Article | Open | Published:

# Lifespan Changes of the Human Brain In Alzheimer’s Disease

## Introduction

Alzheimer’s disease (AD) is the most prevalent form of dementia in persons older than 65 years1. Cognitive impairment, mainly related to memory deficits, is the most common manifestation of this disease2. Available neuroimaging evidence suggests that the neuropathological alterations underlying AD probably begin much earlier than the appearance of clinical symptoms and years before clinical diagnosis3. From these results, it appeared that the pharmacological management was finally implemented in patients with a largely advanced neurodegenerative process, making it difficult to fight against pathological progression. In this context, the concept of disease-modifying therapies is emerging and the search for early biomarkers of these alterations is currently a hot topic of research4.

Neurodegeneration, assessed by the level of cerebral atrophy, is one of these biomarkers. In recent decades, several MRI studies have investigated neurodegeneration in the prodromal phase of Alzheimer’s disease5,6. However, very few of them attempted to investigate the preclinical phase of the disease, the very early asymptomatic phase. Indeed, this type of study is a very challenging task since it requires an imaging database starting before the appearance of the disease, and the corresponding long follow-up study. Therefore, in the past, such studies have been based on subjects with rare autosomal dominant mutations associated with a high risk of developing dementia7,8,9,10 or on longitudinal studies with long follow-up in which brain imaging have been performed before the appearance of clinical symptoms (i.e., memory impairment)11,12,13. In these previous long follow-up studies, the starting point of the neurodegeneration was not determined since incident cases already present brain morphometric differences at baseline 7 or 10 years before the diagnosis13,14,15 or 5 before the apparition of Mild Cognitive Impairment (MCI)16. Finally, the lifespan evolution of these imaging biomarkers many years before cognitive decline also remains unknown.

Determining the timing of the onset of neurodegeneration would require longitudinal MRI datasets with several decades of follow-up. So far, due to obvious technical reasons, such imaging dataset does not exist yet. Consequently, in this paper, we propose to consider alternative questions such as – When does the AD model diverge from the normal aging model? How do the trajectories of biomarker models differ in AD from normal aging? To answer these questions, we present an innovative approach based on an extrapolated lifespan model of AD brain structures using large-scale databases. To this end, we propose to take advantage of the new paradigm of BigData sharing in neuroimaging17 by analyzing publically available databases including subjects from a wide age range covering the entire lifespan. Such analysis can be very valuable since epidemiological studies indicate that late dementia is associated with early exposure to risk factors at midlife, highlighting the need to consider brain biomarkers throughout the entire lifespan18,19,20.

## Material and Methods

### Groups definition

This study aims at comparing normal and pathological models of brain structure across the entire lifespan. To this end, models were estimated on four different groups to generate CN, AD/MCI, AD and MCI trajectories.

• For the CN model, we used the N = 2944 subjects from 9 months to 94 y of the cognitively normal dataset as done in21.

• For the AD/MCI model, we used N = 3262 subjects. We mixed AD patients, with MCI patients and with young CN considered as presymptomatic subjects. We used 426 AD patients (from 55 y to 96 y), 959 MCI patients (from 55 y to 92 y) of the AD/MCI dataset and all the CN younger than 55 y (i.e., 1877 subjects). These subjects are included in the CN used for CN trajectory.

• For the AD model, we used N = 2303 samples. We mixed AD patients with young CN. More precisely, we used 426 AD patients (from 55 y to 96 y) of the AD/MCI datasets and all the CN younger than 55 y (i.e., 1877 subjects).

• For the MCI model, we used N = 2836 samples. Here, 959 MCI patients (from 55 y to 92 y) of the AD/MCI datasets were mixed with all the CN younger than 55 y (i.e., 1877 subjects).

### Datasets description

To study structural changes across the entire lifespan for CN and AD, we aggregated several open access databases to construct two datasets. In the following, CN and AD/MCI datasets will be described. The four previously described groups are built on these datasets.

#### Cognitively normal dataset (N = 2944)

The cognitively normal dataset is composed of the 3296 T1-weighted (T1w) MRI. The composition of the nine open access databases used to build the control dataset is provided in Table 1. As explained in21, after a quality control (see image processing section), only 2944 MRI were kept. The female proportion is 47% for the remaining subjects and the age range is [0.75–94] years.

#### AD/MCI dataset (N = 1385)

The AD/MCI dataset is composed of 426 AD patients and 959 MCI patients extracted from OASIS, AIBL, ADNI1 and ADNI2 databases. Details on clinical criterion for groups definition are provided in28 for ADNI1, ADNI2 and AIBL and in29 for OASIS. After a quality control, only 1385 MRI were kept. The female proportion is 44% for the remaining subjects and the age range is [55–96] years (see Table 2).

In the following, more details are provided about acquisition protocols of the different datasets used in this study.

• C-MIND: 266 images of control subjects from the C-MIND dataset (https://research.cchmc.org/c-mind/) are used in this study. All the 3D T1-weight (T1w) MPRAGE high-resolution MRI were acquired at the same site on a 3 T scanner with spatial resolution of 1 mm3 acquired using a 32 channel SENSE head-coil.

• NDAR: 415 of control subjects from the Database for Autism Research (NDAR) (https://ndar.nih.gov) are used in this study. The T1w 3D MRI were acquired on 1.5 T MRI and 3 T scanners. In our experiments, we used the NIHPD (http://www.bic.mni.mcgill.ca/nihpd/info/data_access.html) dataset and 197 images of control subjects from the Lab Study 19 of National Database for Autism Research. For the NIHPD dataset, the 3D T1w SPGR MRI were acquired at six different sites with 1.5 Tesla systems by General Electric (GE) and Siemens Medical Systems with spatial resolution of 1 mm3. The 3D T1w MPRAGE MRI from the Lab Study 19 were scanned using a 3 T Siemens Tim Trio scanner at each site with spatial resolution of 1 mm3

• ABIDE: 528 control subjects from the Autism Brain Imaging Data Exchange (ABIDE) dataset (http://fcon_1000.projects.nitrc.org/indi/abide/) are used in this study. The MRI are T1w MPRAGE acquired at 20 different sites on 3 T image and the details of acquisition, informed consent, and site-specific protocols are available on the website.

• ICBM: 308 normal subjects from the International Consortium for Brain Mapping (ICBM) dataset (http://www.loni.usc.edu/ICBM/) obtained through the LONI website are used in this study. The T1w MPRAGE MRI were acquired on a 1.5 T Philips GyroScan imaging system (Philips Medical Systems, Best, The Netherlands) with spatial resolution of 1 mm3.

• IXI: 588 normal control from Information eXtraction from Images (IXI) database (http://brain-development.org/ixi-dataset/) are used in this study. The MRI are T1w images collected at 3 sites with 1.5 and 3 T scanners with spatial resolution close to 1 mm3.

• OASIS: 315 control subjects and 98 AD/MCI patients from the Open Access Series of Imaging Studies (OASIS) database (http://www.oasis-brains.org) are used in this study. The MRI are T1w MPRAGE image acquired on a 1.5 T Vision scanner (Siemens, Erlangen, Germany) and resliced at 1 mm3.

• ADNI1: 228 control subjects and 587 AD/MCI patients from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu) phase 1 are used in this study. These baseline MRI are T1w MPRAGE acquired on 1.5 T scanners at 60 different sites across the United States and Canada with reconstructed spatial resolution of 1 mm3.

• ADNI2: 215 control subjects and 621 AD/MCI patients from the ADNI2 database (second phase of the ADNI project) are used in this study. The baseline MRI are T1w MPRAGE acquired on 3 T MR scanners with the standardized ADNI-2 protocol (www.loni.usc.edu) with spatial resolution close to 1 mm3.

• AIBL: 236 control subjects and 112 AD/MCI patients from the Australian Imaging, Biomarkers and Lifestyle (AIBL) database (http://www.aibl.csiro.au/) are used in this study. The baseline MRI are T1w image acquired on 3 T MR scanners with the ADNI protocol (http://adni.loni.ucla.edu/research/protocols/mri-protocols) and with custom MPRAGE sequence on the 1.5 T scanners.

### Image processing

All the considered images were processed with the volBrain pipeline30 (http://volbrain.upv.es). The volBrain system is a web-based online tool providing automatic brain segmentation and generating report summarizing the volumetric results. The full processing time is around 10 minutes. In the past 2 years, volBrain has processed online more than 75.000 brains for approximately 1800 users. In a recent work, we compared volBrain pipeline with two well-known tools used on MR brain analysis (FSL and Freesurfer). We showed significant improvements in terms of both accuracy and reproducibility for intra and inter-scanner scan-rescan acquisitions30. The volBrain processing pipeline includes several steps to improve the quality of the input MR images and to homogenize their contrast and intensity range30. The volBrain pipeline achieves the following preprocessing steps: (1) denoising using spatially adaptive non-local means31, (2) rough inhomogeneity correction using N4 method32, (3) affine registration to MNI152 space using ANTS software33, (4) SPM based fine inhomogeneity correction34 and (5) tissue based intensity standardization35. After preprocessing, the brain is segmented into several structures at different scales. First, the total intracranial volume (TIV) is obtained with NICE method36. Then, tissue classification is performed using the TMS method35 and finally subcortical structures are estimated using the non-local label fusion method37. All the segmentation methods of volBrain are based on a library of 50 experts manually labelled cases (covering almost the entire lifespan). It is worth to note that the used manual hippocampus labeling followed the EADC-ADNI harmonized protocol which is the current consensus protocol for hippocampus segmentation in AD38. More details about volBrain pipeline can be found in30. Finally, a multi-stage quality control (QC) procedure was performed to carefully select subjects included. First, a visual assessment was done for all input images by checking screen shots of one sagittal, one coronal and one axial slice in middle of the 3D volume. Then, a visual assessment of processing quality was carried out by using the volBrain report which provides screenshots for each step of the pipeline. Finally, a last control was performed by individually checking with a 3D viewer all outliers detected using the estimated model (see21 for more details).

### Statistical Analysis

Different model types were considered to estimate the final model of each structure. The candidate models were tested from the simplest to the most complex. A model type was kept as a potential candidate only when simultaneously F-statistic based on ANOVA (i.e., model vs. constant model) was significant (p < 0.05) and when all its coefficients were significant using t-statistic (p < 0.05). As in21, the following model types were used as potential candidates:

1. 1.

Linear model

$$Vol={\beta }_{0}+{\beta }_{1}Age+\varepsilon$$
2. 2.

$$Vol={\beta }_{0}+{\beta }_{1}Age+{\beta }_{2}Ag{e}^{2}+\varepsilon$$
3. 3.

Cubic model

$$Vol={\beta }_{0}+{\beta }_{1}Age+{\beta }_{2}Ag{e}^{2}+{\beta }_{3}Ag{e}^{3}+\varepsilon$$
4. 4.

Linear hybrid model: exponential cumulative distribution for growth with linear model for aging

$$Vol={\beta }_{4}.(1-{e}^{-Age/{\beta }_{5}})+{\beta }_{0}+{\beta }_{1}Age+\varepsilon$$
5. 5.

Quadratic hybrid model: exponential cumulative distribution for growth with quadratic model for aging

$$Vol={\beta }_{4}.(1-{e}^{-Age/{\beta }_{5}})+{\beta }_{0}+{\beta }_{1}Age+{\beta }_{2}Ag{e}^{2}+\varepsilon$$
6. 6.

Cubic hybrid model: exponential cumulative distribution for growth with cubic model for aging

$$Vol={\beta }_{4}.(1-{e}^{-Age/{\beta }_{5}})+{\beta }_{0}+{\beta }_{1}Age+{\beta }_{2}Ag{e}^{2}+{\beta }_{3}Ag{e}^{3}+\varepsilon$$

To select the best model type, we used the Bayesian Information Criterion among kept candidate models – p < 0.05 for ANOVA of the model vs. constant model and p < 0.05 for T-test of all the coefficients. The Bayesian information criterion is a measure providing a trade-off between bias and variance to select the model explaining most of the data with a minimum number of parameters. Moreover, to compensate for variability introduced by head size difference, models were estimated on normalized volume in % of total intracranial volume. Left and right volumes were added to obtain the final volume structure. The prediction bounds were estimated with a confidence level at 95%. This model selection procedure was applied to all the considered structures. In this study, we studied the following brain structures: lateral ventricles, hippocampus, amygdala, caudate, putamen, accumbens, globus pallidus and thalamus. Moreover, tissue classification was used to obtain the global volume of white matter and gray matter. All statistical tests were performed with Matlab© using default parameters. Afterwards, percentage of relative rate of change per year and percentage of abnormality were computed on the estimated models. The relative rate of change in percentage per year was computed as the first derivative of the model divided by the model39 and the abnormality in percentage as the absolute difference between pathological models and control model divided by control model (i.e., absolute relative difference compared to control).

Finally, we studied lifespan classification accuracy of several AD biomarkers. To this end, for each age, a classification was performed with a linear discriminate analysis (LDA) using all the samples in an interval of 10 y (i.e.,+/− 5 years). We used Area Under the Curve (AUC) as classification performance metric. The AUC was estimated through a cross-validation procedure based on a repeated K-fold using 10 iterations and 10-fold. Finally, the average AUC obtained over the 10 repetitions is reported for all ages. Classification experiments were performed with Matlab© using default parameters.

## Results

Figure 1 presents models of all considered structures for AD/MCI and CN groups. This figure shows that hippocampus and amygdala models present marked divergences between AD/MCI and CN, and also indicates that this divergence increases with age. Moreover, the divergence of control and pathological models for these structures occurs early around 40–45 y. Lateral ventricles also exhibit early divergence – starting around 42 y – between both models, however the distance between models decreases at advanced ages. Similarly, the thalamus presents an early but weak divergence that decreases at advanced ages. Pathological models of caudate and accumbens nuclei exhibit accelerated volume decreases from 50–60 y. However, confidence intervals for these structures overlap again after 85 y (see Table 3). For white matter and gray matter, AD/MCI models present an early accelerated aging compared to CN models around 45 y. However, after 80 y, CN models of brain tissues show an accelerated volume decreases. Consequently, confidence intervals of pathological and normal models overlap after 85 years (see Table 3). Finally, normal and pathological models for globus pallidus and putamen present similar trends.

Table 3 shows the age ranges where the confidence interval of the predicted pathological models (i.e., AD, MCI and AD/MCI) do not overlap with the confidence interval of the control models.

First, only hippocampus and amygdala models present non-overlapping confidence intervals after divergence for all the studied pathological modes (i.e., AD/MCI, AD and MCI) (see Table 3). This is also valid for lateral ventricles model but only when using the AD group. For all other considered structures, predicted confidence intervals overlap again at advanced ages around 80–90 y.

Second, hippocampus is the deep gray structure showing the earliest model divergence at 39 y for AD/MCI, 37 y for AD and 42 y for MCI. Then, the lateral ventricles models exhibit a divergence at 42 y for AD/MCI, 39 y for AD and 45 y for MCI. Afterwards, thalamus models diverge from control at 43 y for AD/MCI, 42 y for AD and 45 y for MCI. The divergence of amygdala models occurs at 44 y for AD/MCI, 41 y for AD and 49 y for MCI. Impact on global gray matter and white matter volume is observed later, with models diverging at 45 y and 48 y respectively for AD/MCI, at 46 y and 47 y respectively for AD and at 58 y and 54 y respectively for MCI. Finally, accumbens and caudate models diverge slightly later, but in a similar age range. Putamen and globus pallidus are the only deep gray matter structures for which models do not diverge from CN across the entire lifespan. Indeed, as assessed in Fig. 1, the confidence intervals of normal and pathological models always overlap.

To further analyze well-known AD biomarkers, we propose a second analysis focusing on the hippocampus, the lateral ventricles and the amygdala. Figure 2 presents the lifespan models of these structures for CN, AD and MCI groups. Moreover, relative rate of change and abnormality percentages are provided.

First, divergence points of the models occur earlier for AD than for MCI (see Table 3 for exact time). As expected, models for MCI are in between AD and CN ones. Second, when using relative rate of change, amygdala and lateral ventricle exhibit a more pronounced relative rate of change compared to hippocampus. The maximum relative rate of changes for AD models of these structures are −3.6%/y for AG at 96 y, −2.1%/y for hippocampus at 96 y and 3.4%/y at 42 y for lateral ventricles. Contrary to hippocampus and amygdala, which show an increasing relative rate of change with age, lateral ventricles exhibits enlargement following an inverted U-shape. When considering abnormality percentage, an earlier abnormality increase is observed for hippocampus than for lateral ventricles and amygdala. This abnormality reaches a maximum of 32% for the AD model at 96 y. Abnormality appears later in life for lateral ventricles and amygdala and follows very different patterns for both. The lateral ventricles abnormality follows an inverted U-shape with a maximum of 47% at 63 y for the AD model. The amygdala abnormality has similar trend to that of the hippocampus abnormality. Amygdala volume reaches 40% of abnormality at 96 y for the AD model. Therefore, while hippocampus abnormality starts first, amygdala presents a greater abnormality at advanced age. Moreover, the abnormality observed in lateral ventricles is also important but its maximum is reached at 65 y. Afterwards, percentage of abnormality of lateral ventricles decreases to end at 19% at 96 y for the AD model. Therefore, at late age, the lateral ventricles show lower abnormalities than those of the hippocampus and the amygdala at the same ages.

Finally, we propose a third analysis to estimate classification performance of the studied AD biomarkers according to the period of life. In the past, a large number of studies has been dedicated to AD patient classification for computer-aided diagnosis purpose5,40,41,42,43,44,45. In such studies, the classification performance of different biomarkers/features is compared to estimate their capability to distinguish patients from control subjects46,47,48. However, the evolution of the classification performance according to the period of life has not been studied. Therefore, we propose to estimate classification accuracy of hippocampus, lateral ventricles and amygdala volumes across the lifespan. Figure 3 presents the evolution of AUC across lifespan.

First, we can notice that the evolution of classification performance based on hippocampus and amygdala volumes are similar for the AD population with a plateau around 89% of AUC starting at 72 y. In contrast, for the MCI population, AUC trajectory of both regions does not behave similarly. Indeed, AUC increases faster and earlier for hippocampus volume than for amygdala volume. However, at the end, AUC trajectory for both regions converge to similar maximum around 73% at 80 y for hippocampus and around 72% at 80 y for amygdala. These AUC values are in line with literature dedicated to automatic classification of patients40,45. Finally, AUC based on lateral ventricles follows an inverted U-shape for AD population with a maximum of 76% around 65 y. For the MCI population, AUC shows a plateau around 60% from 60 y to 80 y.

## Discussion

In this paper, we investigate lifespan changes of the human brain in Alzheimer’s disease. The main novelty of this work is to propose pathological trajectory of brain structure over the entire lifespan. Consequently, it is possible to estimate when our normal and pathological models diverge. Moreover, our models enable to follow the dynamic of biomarker abnormality over the entire evolution of the pathology.

Our lifespan analysis using inferred models of brain trajectory in AD indicates that the hippocampus is the brain structure that exhibits the earliest divergence between cognitively normal model and pathological model. This model divergence is detectable early in life, at 39 y for the AD/MCI and at 37 y for the AD. The hippocampus model shares similar trends with another temporal lobe region, the amygdala, which presents model divergence at 44 y for AD/MCI and at 40 y for AD. It is noticeable that amygdala model is undergoing larger changes proportionally to its size compared to hippocampus. Finally, the lateral ventricles model presents an early divergence at 42 y for AD/MCI and at 39 y for AD. However, lateral ventricles enlargement occurring during normal aging reduces the abnormality of this structure after 60 y. Finally, the thalamus model shows early divergence at 43 y for AD/MCI and at 42 y for AD.

Our results presenting the hippocampus model as the first brain region diverging from normal aging model is in accordance with previous morphometric studies focused on the prodromal phase of the disease12,13,14,15,49,50. It is also in accordance with histopathological studies showing the temporal lobe as the starting point of the neurodegenerative process in AD51. In long follow-up studies, authors observed that incident cases of AD present morphometric difference in the hippocampus at least 10 years before the diagnosis13,14,21. Hippocampal atrophy has also been reported in several different transgenic mouse models of AD52, such as the amyloid precursor protein/presenilin2/Tau model, in which the volume reduction progresses with the pathology53, and it is associated with an enlargement of the lateral ventricles. In vivo MRI-measure of grey matter atrophy in AD has been described as a surrogate of the amount of neurofibrillary tangle revealed by post-mortem immunohistochemistry in human54. As this observation has been done in aging AD-diagnosed subjects, we could not ascertain that the mid-life atrophy revealed by our model is the result of the same Tau-pathophysiological process. Nevertheless, some recent studies revealed that temporal grey matter atrophy is associated to temporal Tau accumulation55,56 in clinically normal subjects. Moreover, a part of the effects of Tau accumulation on cognition is mediated by the temporal grey matter atrophy. On the other side, amyloid burden would be an independent aggravating factor of the process impacting temporal lobe regions, with an additive effect on cognition57. In fact, amyloid deposition in healthy elderly subjects, measured by the Pittsburgh compound B positron emission tomography, has been shown in a longitudinal study to be associated with later hippocampal atrophy and memory impairment58. All these results indicate that AD results from the cooccurrence of different pathological process, partially independent from each other and presenting different time course depending on the interplay between genetic and environmental factors. The time relationship between the different neuroimaging biomarkers is currently the subject of intensive research to describe early pathophysiological process associated to the disease. In addition to the atrophy and the Tauopathy, AD brain presents amyloid deposition. During the asymptomatic phase of the disease these biomarkers could be concomitantly or separately observed in vivo, leading to a proposed Amyloid/Tau/Neurodegeneration classification scheme to describe subjects during this silent phase59.

Based on our study, the volume of the lateral ventricles is also an early biomarker of AD, since its model diverges at 42 y for AD/MCI and at 39 y for AD. The potential of using lateral ventricles volume as AD biomarkers has been previously mentioned over restricted periods27,71. In this study, by analyzing the lifespan evolution of lateral ventricles abnormality, we showed that lateral ventricles abnormality decreases after 65 y. Therefore, the use of this biomarker is difficult for the late onset cases due to important lateral ventricles enlargement occurring during normal aging. However, it may be useful to discriminate cases around 65 y, an early age at which the AD diagnosis is particularly relevant because future potential treatment could be more effective in the early phases of the disease72. The importance of taking into account volume increase at advanced age in normal aging has been previously mentioned26. In the present study, early divergence of the models has been also observed for thalamus around 42 y. Thalamic atrophy was previously reported in AD literature73,74. However, we found that thalamus abnormality was very small (4.6% at 81 y). This may explain why only a small number of studies have mentioned that this structure could be affected by AD, because a large number of subjects are needed to detect such subtle atrophy.

In this study, to overcome the absence of longitudinal datasets with several decades of follow-up, we processed a massive number of cross-sectional MRI to generate a model of the volume trajectory in several brain structures in AD across the entire lifespan. We acknowledge that the use of cross-sectional data to analyze a dynamic process is not optimal. However, previous studies demonstrated that cross-sectional and longitudinal approaches produce similar age-related patterns in normal aging75 and similar models in AD76. Compared to previous longitudinal investigations, our results based on cross-sectional models are consistent with most previous findings on the importance and timing of hippocampus, amygdala and lateral ventricles alterations27,77,78. Although different in nature, it can be interesting to compare relative rates of change obtained with our models and annual rates of atrophy estimated on longitudinal data. The obtained values for relative rates of change are at the lower bound of expected range of annual atrophy rate reported in previous longitudinal literature77,79,80. However, our relative rates of change fits very well with the longitudinal rate of atrophy recently estimated with a multi-atlas method known to provide a less biased estimation81.

With respect to the estimated point of divergence between the CN and AD models, there is no longitudinal or cross-sectional MRI-based study literature over the lifetime with which to compare. However, our estimated point of divergence between models can be put into perspective with the detection of first AD signs in long follow-up longitudinal studies. First, prospective and longitudinal studies dedicated to autosomal dominant AD detected hippocampal atrophy up to 15 years before symptom onset10. Moreover, long follow-up population-based studies tracking cognition estimated that declines start from few years up to several decades before AD diagnosis82,83,84,85. Therefore, these results seem to confirm the presence of a long-lasting period of silence in AD, as discussed in86. Our models indicate that the age of 40 y is a critical period in the onset of the temporal lobe alteration. Midlife lifestyle factors such as diet20, physical activity87 and exposure to risk factors88 are associated with the risk of developing late life dementia; the life style factors of this critical period may specifically impact hippocampus leading to an increase vulnerability to dementia19. Consequently, exposure to risk factors (such as diabetes and smoking) occurring at this lifetime period should be considered in future studies to evaluate their implication in the early atrophy process.

We also proposed a classification analysis in order to highlight the importance of considering lifespan trajectory when evaluating biomarker performance. While most of the studies dedicated to automatic patient classification present only a global accuracy, our experience shows that biomarker efficiency changes according to the period of life. Therefore, computer-aided diagnosis study should test the accuracy of the biomarker at different time points.

Finally, our pathological model present three main limitations. First, we used normal control for the period [0–55] years. However, there is no alternative to simulate the AD model before 55 years. Second, by mixing AD and MCI subjects in the same model, we included false positive subjects in the pathological model since we do know that only a part of the MCI subjects will develop AD. However, we have also analyzed the AD and MCI models separately for the most relevant structures. Finally, in our model we consider that brain alteration caused by AD is a progressive and smooth phenomenon. The transition from CN to AD over the lifespan is viewed as a continuum. Such approach is not well-suited if brain modifications occurs suddenly. However, evidences such as slow accumulation of Amyloid-β3,22,23 or smooth and progressive atrophy of brain3,10 seem to assess that brain alterations occurs smoothly and progressively in AD.

## Conclusion

In this work, to analyze when AD brain model diverges from the cognitively normal model we build extrapolated lifespan models of AD brain structures by combining multiple large-scale databases. We found an early divergence of AD model from control model for hippocampus before 40 y followed by lateral ventricles and amygdala around 40 y. Moreover, we observed a similar abnormality evolution for the hippocampus and the amygdala. Finally, the model of ventricular enlargement shows that the volume of lateral ventricles reaches a peak of abnormality at 65 y before decreasing due to important enlargement in normal aging.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## References

1. 1.

Lobo, A. et al. Prevalence of dementia and major subtypes in Europe: a collaborative study of population-based cohorts. Neurology 54, S4 (2000).

2. 2.

Barnes, J. et al. Alzheimer’s disease first symptoms are age dependent: evidence from the NACC dataset. Alzheimer’s & dementia 11, 1349–1357 (2015).

3. 3.

Jack, C. R. et al. Tracking pathophysiological processes in Alzheimer’s disease: an updated hypothetical model of dynamic biomarkers. The Lancet Neurology 12, 207–216 (2013).

4. 4.

Nestor, P. J., Scheltens, P. & Hodges, J. R. Advances in the early detection of Alzheimer’s disease. Nature medicine 10 (2004).

5. 5.

Davatzikos, C., Fan, Y., Wu, X., Shen, D. & Resnick, S. M. Detection of prodromal Alzheimer’s disease via pattern classification of magnetic resonance imaging. Neurobiology of aging 29, 514–523 (2008).

6. 6.

Bakkour, A., Morris, J. C. & Dickerson, B. C. The cortical signature of prodromal AD Regional thinning predicts mild AD dementia. Neurology 72, 1048–1055 (2009).

7. 7.

Chan, D. et al. Change in rates of cerebral atrophy over time in early-onset Alzheimer’s disease: longitudinal MRI study. The Lancet 362, 1121–1122 (2003).

8. 8.

Ridha, B. H. et al. Tracking atrophy progression in familial Alzheimer’s disease: a serial MRI study. The Lancet Neurology 5, 828–834 (2006).

9. 9.

Sala-Llonch, R., Bartrés-Faz, D. & Junqué, C. Reorganization of brain networks in aging: a review of functional connectivity studies. Frontiers in psychology 6 (2015).

10. 10.

Bateman, R. J. et al. Clinical and biomarker changes in dominantly inherited Alzheimer’s disease. New England Journal of Medicine 367, 795–804 (2012).

11. 11.

Dickerson, B. et al. Alzheimer-signature MRI biomarker predicts AD dementia in cognitively normal adults. Neurology 76, 1395–1402 (2011).

12. 12.

Miller, M. I. et al. The diffeomorphometry of temporal lobe structures in preclinical Alzheimer’s disease. NeuroImage: Clinical 3, 352–360 (2013).

13. 13.

Bernard, C. et al. Time course of brain volume changes in the preclinical phase of Alzheimer’s disease. Alzheimer’s & Dementia 10, 143–151. e141 (2014).

14. 14.

den Heijer, T. et al. A 10-year follow-up of hippocampal volume on magnetic resonance imaging in early dementia and cognitive decline. Brain 133, 1163–1172 (2010).

15. 15.

Coupé, P. et al. Detection of Alzheimer’s disease signature in MR images seven years before conversion to dementia: Toward an early individual prognosis. Hum Brain Mapp 36, 4758–4770, https://doi.org/10.1002/hbm.22926 (2015).

16. 16.

Albert, M. et al. Predicting progression from normal cognition to mild cognitive impairment for individuals at 5 years. Brain (2018).

17. 17.

Poldrack, R. A. & Gorgolewski, K. J. Making big data open: data sharing in neuroimaging. Nature neuroscience 17, 1510–1517 (2014).

18. 18.

Solomon, A. et al. Serum cholesterol changes after midlife and late-life cognition twenty-one-year follow-up study. Neurology 68, 751–756 (2007).

19. 19.

Debette, S. et al. Midlife vascular risk factor exposure accelerates structural brain aging and cognitive decline. Neurology 77, 461–468 (2011).

20. 20.

Tolppanen, A.-M. et al. Midlife and late-life body mass index and late-life dementia: results from a prospective population-based cohort. Journal of Alzheimer’s Disease 38, 201–209 (2014).

21. 21.

Coupe, P., Catheline, G., Lanuza, E. & Manjon, J. V. & Alzheimer’s Disease Neuroimaging, I. Towards a unified analysis of brain maturation and aging across the entire lifespan: A MRI analysis. Hum Brain Mapp 38, 5501–5518, https://doi.org/10.1002/hbm.23743 (2017).

22. 22.

Villemagne, V. L. et al. Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer’s disease: a prospective cohort study. The Lancet Neurology 12, 357–367 %@1474–4422 (2013).

23. 23.

Villemagne, V. L. et al. Longitudinal assessment of Aβ and cognition in aging and Alzheimer disease. Annals of neurology 69, 181–192 (2011).

24. 24.

Poulin, S. P. et al. Amygdala atrophy is prominent in early Alzheimer’s disease and relates to symptom severity. Psychiatry Research: Neuroimaging 194, 7–13 (2011).

25. 25.

Jack, C. R. et al. Medial temporal atrophy on MRI in normal aging and very mild Alzheimer’s disease. Neurology 49, 786–794 (1997).

26. 26.

Apostolova, L. G. et al. Hippocampal atrophy and ventricular enlargement in normal aging, mild cognitive impairment and Alzheimer’s disease. Alzheimer disease and associated disorders 26, 17 (2012).

27. 27.

Nestor, S. M. et al. Ventricular enlargement as a possible measure of Alzheimer’s disease progression validated using the Alzheimer’s disease neuroimaging initiative database. Brain 131, 2443–2454 (2008).

28. 28.

Petersen, R. C. et al. Alzheimer’s disease Neuroimaging Initiative (ADNI) clinical characterization. Neurology 74, 201–209 (2010).

29. 29.

Marcus, D. S. et al. Open Access Series of Imaging Studies (OASIS): cross-sectional MRI data in young, middle aged, nondemented, and demented older adults. Journal of cognitive neuroscience 19, 1498–1507 (2007).

30. 30.

Manjon, J. V. & Coupe, P. volBrain: An Online MRI Brain Volumetry System. Front Neuroinform 10, 30, https://doi.org/10.3389/fninf.2016.00030 (2016).

31. 31.

Manjon, J. V., Coupe, P., Marti-Bonmati, L., Collins, D. L. & Robles, M. Adaptive non-local means denoising of MR images with spatially varying noise levels. J Magn Reson Imaging 31, 192–203, https://doi.org/10.1002/jmri.22003 (2010).

32. 32.

Tustison, N. J. et al. N4ITK: improved N3 bias correction. IEEE Trans Med Imaging 29, 1310–1320, https://doi.org/10.1109/TMI.2010.2046908 (2010).

33. 33.

Avants, B. B. et al. A reproducible evaluation of ANTs similarity metric performance in brain image registration. Neuroimage 54, 2033–2044 (2011).

34. 34.

Ashburner, J. & Friston, K. J. Unified segmentation. Neuroimage 26, 839–851, https://doi.org/10.1016/j.neuroimage.2005.02.018 (2005).

35. 35.

Manjón, J. V., Tohka, J. & Robles, M. Improved estimates of partial volume coefficients from noisy brain MRI using spatial context. Neuroimage 53, 480–490 (2010).

36. 36.

Manjon, J. V. et al. Nonlocal intracranial cavity extraction. Int J Biomed Imaging 2014, 820205, https://doi.org/10.1155/2014/820205 (2014).

37. 37.

Coupe, P. et al. Patch-based segmentation using expert priors: application to hippocampus and ventricle segmentation. Neuroimage 54, 940–954, https://doi.org/10.1016/j.neuroimage.2010.09.018 (2011).

38. 38.

Frisoni, G. B. et al. The EADC-ADNI Harmonized Protocol for manual hippocampal segmentation on magnetic resonance: evidence of validity. Alzheimer’s & Dementia 11, 111–125 (2015).

39. 39.

Solow, R. M. A contribution to the theory of economic growth. The quarterly journal of economics 70, 65–94 %@1531–4650 (1956).

40. 40.

Coupe, P. et al. Scoring by nonlocal image patch estimator for early detection of Alzheimer’s disease. Neuroimage Clin 1, 141–152, https://doi.org/10.1016/j.nicl.2012.10.002 (2012).

41. 41.

Cuingnet, R. et al. Automatic classification of patients with Alzheimer’s disease from structural MRI: a comparison of ten methods using the ADNI database. Neuroimage 56, 766–781, https://doi.org/10.1016/j.neuroimage.2010.06.013 (2011).

42. 42.

Eskildsen, S. F. et al. Prediction of Alzheimer’s disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. Neuroimage 65, 511–521 (2013).

43. 43.

Eskildsen, S. F. et al. Structural imaging biomarkers of Alzheimer’s disease: predicting disease progression. Neurobiology of aging 36, S23–S31 (2015).

44. 44.

Tong, T. et al. A Novel Grading Biomarker for the Prediction of Conversion From Mild Cognitive Impairment to Alzheimer’s Disease. IEEE Transactions on Biomedical Engineering 64, 155–165 (2017).

45. 45.

Wolz, R. et al. Multi-method analysis of MRI images in early diagnostics of Alzheimer’s disease. PLoS One 6, e25446, https://doi.org/10.1371/journal.pone.0025446 (2011).

46. 46.

Bron, E. E. et al. Standardized evaluation of algorithms for computer-aided diagnosis of dementia based on structural MRI: the CADDementia challenge. Neuroimage 111, 562–579, https://doi.org/10.1016/j.neuroimage.2015.01.048 (2015).

47. 47.

Chaddad, A., Desrosiers, C., Hassan, L. & Tanougast, C. Hippocampus and amygdala radiomic biomarkers for the study of autism spectrum disorder. BMC Neurosci 18, 52, https://doi.org/10.1186/s12868-017-0373-0 (2017).

48. 48.

Chaddad, A., Desrosiers, C. & Toews, M. Multi-scale radiomic analysis of sub-cortical regions in MRI related to autism, gender and age. Sci Rep 7, 45639, https://doi.org/10.1038/srep45639 (2017).

49. 49.

Apostolova, L. G. et al. Subregional hippocampal atrophy predicts Alzheimer’s dementia in the cognitively normal. Neurobiology of aging 31, 1077–1088 (2010).

50. 50.

Younes, L., Albert, M., Miller, M. I. & Team, B. R. Inferring changepoint times of medial temporal lobe morphometric change in preclinical Alzheimer’s disease. NeuroImage: Clinical 5, 178–187 (2014).

51. 51.

Braak, H. & Braak, E. Neuropathological stageing of Alzheimer-related changes. Acta neuropathologica 82, 239–259 (1991).

52. 52.

Badea, A. et al. The fornix provides multiple biomarkers to characterize circuit disruption in a mouse model of Alzheimer’s disease. NeuroImage 142, 498–511 (2016).

53. 53.

Micotti, E. et al. Striatum and entorhinal cortex atrophy in AD mouse models: MRI comprehensive analysis. Neurobiology of aging 36, 776–788 (2015).

54. 54.

Whitwell, J. L. et al. MRI correlates of neurofibrillary tangle pathology at autopsy A voxel-based morphometry study. Neurology 71, 743–749 (2008).

55. 55.

Iaccarino, L. et al. Local and distant relationships between amyloid, tau and neurodegeneration in Alzheimer’s Disease. NeuroImage: Clinical 17, 452–464 (2018).

56. 56.

Das, S. R. et al. Longitudinal and cross-sectional structural magnetic resonance imaging correlates of AV-1451 uptake. Neurobiology of aging 66, 49–58 (2018).

57. 57.

Knopman, D. S. et al. Joint associations of β-amyloidosis and cortical thickness with cognition. Neurobiology of aging 65, 121–131 (2018).

58. 58.

Doré, V. et al. Cross-sectional and longitudinal analysis of the relationship between Aβ deposition, cortical thickness, and memory in cognitively unimpaired individuals and in Alzheimer disease. JAMA neurology 70, 903–911 (2013).

59. 59.

Jack, C. R. et al. A/T/N: an unbiased descriptive classification scheme for Alzheimer disease biomarkers. Neurology 87, 539–547 (2016).

60. 60.

Cavedo, E. et al. Local amygdala structural differences with 3T MRI in patients with Alzheimer disease. Neurology 76, 727–733 (2011).

61. 61.

Qiu, A., Fennema-Notestine, C., Dale, A. M., Miller, M. I. & Alzheimer’s Disease Neuroimaging, I. Regional shape abnormalities in mild cognitive impairment and Alzheimer’s disease. Neuroimage 45, 656–661 (2009).

62. 62.

Lin, T.-W. et al. Neurodegeneration in amygdala precedes hippocampus in the APPswe/PS1dE9 mouse model of Alzheimer’s disease. Current Alzheimer Research 12, 951–963 (2015).

63. 63.

Phelps, E. A. Human emotion and memory: interactions of the amygdala and hippocampal complex. Current opinion in neurobiology 14, 198–202 (2004).

64. 64.

Kumfor, F. et al. Degradation of emotion processing ability in corticobasal syndrome and Alzheimer’s disease. Brain 137, 3061–3072 (2014).

65. 65.

De Olmos, J. S. In The Human Nervous System (Second Edition) Ch. 22, 739–868 (2004).

66. 66.

Tabert, M. H. et al. A 10‐item smell identification scale related to risk for Alzheimer’s disease. Annals of neurology 58, 155–160 (2005).

67. 67.

Serby, M., Larson, P. & Kalkstein, D. The nature and course of olfactory deficits in Alzheimer’s disease. The American journal of psychiatry 148, 357 (1991).

68. 68.

Djordjevic, J., Jones-Gotman, M., De Sousa, K. & Chertkow, H. Olfaction in patients with mild cognitive impairment and Alzheimer’s disease. Neurobiology of aging 29, 693–706 (2008).

69. 69.

Price, J. L., Davis, P., Morris, J. & White, D. The distribution of tangles, plaques and related immunohistochemical markers in healthy aging and Alzheimer’s disease. Neurobiology of aging 12, 295–312 (1991).

70. 70.

Ohm, T. & Braak, H. Olfactory bulb changes in Alzheimer’s disease. Acta neuropathologica 73, 365–369 (1987).

71. 71.

Carmichael, O. T. et al. Cerebral ventricular changes associated with transitions between normal cognitive function, mild cognitive impairment, and dementia. Alzheimer disease and associated disorders 21, 14 (2007).

72. 72.

Prince, M., Bryce, R. & Ferri, C. World Alzheimer Report 2011: The benefits of early diagnosis and intervention. (Alzheimer’s Disease International, 2011).

73. 73.

De Jong, L. W. et al. Strongly reduced volumes of putamen and thalamus in Alzheimer’s disease: an MRI study. Brain 131, 3277–3285 (2008).

74. 74.

Braak, H. & Braak, E. Alzheimer’s disease affects limbic nuclei of the thalamus. Acta neuropathologica 81, 261–268 (1991).

75. 75.

Fjell, A. M. et al. Critical ages in the life course of the adult brain: nonlinear subcortical aging. Neurobiol Aging 34, 2239–2247, https://doi.org/10.1016/j.neurobiolaging.2013.04.006 (2013).

76. 76.

Fotenos, A. F., Snyder, A. Z., Girton, L. E., Morris, J. C. & Buckner, R. L. Normative estimates of cross-sectional and longitudinal brain volume decline in aging and AD. Neurology 64, 1032–1039 (2005).

77. 77.

Fjell, A. M. et al. One-year brain atrophy evident in healthy aging. Journal of Neuroscience 29, 15223–15231 (2009).

78. 78.

Jack, C. R. et al. Comparison of different MRI brain atrophy rate measures with clinical disease progression in AD. Neurology 62, 591–600 (2004).

79. 79.

Barnes, J. et al. A meta-analysis of hippocampal atrophy rates in Alzheimer’s disease. Neurobiology of aging 30, 1711–1723 (2009).

80. 80.

McDonald, C. R. et al. Regional rates of neocortical atrophy from normal aging to early Alzheimer disease. Neurology 73, 457–465 (2009).

81. 81.

Sankar, T. et al. Your algorithm might think the hippocampus grows in Alzheimer’s disease: Caveats of longitudinal automated hippocampal volumetry. Human Brain Mapping 38, 2875–2896 (2017).

82. 82.

Small, B. J., Fratiglioni, L., Viitanen, M., Winblad, B. & Bäckman, L. The course of cognitive impairment in preclinical Alzheimer disease: three-and 6-year follow-up of a population-based sample. Archives of neurology 57, 839–844 (2000).

83. 83.

La Rue, A. & Jarvik, L. F. Cognitive function and prediction of dementia in old age. The International Journal of Aging and Human Development 25, 79–89 (1987).

84. 84.

Elias, M. F. et al. The preclinical phase of Alzheimer disease: a 22-year prospective study of the Framingham Cohort. Archives of neurology 57, 808–813 (2000).

85. 85.

Snowdon, D. A. et al. Linguistic ability in early life and cognitive function and Alzheimer’s disease in late life: Findings from the Nun Study. Jama 275, 528–532 (1996).

86. 86.

Dubois, B. et al. Preclinical Alzheimer’s disease: definition, natural history, and diagnostic criteria. Alzheimer’s & Dementia 12, 292–323 (2016).

87. 87.

Krell-Roesch, J. et al. Leisure-Time Physical Activity and the Risk of IncidentDementia: The Mayo Clinic Study of Aging. Journal of Alzheimer’s Disease, 1–7 (2018).

88. 88.

Rusanen, M., Kivipelto, M., Quesenberry, C. P., Zhou, J. & Whitmer, R. A. Heavy smoking in midlife and long-term risk of Alzheimer disease and vascular dementia. Archives of internal medicine 171, 333–339 (2011).

## Author information

### Affiliations

1. #### University Bordeaux, LaBRI, UMR 5800, PICTURA, F-33400, Talence, France

• Pierrick Coupé
2. #### CNRS, LaBRI, UMR 5800, PICTURA, F-33400, Talence, France

• Pierrick Coupé
3. #### Instituto Universitario de Tecnologías de la Información y Comunicaciones (ITACA), Universitat Politècnica de València, Camino de Vera s/n, 46022, Valencia, Spain

• José Vicente Manjón
4. #### University Valencia, Department of Cell Biology, Burjassot, 46100, Valencia, Spain

• Enrique Lanuza
5. #### University Bordeaux, CNRS, EPHE, PSL, INCIA, UMR 5283, F-33000, Bordeaux, France

• Gwenaelle Catheline

### Contributions

All the authors conceived the original idea of this paper and planned the experiments. P.C. and J.M. developed the image processing pipeline, performed the processing and the statistical analysis. E.L. and G.C. contributed to the interpretation of the results. All authors wrote the manuscript, discussed the results, provided critical feedback and helped shape the research and analysis.

### Competing Interests

The authors declare no competing interests.

### Corresponding author

Correspondence to Pierrick Coupé.

### DOI

https://doi.org/10.1038/s41598-019-39809-8