly poorer on all neuropsychological tests except for the TMT-A. All patients performed within normal ranges on tests of attention (the Brief Test of Attention) and logical thinking (part 4 of the Leistungsprüfungssystem; Table 1). However, nine patients had pathological scores on the TMT-B and five patients also on the TMT-A. While the predominant impairment in all patients was in the memory domain, five had a purely amnestic MCI subtype, and nine could be classified as multi-domain MCI. There was no difference between placebo and rivastigmine with respect to subjective complaints at the beginning of the session [t(29) = −0.643, P = 0.525]. However, there was a non-significant difference between placebo and rivastigmine regarding the increase of complaints after administration relative to baseline [t(29) = 1.84, P = 0.076]. Groups did not differ with respect to an increased number of complaints under rivastigmine [controls 0.875, SD 5.464; MCI 3.286, SD 6.426, t(28) = −1.111, P = .276].
Behavioural performance
Performance on the item memory task, quantified as d′, was normally distributed and differed significantly between groups without pharmacological stimulation [controls 2.651, SD 0.561, patients 1.416 SD 0.617, t(28) = 5.746, P < 0.001], after administration of placebo [controls 2.714, SD 0.584; MCI 1.379, SD 0.677; t(28) = 5.797, P < 0.001], and after administration of rivastigmine [controls 2.53, SD 0.536; MCI 1.459, SD 0.789; t(28) = 4.397, P < 0.001]. There was a non-significant group difference of the effect of rivastigmine on d′ [controls −0.184, SD 0.444; MCI 0.08 SD = 0.439; t(28) = −1.632, P = 0.114].
Behavioural data: cortical cholinergic system integrity and the effect of rivastigmine
AChE activity, quantified as k3 of MP4A, was significantly decreased in MCI compared to controls (controls mean = 0.064, SD 0.006; MCI mean = 0.057, SD 0.007, t(28) = 3.105, P = 0.004). This effect was most pronounced in the lateral temporal, parietal and occipital lobes, but also encompassed the hippocampi, adjacent medial temporal structures, and frontal areas (Fig. 1 and Table 2) in two clusters of 39 720 and 26 voxels, respectively. For a detailed list of the structures comprising the clusters and the number of significant voxels within each structure see Supplementary Table 3. A voxel-wise analysis revealed no areas where AChE activity was greater in MCI than in controls.
Cortical AChE activity is reduced in MCI due to Alzheimer’s disease
| Structure | Local maxima | ||||
|---|---|---|---|---|---|
| Side | P (FWE-corr.) | x | y | z | |
| Inferior temporal gyrus | Right | <0.001 | 62 | −24 | −24 |
| Middle temporal gyrus | Right | <0.001 | 56 | −14 | −16 |
| Inferior temporal gyrus | Right | <0.001 | 60 | −36 | −20 |
| Right | <0.001 | 62 | −40 | −18 | |
| Right | <0.001 | 56 | −58 | −16 | |
| Planum polare | Right | <0.001 | 40 | −22 | 0 |
| Middle temporal gyrus | Left | <0.001 | −54 | −26 | −8 |
| Planum polare | Left | <0.001 | −42 | −20 | −6 |
| Middle temporal gyrus | Left | <0.001 | −54 | −30 | −8 |
| Left | <0.001 | −52 | −12 | −14 | |
| Left | <0.001 | −54 | −16 | −12 | |
| Left | <0.001 | −48 | −24 | −10 | |
| Structure | Local maxima | ||||
|---|---|---|---|---|---|
| Side | P (FWE-corr.) | x | y | z | |
| Inferior temporal gyrus | Right | <0.001 | 62 | −24 | −24 |
| Middle temporal gyrus | Right | <0.001 | 56 | −14 | −16 |
| Inferior temporal gyrus | Right | <0.001 | 60 | −36 | −20 |
| Right | <0.001 | 62 | −40 | −18 | |
| Right | <0.001 | 56 | −58 | −16 | |
| Planum polare | Right | <0.001 | 40 | −22 | 0 |
| Middle temporal gyrus | Left | <0.001 | −54 | −26 | −8 |
| Planum polare | Left | <0.001 | −42 | −20 | −6 |
| Middle temporal gyrus | Left | <0.001 | −54 | −30 | −8 |
| Left | <0.001 | −52 | −12 | −14 | |
| Left | <0.001 | −54 | −16 | −12 | |
| Left | <0.001 | −48 | −24 | −10 | |
P < 0.05, FWE-corrected with cluster free threshold enhancement. Side: left or right hemisphere.
Cortical AChE activity is reduced in MCI due to Alzheimer’s disease
| Structure | Local maxima | ||||
|---|---|---|---|---|---|
| Side | P (FWE-corr.) | x | y | z | |
| Inferior temporal gyrus | Right | <0.001 | 62 | −24 | −24 |
| Middle temporal gyrus | Right | <0.001 | 56 | −14 | −16 |
| Inferior temporal gyrus | Right | <0.001 | 60 | −36 | −20 |
| Right | <0.001 | 62 | −40 | −18 | |
| Right | <0.001 | 56 | −58 | −16 | |
| Planum polare | Right | <0.001 | 40 | −22 | 0 |
| Middle temporal gyrus | Left | <0.001 | −54 | −26 | −8 |
| Planum polare | Left | <0.001 | −42 | −20 | −6 |
| Middle temporal gyrus | Left | <0.001 | −54 | −30 | −8 |
| Left | <0.001 | −52 | −12 | −14 | |
| Left | <0.001 | −54 | −16 | −12 | |
| Left | <0.001 | −48 | −24 | −10 | |
| Structure | Local maxima | ||||
|---|---|---|---|---|---|
| Side | P (FWE-corr.) | x | y | z | |
| Inferior temporal gyrus | Right | <0.001 | 62 | −24 | −24 |
| Middle temporal gyrus | Right | <0.001 | 56 | −14 | −16 |
| Inferior temporal gyrus | Right | <0.001 | 60 | −36 | −20 |
| Right | <0.001 | 62 | −40 | −18 | |
| Right | <0.001 | 56 | −58 | −16 | |
| Planum polare | Right | <0.001 | 40 | −22 | 0 |
| Middle temporal gyrus | Left | <0.001 | −54 | −26 | −8 |
| Planum polare | Left | <0.001 | −42 | −20 | −6 |
| Middle temporal gyrus | Left | <0.001 | −54 | −30 | −8 |
| Left | <0.001 | −52 | −12 | −14 | |
| Left | <0.001 | −54 | −16 | −12 | |
| Left | <0.001 | −48 | −24 | −10 | |
P < 0.05, FWE-corrected with cluster free threshold enhancement. Side: left or right hemisphere.
A two-way mixed ANCOVA including changes in well-being across treatments as covariate of no interest revealed a group × treatment interaction which missed significance [F(1,27) = 4.228, P = 0.05]. To operationalize the cholinergic deficit in MCI, k3 was averaged across voxels, where it was significantly lower in patients than in controls. When including the resulting values as covariate in the ANCOVA, the group × treatmentinteraction was not significant [F(1,26) = 2.874, P = 0.102]. However, when including hippocampal grey matter density as a covariate of no interest, there was a significant group × treatment interaction [F(1,26) = 5.581, P = 0.026]. Including episodic memory performance at delayed recall, i.e. a measure of hippocampal function, as a covariate of no interest yielded an even stronger effect [F(1,26) = 10.787, P = 0.003]. This finding was independent of age and white matter lesion load (see Supplementary material for details).
Functional MRI results
Subsequent memory effect across groups
Across groups, increased neural activation on subsequently remembered items was observed in the left anterior hippocampus (maximum at MNI coordinates x = −30, y = −8, z = −22, Fig. 2 and Table 3). Mainly in the left but also in the right fusiform gyri, greater neural activity for subsequently remembered items was observed (Fig. 2 and Table 3). In the posterior cingulate cortex (PCC), decreased neural activation was seen for subsequently remembered items (Fig. 2 and Table 3).
Local maxima of the subsequent memory effect
| Anatomical structure | Contrast | Local maxima | Effect | |||||
|---|---|---|---|---|---|---|---|---|
| Side | Voxels | P (FWE- corr.) | x | y | z | |||
| Amygdala and hippocampus | SME | Left | 110 | 0.015 | −30 | −8 | −22 | Positive |
| Fusiform gyrus | SME | Left | 161 | 0.01 | −42 | −24 | −26 | Positive |
| Left | 95 | 0.028 | −36 | −58 | −16 | Positive | ||
| Right | 46 | 0.021 | 34 | −32 | −22 | Positive | ||
| SME (group difference: HC > MCI) | Left | 188 | 0.003 | −36 | −58 | −18 | HC > MCI | |
| Pharmacological effect (group difference: HC < MCI) | Right | 11 | 0.038 | 34 | −36 | −24 | HC < MCI | |
| PCC | SME | Left | 445 | <0.001 | 2 | −22 | 40 | Negative |
| Left | 17 | 0.042 | 0 | −44 | 20 | Negative | ||
| Left | 14 | 0.042 | 8 | −52 | 32 | Negative | ||
| Left | 13 | 0.045 | −10 | −48 | 34 | Negative | ||
| SME (group difference: HC < MCI) | 35 | 0.042 | 0 | −42 | 32 | HC < MCI | ||
| Pharmacological effect (group difference: HC > MCI) | Right | 14 | 0.045 | 8 | −42 | 36 | HC > MCI | |
| Anatomical structure | Contrast | Local maxima | Effect | |||||
|---|---|---|---|---|---|---|---|---|
| Side | Voxels | P (FWE- corr.) | x | y | z | |||
| Amygdala and hippocampus | SME | Left | 110 | 0.015 | −30 | −8 | −22 | Positive |
| Fusiform gyrus | SME | Left | 161 | 0.01 | −42 | −24 | −26 | Positive |
| Left | 95 | 0.028 | −36 | −58 | −16 | Positive | ||
| Right | 46 | 0.021 | 34 | −32 | −22 | Positive | ||
| SME (group difference: HC > MCI) | Left | 188 | 0.003 | −36 | −58 | −18 | HC > MCI | |
| Pharmacological effect (group difference: HC < MCI) | Right | 11 | 0.038 | 34 | −36 | −24 | HC < MCI | |
| PCC | SME | Left | 445 | <0.001 | 2 | −22 | 40 | Negative |
| Left | 17 | 0.042 | 0 | −44 | 20 | Negative | ||
| Left | 14 | 0.042 | 8 | −52 | 32 | Negative | ||
| Left | 13 | 0.045 | −10 | −48 | 34 | Negative | ||
| SME (group difference: HC < MCI) | 35 | 0.042 | 0 | −42 | 32 | HC < MCI | ||
| Pharmacological effect (group difference: HC > MCI) | Right | 14 | 0.045 | 8 | −42 | 36 | HC > MCI | |
P < 0.05, FWE-corrected with cluster free threshold enhancement. Side: left or right hemisphere.
Local maxima of the subsequent memory effect
| Anatomical structure | Contrast | Local maxima | Effect | |||||
|---|---|---|---|---|---|---|---|---|
| Side | Voxels | P (FWE- corr.) | x | y | z | |||
| Amygdala and hippocampus | SME | Left | 110 | 0.015 | −30 | −8 | −22 | Positive |
| Fusiform gyrus | SME | Left | 161 | 0.01 | −42 | −24 | −26 | Positive |
| Left | 95 | 0.028 | −36 | −58 | −16 | Positive | ||
| Right | 46 | 0.021 | 34 | −32 | −22 | Positive | ||
| SME (group difference: HC > MCI) | Left | 188 | 0.003 | −36 | −58 | −18 | HC > MCI | |
| Pharmacological effect (group difference: HC < MCI) | Right | 11 | 0.038 | 34 | −36 | −24 | HC < MCI | |
| PCC | SME | Left | 445 | <0.001 | 2 | −22 | 40 | Negative |
| Left | 17 | 0.042 | 0 | −44 | 20 | Negative | ||
| Left | 14 | 0.042 | 8 | −52 | 32 | Negative | ||
| Left | 13 | 0.045 | −10 | −48 | 34 | Negative | ||
| SME (group difference: HC < MCI) | 35 | 0.042 | 0 | −42 | 32 | HC < MCI | ||
| Pharmacological effect (group difference: HC > MCI) | Right | 14 | 0.045 | 8 | −42 | 36 | HC > MCI | |
| Anatomical structure | Contrast | Local maxima | Effect | |||||
|---|---|---|---|---|---|---|---|---|
| Side | Voxels | P (FWE- corr.) | x | y | z | |||
| Amygdala and hippocampus | SME | Left | 110 | 0.015 | −30 | −8 | −22 | Positive |
| Fusiform gyrus | SME | Left | 161 | 0.01 | −42 | −24 | −26 | Positive |
| Left | 95 | 0.028 | −36 | −58 | −16 | Positive | ||
| Right | 46 | 0.021 | 34 | −32 | −22 | Positive | ||
| SME (group difference: HC > MCI) | Left | 188 | 0.003 | −36 | −58 | −18 | HC > MCI | |
| Pharmacological effect (group difference: HC < MCI) | Right | 11 | 0.038 | 34 | −36 | −24 | HC < MCI | |
| PCC | SME | Left | 445 | <0.001 | 2 | −22 | 40 | Negative |
| Left | 17 | 0.042 | 0 | −44 | 20 | Negative | ||
| Left | 14 | 0.042 | 8 | −52 | 32 | Negative | ||
| Left | 13 | 0.045 | −10 | −48 | 34 | Negative | ||
| SME (group difference: HC < MCI) | 35 | 0.042 | 0 | −42 | 32 | HC < MCI | ||
| Pharmacological effect (group difference: HC > MCI) | Right | 14 | 0.045 | 8 | −42 | 36 | HC > MCI | |
P < 0.05, FWE-corrected with cluster free threshold enhancement. Side: left or right hemisphere.
The SME. Across groups, successful encoding of visual stimuli was associated with activation of temporal lobe structures and deactivation in the posterior cingulate cortex. P < 0.05, FWE-corrected with cluster free threshold enhancement. A = anterior; FusG = fusiform gyrus; HC = hippocampus; PE = parameter estimates; R = right. Slice labels indicate MNI coordinates.
Group differences in the subsequent memory effect
No group difference in the SME was observed in the hippocampus. In the left fusiform gyrus, controls showed greater neural activation on subsequently remembered items than MCI patients (Fig. 3 and Table 3). In the PCC, greater deactivation on subsequently remembered items was observed in controls than in MCI patients (Fig. 3 and Table 3).
Neural activation for successfully encoded content differs between patients with MCI and cognitively normal individuals. In the PCC, patients did not show deactivation. Conversely, in the fusiform gyrus neural activation was greater in controls than in the patient group. P < 0.05, FWE-corrected with cluster free threshold enhancement. A = anterior; CN = cognitively normal; FusG = fusiform gyrus; PE = parameter estimates; R = right. Slice labels indicate MNI coordinates.
Group differences in the effect of rivastigmine on neural activation
In response to rivastigmine, MCI patients showed an increase and controls showed a decrease of neural activation in the right fusiform gyrus (Fig. 4 and Table 3). In the PCC, patients showed a decrease in neural activation, while controls exhibited an increase (Fig. 4 and Table 3). No difference between groups in the pharmacological effect on the SME was observed in the hippocampus.
Cholinergic stimulation differentially affects the neural SME in MCI patients and cognitively normal individuals. (A) Group difference in treatment effect (controlrivastigmine − controlplacebo) − (MCIrivastigmine − MCIplacebo) in the right fusiform gyrus. (B) The bar graph depicts the change in neural activation under cholinergic stimulation (Δrivastigmine − placebo) in the fusiform gyrus, illustrating that activation increased in patients and decreased in controls. (C) In the fusiform gyrus this change in neural activation was negatively associated with local acetylcholinesterase activity. (D) Group difference in treatment effect (controlrivastigmine − controlplacebo) − (MCIrivastigmine − MCIplacebo) in the PCC. (E) This bar graph depicts the change in neural activation under cholinergic stimulation (Δrivastigmine − placebo) in the PCC: an increase in controls and a decrease in MCI patients. (F) In the PCC, the change in neural activation under cholinergic stimulation was positively associated with local acetylcholinesterase activity. Statistical maps were FWE-corrected with cluster free threshold enhancement (P < 0.05). CN = cognitively normal; FusG = fusiform gyrus; k3 = hydrolysis rate constant of PET-tracer MP4A; PE = parameter estimates; R = right,. Slice labels indicate MNI coordinates.
Effect of rivastigmine on neural activation and its relationship to local acetylcholinesterase activity
Average parameter estimates extracted from the areas showing a significant group difference in the pharmacological effect on neural activity were highly correlated with local cortical k3 (fusiform gyrus: Spearman’s rho = −0.476, P = 0.008, PCC: Spearman’s rho = 0.481, P = 0.007), as assessed with a correlation analysis including both groups. This effect remained when accounting for local grey matter density and group as covariates of no interest in a multiple regression analysis (fusiform gyrus: β = −0.394, P = 0.019; PCC: β = 0.359, P = 0.035; Fig. 4). Within the control group, a multiple regression showed the same pattern with a stronger positive association in PCC (β = 0.784, P = 0.01), but a negative trend in the fusiform gyrus, which missed significance (β = −0.499, P = 0.056). In the patient group, the associations had the same direction but were not significant in either region (PCC: β = 0.273, P = 0.380; fusiform gyrus: β = −0.341, P = 0.174). This finding was independent of age and white matter lesion load (see Supplementary material for details).
Discussion
We observed that neural activation underlying episodic memory encoding depends on local AChE activity as a marker for the integrity of the cortical cholinergic system. AChE activity was significantly reduced in MCI due to Alzheimer’s disease. Characteristic neural effects of the SME—i.e. activation of the fusiform gyrus and deactivation in the PCC—were attenuated in patients, but partially normalized under cholinergic stimulation. This positive pharmacological effect of rivastigmine on neural activation was significantly correlated with independently assessed local AChE activity. In response to rivastigmine, greater deactivation in the PCC and activation in the fusiform gyrus were associated with lower AChE activity in the respective regions. The effect of rivastigmine on memory also depended on the cortical cholinergic deficit in MCI. There was a significant group × treatment interaction effect of rivastigmine on memory performance, which was not present when accounting for the group difference in AChE activity. Interestingly, this interaction was especially significant when accounting for hippocampal structure (hippocampal grey matter density) and function (baseline verbal memory performance). The integrity of the hippocampus and the reduction of cortical AChE activity thus appear to have separate influences on the effect of rivastigmine on memory.
Group difference in acetylcholinesterase activity: the cortical cholinergic deficit in mild cognitive impairment
As previously described (Haense et al., 2012; Marcone et al., 2012), the most significant decrease in AChE activity was observed in the lateral temporal lobes (i.e. the middle and inferior temporal gyrus as well as the planum polare). AChE reduction was less pronounced in the hippocampal formation. This may be attributable to the different origins of cholinergic neurons providing efferents to neocortical regions and the hippocampal formation: while the neocortex and amygdala are supplied by cholinergic neurons in the posterior basal nucleus of Meynert, the hippocampus receives its cholinergic input from the medial septal nuclei and the diagonal band of Broca (Mesulam and Geula, 1988b). Since Alzheimer pathology initially and most severely affects cholinergic neurons in the posterior basal nucleus of Meynert (Sassin et al., 2000) a more pronounced AChE decrease in the lateral temporal cortex compared to the hippocampal formation is likely to reflect differential effects of Alzheimer’s disease related degeneration of separate cholinergic nuclei.
In the neocortex and the hippocampus, AChE is found in the axons of cholinergic neurons that originate from the basal forebrain (Mesulam and Geula, 1988b), and to a lesser extent in AChE-rich pyramidal neurons, both of which show a significant reduction in Alzheimer’s disease (Mesulam and Geula, 1988a). Since the majority of the cholinergic axons in the cortex originate from the basal forebrain, cortical AChE activity also provides information about the integrity of cholinergic tracts (Mesulam and Geula, 1988b).
Other measures of cholinergic neurotransmission, i.e. the presynaptic vesicular acetylcholine transporter (VChAT) and nicotinic acetylcholine receptors, show similar alterations to AChE: their levels are decreased in PET studies of MCI and early dementia due to Alzheimer’s disease, albeit to a lesser extent than in later stages (Kuhl et al., 1996; Nordberg, 2001; Rinne et al., 2003; Mazère et al., 2008; Haense et al., 2012; Marcone et al., 2012). Choline acetyltransferase (ChAT), another key enzyme in cholinergic neurotransmission, is an exception to this pattern. While it is generally co-localized with AChE in cholinergic neurons in normal ageing and Alzheimer’s disease (Mesulam and Geula, 1992; Geula and Mesulam, 1996), it has been found to be increased in MCI due to Alzheimer’s disease in the hippocampus and frontal lobe (DeKosky et al., 2002; Ikonomovic et al., 2003). This increase may be due to reactive synaptogenesis arising from cholinergic neurons in the medial septal nuclei following a loss of glutamatergic input from the entorhinal cortex to the hippocampus (Cotman et al., 1973; Savaskan and Nitsch, 2001). Importantly, an increase of ChAT is not necessarily at odds with a degeneration of the cortical cholinergic system, as cholinergic fibre abnormalities observed in ageing and Alzheimer’s disease include an enlargement of nerve terminals immunoreactive to ChAT (Geula et al., 2008). Interestingly, neuronal counts (Gilmor et al., 1999) and AChE activity (Herholz et al., 2004) in the basal forebrain remain normal in MCI and mild dementia, although there is evidence from macroscopic MRI studies indicating a progressive atrophy of the basal forebrain in ageing and Alzheimer’s disease (Grothe et al., 2012). In summary, AChE and most other cholinergic markers provide a relatively consistent picture of a progressive degeneration of the cholinergic inputs to the cortex and the hippocampus.
The pharmacological modulation of the subsequent memory effect depends on local acetylcholinesterase activity
In patients and controls, cholinergic stimulation induced opposite effects on the SME. In the fusiform gyrus, activation increased in MCI but decreased in controls. Conversely, in the PCC, a region typically deactivated during episodic memory encoding, rivastigmine administration elicited a greater deactivation in MCI but reduced the deactivation in controls. These results are in line with previous studies, reporting increased activation in the fusiform gyrus in Alzheimer’s disease in response to acute (Rombouts et al., 2002; Kircher et al., 2005) and chronic cholinergic stimulation (Pa et al., 2013) and improved deactivation in postero-medial regions under chronic administration of donepezil in MCI (Risacher et al., 2013). Crucially, in our data the direction and size of the effect of cholinergic stimulation on neural activation depended on local AChE activity: low local AChE activity was associated with increased neural activation in the fusiform gyrus under rivastigmine. In the PCC, by contrast, reduced AChE activity was associated with greater deactivation under rivastigmine. In other words, data suggest that cholinergic stimulation may only exert its beneficial effects on neural activation in the presence of a local cholinergic deficit. Likewise, detrimental effects of rivastigmine on neural activation and behavioural performance in controls (in contrast to the beneficial effects in MCI) can best be explained by an ‘inverted-U’ response, often observed in pharmacotherapy (Yerkes and Dodson, 1908; Diamond et al., 2007): While controls, who already operated close to optimal levels at baseline, performed poorer after drug administration, performance in patients with a cholinergic deficit was enhanced (Kukolja et al., 2009; Bentley et al., 2011).
Generally, cholinergic input to the cerebral cortex is of a modulatory nature, reducing background activity, increasing the responsiveness of cortical neurons to sensory input (Hasselmo et al., 1992; Patil and Hasselmo, 1999; Mesulam, 2004), and facilitating long-term potentiation (Huerta and Lisman, 1993; Dringenberg et al., 2007). The observed changes in neural activation are in line with findings that cholinergic stimulation optimizes processing of task-relevant stimuli (Sarter et al., 2005) and facilitates a shift from an internally focused state with an active default-mode network to the processing of external stimuli (Hahn et al., 2007; Bentley et al., 2011). During initial episodic memory formation, neural resources are redistributed, enhancing activity in regions relevant for encoding at the cost of activity in other regions (Bentley et al., 2009; Kukolja et al., 2009). Cholinergic stimulation supports this redistribution by enhancing activations in the fusiform gyrus, part of the encoding network (Cabeza and Nyberg, 2000; Kim, 2011), and deactivations in the PCC.
We did not observe an effect of rivastigmine on neural activation in the medial temporal lobe. The SME in the hippocampus has been shown to increase after chronic, but not acute administration of the AChE inhibitor galantamine (Goekoop et al., 2004). The most parsimonious explanation for the discrepancy of our finding of no rivastigmine effect on neural activation in the medial temporal lobe and studies which showed increased hippocampal activation after acute administration is that those studies used tasks emphasizing spatial context association (Kukolja et al., 2009), or navigation (Grön et al., 2006), placing a higher demand on hippocampal function.
The neural subsequent memory effect differs between groups
Across groups a positive SME was observed in the left anterior hippocampus and the fusiform gyrus bilaterally. This lateralization has been demonstrated in several studies for item memory (Spaniol et al., 2009; Kim, 2011). In accordance with the literature, in the PCC the SME was negative (Kim, 2011). In the fusiform gyrus, the SME was greater in controls than in patients as has been reported in early Alzheimer’s disease (Golby et al., 2005; Kircher et al., 2005). In the PCC on the other hand, a less negative SME was observed in MCI than in controls, which is a consistent finding in studies examining episodic memory in MCI (Petrella et al., 2007; Pihlajamäki et al., 2010; Risacher et al., 2013; Nellessen et al., 2015).
The SME in the hippocampal formation did not differ between groups. Previous studies have found both increased (Trivedi et al., 2008) and decreased (Pariente et al., 2005) SMEs in MCI compared to controls. Findings are inconsistent regarding changes in memory-related hippocampal activation in MCI, which has been attributed to different patterns depending on disease stage. Compensatory hyperactivation may precede decreased activation at later stages (Nellessen et al., 2015).
The fact that hippocampal activation did not differ between groups may also reflect load dependency of neural activation. At low loads, patients with MCI have been shown to exhibit greater neural activation than controls in the hippocampus, while the opposite pattern was observed at high cognitive loads (de Rover et al., 2011). In analogy to the ‘compensation-related utilization of neural circuits hypothesis’ (CRUNCH) (Reuter-Lorenz and Cappell, 2008) of cognitive ageing, it has been argued that in MCI additional neural resources are recruited as a compensatory mechanism, which breaks down when cognitive load passes a threshold. Hence, when neural reserves are exhausted activation is lower in MCI than in cognitively normal individuals (Grady, 2012).
Factors influencing the behavioural response to rivastigmine
As expected, baseline performance on the item memory task was significantly worse in MCI patients, but also after application of placebo and rivastigmine. There was a weak trend for a group difference in the change in memory performance, but no group × treatment interaction. While participants with severe adverse reactions were excluded from analysis a priori, milder side-effects, which are common under chronic (Birks and Flicker, 2006) and acute cholinergic pharmacotherapy (Bentley et al., 2008), were not accounted for in this analysis. When correcting for such subtle changes in well-being, the group × treatment interaction barely missed significance, suggesting an association with the effect of rivastigmine on performance. In other words, even mild side-effects of AChE inhibition may interfere with its effects on memory performance.
To confirm the relevance of the cortical cholinergic deficit in MCI, i.e. the reduction of cortical AChE especially in lateral temporal lobes, AChE values extracted from these areas were included in the ANCOVA as a covariate of no interest. This decreased the strength of the group × treatment interaction, leaving it insignificant. This pattern is consistent with a close association of the cortical cholinergic deficit and the presence of Alzheimer’s disease.
As damage to memory relevant medial temporal lobe structures can already be observed at the MCI stage of Alzheimer’s disease (Nickl-Jockschat et al., 2012; Wolk et al., 2017), we examined whether the integrity of the hippocampus has a separate influence on the behavioural effect of rivastigmine. When including baseline episodic memory performance as measure of hippocampal function, the group × treatment interaction was highly significant. A similar effect was observed when replacing verbal episodic memory performance with hippocampal grey matter density to reflect hippocampal structure. It thus appears that the cortical cholinergic deficit and hippocampal integrity are separately associated with the response of memory performance to cholinergic stimulation: a benefit of AChE inhibition was only observed if cortical AChE activity was reduced but hippocampal structures were relatively intact. This could be of prognostic value when deciding to treat MCI due to Alzheimer’s disease with AChE inhibitors: a patient with little hippocampal atrophy, which can be assessed on a clinical MRI scan, may be more likely to benefit from AChE inhibition than one with more pronounced medial temporal lobe atrophy.
There is also the possibility that the change in performance is not a consequence of enhanced memory-related neural activation, but improved stimulus processing. This is corroborated by the pattern of the pharmacological effect on neural activation. Activation increased in the fusiform gyrus, which plays a central role in the processing of visual stimuli (Grill-Spector et al., 2001; Garoff et al., 2005) and decreased in the PCC as part of the default-mode network (Fox and Raichle, 2007), but remained unaltered in the hippocampal formation, a key structure in episodic memory (Bird and Burgess, 2008). The notion that AChE inhibition mainly affects stimulus processing and attention, thereby only indirectly improving memory, could explain why large clinical trials consistently report a significant improvement of global cognition under AChE inhibitors in Alzheimer’s dementia, but not specifically of episodic memory (Birks, 2006).
Limitations
Our study sample size was comparatively small because of a significant number of dropouts, which were largely attributable to the complex study design. However, a consecutive bias on the results is unlikely, as the total number of dropouts (controls: six, patients: six) due to side-effects (controls: two, patients: two) was identical in both groups. Although the sample size studied is relatively small, our results are consistent with previous studies of the cholinergic system using PET or pharmacological challenges. Another limitation lies in the fact that an acute single rivastigmine challenge was applied only. Therefore, it remains to be elucidated whether our results regarding cortical AChE activity and the response to chronic cholinergic treatment in MCI due to Alzheimer’s disease extend to chronic application of rivastigmine, which warrants future inverstigations.
Conclusions
The present data demonstrate that a cortical cholinergic deficit is already present at the MCI stage of Alzheimer’s disease. It affects mainly lateral temporal and parietal cortices and predicts the response to cholinergic stimulation at both the neural and behavioural levels. However, our data also indicate that the cholinergic deficit is one of several factors that influence the response to cholinergic stimulation. We provide evidence that improvement of memory performance depends on the presence of a local cholinergic deficit in task-relevant structures, treatment side-effects, and the functional and structural integrity of the hippocampus. Data suggest that the presence of a cortical cholinergic deficit and relatively intact hippocampal function may be prognostic factors for the response of memory to AChE inhibitors, a clinical hypothesis that warrants further investigation.
Acknowledgements
We are grateful to our colleagues from the Clinical Trials Center Cologne (Zentrum für Klinische Studien Köln), the Department of Nuclear Medicine, and the former Max-Planck-Institute for Neurological Research (now Max-Planck-Institute for Metabolism Research), in particular D. Yves von Cramon, Rudolf Graf, and Matthias Schmidt, for their valuable support.
Funding
This work was supported by a grant from the Medical Faculty of the University of Cologne (Forschungspool Klinische Studien, # 2620000301), the Marga and Walter Boll Foundation, Kerpen, Germany, to G.R.F. and J.K., and the Brandau-Laibach Foundation, Cologne, Germany, to J.K.
Supplementary material
Supplementary material is available at Brain online.
Abbreviations
- AChE
acetylcholinesterase
- MCI
mild cognitive impairment
- MP4A
11C-N-methyl-4-piperidyl acetate
- PCC
posterior cingulate cortex
- SME
subsequent memory effect
- VBM
voxel based morphometry
