Amyloid-independent atrophy patterns predict time to progression to dementia in mild cognitive impairment

Subjects with mild cognitive impairment (MCI) are at increased risk of dementia with annual conversion rates of 10–15% [1, 2]. According to the NIA–AA research criteria [3], a diagnosis of MCI due to AD requires the presence of amyloid pathology as measured in cerebrospinal fluid (CSF) or on amyloid PET. Since amyloid reaches a plateau relatively early in the disease course [4‐6], it has limited prognostic value for the time to onset of dementia. Neuronal injury markers, such as brain atrophy measured with structural MRI, are more closely related to cognitive impairment and could thus be useful for estimating time to clinical progression [7‐10]. Previous studies found that hippocampal atrophy can be used to predict time to dementia in MCI patients with positive amyloid markers [11, 12]. However, other brain regions may also be valuable for the prediction of progression, as indicated by voxel-based morphometry studies [13‐15]. Two previous studies have used hypothesis-free voxel-level survival analyses to show that decreased gray matter in the medial temporal lobe and posterior cingulate cortex can predict time to conversion to AD-type dementia in nondemented subjects [16, 17]. However, it is still unclear whether such predictive atrophy patterns are specific for amyloid pathology.

In this study, we performed unbiased voxel-wise survival analyses to detect brain regions where decreased gray matter volume is associated with time to progression to dementia, and examined whether patterns associated with progression to dementia were dependent on amyloid status in a large sample of subjects with MCI.

Of the 276 subjects with MCI, 122 (44%) progressed to dementia. Among those who progressed to dementia, 104 (85%) subjects progressed to AD-type dementia and 18 (15%) subjects to another type of dementia (four fronto-temporal dementia, eight vascular dementia, one mixed vascular and AD, three Lewy body dementia, and two dementia unspecified). Clinical characteristics of subjects are summarized in Table 1. Subjects who progressed were on average older, had lower baseline scores on the Mini-Mental State Examination (MMSE), and had lower baseline CSF Aβ_1–42 than subjects who remained stable. The groups had similar follow-up times. Subjects who were amyloid positive had a higher risk of progressing to dementia compared to amyloid-negative subjects (HR 2.4, p < 0.001). When stratifying for amyloid status, 160 subjects were amyloid positive and 99 (62%) of them showed clinical progression. Of those amyloid-positive subjects who progressed, most subjects progressed to AD-type dementia (n = 94, 95%). A total of 116 subjects were amyloid negative, of whom 23 (20%) subjects progressed. Amyloid-negative subjects more often progressed to non-AD dementias (57%) than AD-type dementia (43%). Within the amyloid-positive group, continuous CSF Aβ_1–42 levels were unrelated to progression to dementia (HR 1.0, p = 0.8). Within the amyloid-negative group, continuous CSF Aβ_1–42 levels were associated with an increased risk of progression to dementia (HR 2.2, p = 0.01).

The main finding of this article is that a widespread pattern of decreased gray matter volume, beyond the hippocampal region, is predictive of time to progression to dementia in subjects with MCI. The presence of amyloid pathology was also a predictor of time to progression to dementia. The pattern of decreased gray matter volume that was predictive of progression was mostly similar in amyloid-positive and amyloid-negative subjects.

The cortical pattern of decreased gray matter volume predicting progression to dementia in subjects with MCI included the temporal, parietal, and frontal regions. Most of the voxels predictive of progression to dementia are located in brain regions that typically show atrophy and hypometabolism as a sign of neurodegeneration in subjects with AD [30, 31], and are known to be pathologically involved in AD [32]. Our results are largely in line with previous voxel-based morphometry studies comparing MCI subjects who progress to dementia with those who remain stable over time. These studies have identified the medial and lateral temporal lobe [13‐15] and also the parietal and frontal areas [15] to be associated with progression. Only two studies have so far also applied voxel-level survival analysis to structural MRI data. Using this method, intersubject variability in rates of clinical progression and follow-up time can be taken into account. Those studies also reported that atrophy in (mesial) temporal and posterior cingulate structures is predictive of cognitive decline [16, 17]. The patterns of low gray matter volume predictive of decline observed in our study were more extensive than in these previous studies and additionally revealed the frontal and more widespread parietal regions to be involved. This could partially be explained by our higher power due to the larger sample size. Also, Zeifman et al. [17] examined clinical progression from a cognitively normal stage, whereas our sample comprised subjects with MCI at baseline, who have more atrophy and a higher probability to progress to dementia within a follow-up time of a few years.

The regional patterns of decreased gray matter volume predictive of progression to dementia in subjects with evidence of amyloid pathology were largely similar to those found in the whole sample, and were located in regions known to be associated with Alzheimer pathology. This is in line with the notion that subjects with MCI and evidence of amyloid plaques are at an early stage of AD [3]. In addition, we found that decreased gray matter volume in most of these regions was also associated with progression to dementia in MCI subjects with normal amyloid at baseline. This suggests that the regions of decreased gray matter volume we found to be predictive for time to dementia are not specific for amyloid pathology. These findings are in line with MRI studies using a priori defined areas, which have also found that brain regions typically atrophied in AD were predictive of cognitive decline in MCI subjects both with and without evidence of amyloid pathology [7, 9, 11]. We further extend those results by showing that a more widespread anatomical pattern of decreased gray matter volume is related to cognitive decline, independent of the presence of amyloid pathology.

A possible explanation for our results is that the subjects who progressed within the amyloid-negative group do have underlying AD pathology, but are above the cut-off point for amyloid positivity. CSF Aβ_1–42 levels, although well above the threshold, were significantly lower in subjects who progressed than in those who remained stable (Table 1). Further characterizing the subjects who progressed within the amyloid-negative group revealed that those subjects who progressed to clinical AD type dementia also had lower baseline CSF Aβ_1–42 than those who progressed to other dementias (740 ± 129 vs 930 ± 158, p < 0.01). Possibly, these subjects might have reached abnormal CSF Aβ_1–42 levels at the time of dementia diagnosis (not tested), since several studies have shown that higher amyloid burden at baseline, even within the normal range, is associated with future amyloid depositions [10, 33, 34]. An alternative explanation is that subjects who progressed to clinical dementia in the absence of amyloid pathology are akin to the concept of suspected non-amyloid pathology (SNAP). This concept has been constructed for subjects who present with evidence of AD-like neurodegeneration in the absence of amyloid pathology [35]. We studied subjects with MCI, who suffer from cognitive impairment, suggesting that a neurodegenerative process has already caused neuronal damage, but the underlying cause might not be AD. There are several disorders that show a spatially overlapping atrophy pattern with regions that are typically affected in AD. For example, the medial temporal lobe is involved in hippocampal sclerosis and TDP43 pathology can be associated with widespread cortical atrophy [36‐38]. The association between regional amyloid deposition in the brain and cortical atrophy is still unclear [32, 39]. Our results support the idea that some brain regions are selectively vulnerable to pathological factors in general, and that amyloid pathology is one possible initiator of a common neurodegenerative pathway [40].

Although the overall anatomical pattern of decreased gray matter volume associated with time to progression was largely similar in amyloid-positive and amyloid-negative subjects, some subtle differences can be appreciated in Fig. 2 and Additional file 2: Figure S2. In the ventromedial prefrontal cortex, HRs were higher in amyloid-negative subjects compared to amyloid-positive subjects. Furthermore, the right fusiform gyrus was only significantly associated with time to progression in amyloid-negative subjects. This region has previously been included in an AD signature cortical thickness measure [41]. Whether decreased gray matter volume in these regions could be an indication of non-AD pathology in subjects with MCI will need to be examined in larger studies.

A potential limitation of this study is that the data were acquired at a clinical center over a relatively long period of time, during which the dementia work-up protocols changed, diagnostic criteria evolved, and higher field-strength MRI scanners were implemented. Although we have corrected for the potential influences where possible, we cannot exclude the possibility that this variability might have led to an underestimation of our results. Still, this can also be considered a strong point of our study, as it supports the robustness of our results. Another limitation is that we defined the follow-up time when the clinical diagnosis changed as the outcome parameter for the survival analyses. Since follow-up times were not strictly standardized but rather based on clinical judgment and timing of a yearly appointment, this might have biased the results. However, we think this is unlikely since the time of follow-up, with an average of 2.5 ± 1.5 and ranging up to 11 years, was similar for subjects who progressed and those who remained stable.

Widespread decreases in gray matter volume are useful for the prediction of clinical progression and time to dementia in subjects with MCI. Findings were largely similar in subjects with and without evidence of amyloid pathology. This leads us to consider that although brain atrophy does not seem specific for the underlying pathology, it is a useful marker that reflects incipient dementia and thereby is valuable for predicting clinical progression.