Scientists are becoming more nuanced in how they use amyloid scans—not just to detect the presence of Alzheimer’s pathology, but also to pinpoint disease stage. At this year’s Alzheimer’s Association International Conference, held July 13–18 in Los Angeles, researchers led by Niklas Mattsson and Oskar Hansson at Lund University, Sweden, debuted a new staging scheme. Using longitudinal data from 741 participants in the Alzheimer’s Disease Neuroimaging Initiative, including cerebrospinal fluid Aβ42 as well as PET, the researchers defined four stages of amyloid accumulation. People at stage zero had a low risk of developing plaques, but those at higher stages were likely to move to the next-higher stage within a few years, suggesting the system reflects disease progression.

  • Using longitudinal PET and CSF, scientists define four stages of amyloid accumulation.
  • In stage 1, plaques form in the precuneus and posterior cingulate. Global PET scans turn positive at stage 2 or 3.
  • The scheme could help identify people with very early AD for research.

Their staging system differs notably from previous models derived from cross-sectional imaging and neuropathology data. It flags amyloid accumulation in the precuneus and posterior cingulate as the earliest signs of AD, long before the brain’s overall amyloid burden becomes positive on a PET scan. “If you wanted to run a very early AD trial, you could use this staging system to select participants, rather than a global amyloid cut point,” Mattsson told Alzforum.

Others agree. “This elegant approach provides critical information on how we can discern individuals in the amyloid-negative spectrum who are likely to progress to preclinical Alzheimer’s disease,” Heidi Jacobs at Massachusetts General Hospital, Boston, wrote to Alzforum (full comment below). Arthur Toga at the University of Southern California, Los Angeles, noted, “The approach could have significant utility for tracking disease progression in a clinical setting.” Mattsson and colleagues described the scheme in the July 17 JAMA Neurology.

Amyloid Staging. Longitudinal PET scans reveal regions of early (green), intermediate (blue), and late (red) amyloid accumulation. [Courtesy of Mattsson et al., ©2018 American Medical Association.]

To develop their scheme, the Lund group made use of their previous finding that CSF Aβ42 drops up to 10 years before an amyloid PET scan crosses the threshold for global positivity (Aug 2016 conference news; Palmqvist et al., 2016). These data suggested that CSF-positive people were accumulating amyloid in select brain regions. The researchers wondered if they could identify those regions of early buildup. In their initial study, CSF-positive yet whole-brain PET-negative ADNI participants indeed deposited amyloid only in specific regions, including the precuneus and posterior cingulate (Nov 2017 news).

Mattsson and colleagues extended those findings to develop a longitudinal staging system. First, they selected 641 ADNI participants who had CSF data and at least two florbetapir PET scans, and stratified them by amyloid positivity. Among this group, 288 were negative on both CSF and PET and were classified as non-accumulators; 69 were CSF-positive but PET-negative and were deemed early-stage accumulators; and 274 were positive on both—the late-stage accumulators. Ten discordant people were CSF-negative and PET-positive.

The researchers then examined longitudinal amyloid scans from the 69 early stage accumulators. They found six brain regions—the precuneus, posterior and isthmus cingulate, insula, and medial and lateral orbitofrontal cortices—where amyloid load was increasing compared to non-accumulators. Amyloid positivity in these regions, which form part of the brain’s default mode network (DMN), defined stage 1. Next, accumulation of amyloid in a large number of regions, including the parahippocampus, medial and inferior temporal lobes, inferior parietal lobe, and superior parietal, temporal, and frontal regions, marked stage 2. Most of these regions are known sites of pathology in early AD. Finally, in late-stage accumulators, amyloid piled up in precentral, postcentral, paracentral, lingual, and pericalcarine cortices. Amyloid in these regions defined stage 3 (see image above).

How did the ADNI cohort break down across these stages? For this, the researchers included another 100 ADNI volunteers who lacked CSF data but had at least two florbetapir scans, for a total of 741 participants. The full cohort comprised 304 cognitively healthy controls, 384 people with MCI, and 53 with AD dementia. Ninety-eight percent of participants fell cleanly into one of the four stages. More than half were at stage zero, three percent at stage 1, 11 percent at stage 2, and 30 percent at stage 3. The stages roughly corresponded to cognitive status, with 70 percent of controls at stage zero and 80 percent of AD patients at stage 3. Still, there were plenty of exceptions: 16 percent of controls were at stage 3, and 17 percent of AD patients were at stage zero.

The authors repeated the analysis with data from the Swedish BioFINDER cohort. This longitudinal biomarker study uses flutemetamol rather than florbetapir for amyloid scans. In a cross-sectional set of 306 healthy controls and 168 people with MCI, 98 percent of them fit unambiguously into one of the four stages. The percentages for each stage and cognitive group were similar to those in ADNI.

Notably, the staging patterns closely matched those seen in longitudinal PET scans from the Dominantly Inherited Alzheimer Network, where deposition occurred first in the precuneus, then in the posterior cingulate and medial orbitofrontal cortex (Feb 2018 news). 

They also agree with other longitudinal PET data presented at AAIC. Michelle Farrell of Massachusetts General Hospital reported that among 265 cognitively healthy adults in the Harvard Aging Brain Study who had repeated PET scans, amyloid accumulated earliest in the precuneus, isthmus and anterior cingulate cortex, medial orbitofronal cortex, and middle and inferior temporal lobe. Likewise, Gemma Salvadó of Barcelonaβeta Brain Research Center in Barcelona, Spain, showed congruent data from the European Amyloid Imaging to Prevent Alzheimer’s Disease study. Her team developed a staging model based on PET data from 224 cognitively healthy participants in Barcelonaβeta’s Alzheimer and Families (ALFA) project, then applied it to 870 PET images from the ALFA, ADNI, ABIDE, and EMIF-AD cohorts. The first areas to accumulate plaques were the precuneus, anterior cingulate cortex, and orbitofrontal cortex.

On the other hand, the Lund group’s findings only partially overlap with cross-sectional staging schemes, including classic neuropathology data and a recent PET amyloid study from Michel Grothe and colleagues at the German Center for Neurodegenerative Diseases in Rostock (Braak and Braak, 1991; Oct 2017 news). Grothe detected the earliest amyloid deposition in neocortical regions such as the temporal lobe, parietal operculum, and anterior cingulate, but did not pick out precuneus and posterior cingulate as early sites.

“The strength of the Mattsson et al. approach is that they utilize longitudinal PET data, which can give a more dynamic picture than static states do,” Rachel Buckley at MGH wrote to Alzforum (full comment below).

Is this new staging scheme biologically meaningful? Several pieces of evidence argue that it is, Mattsson said. For one thing, the amyloid PET stages correlated with other biomarkers. People at stage 1 or higher had low CSF Aβ42 and high phospho-tau compared with controls. In those at stage 2 or higher, CSF total tau ramped up, while at stage 3 brain atrophy did. Cognitive decline began in stage 2.

Another indication that the staging system is valid is that it predicted progression. People at stage zero had a 15 percent risk of progressing to a higher stage over an average of four years, while those in stage 1 ran a 71 percent risk and stage 2, a 53 percent risk. This fast rate of progression may explain why relatively few people are found in stage 1 or 2 at any given time, the authors speculated.

A final piece of evidence is that the combined brain regions that define each stage had distinct patterns of gene expression. Using data from the Allen Human Brain Atlas, the researchers found classes of genes that were differentially expressed among the regions. They were linked to voltage-gated ion channels, neuropeptide and glutamate signaling, lipid transport, and axon guidance. The data hinted at biological factors that may underlie the selective vulnerability to amyloid accumulation of brain regions associated with different stages, Hansson said.

“The relationship to regional gene expression areas was interesting, as this may hint at the ‘why’ [of amyloid accumulation] as well as the ‘where’,” Toga wrote to Alzforum.—Madolyn Bowman Rogers


  1. This is a very interesting paper that has attempted to base amyloid staging on both CSF and PET biomarkers, which is unlike the previous approaches (e.g. Grothe et al., 2017). In this current approach, Niklas Mattsson and colleagues focused on applying the previously proposed model of “CSF abnormality comes first, followed by PET abnormality,” to group individuals into “early” or “late” amyloid accumulators (with early accumulators defined by baseline CSF+/PET- and late defined by baseline CSF+/PET+). Once these baseline groupings were applied, longitudinal rates of accumulation in PET were harnessed to define those regions that were considered most salient in each of the stages. This is distinct from Michel Grothe’s previous work that used a standard approach applied in pathological staging, that is, identifying the most frequently “positive” regions in each stage.

    I think that the strength of the Mattsson approach is that they are utilizing the longitudinal PET data, which adds a more dynamic picture to these static states. There are many ways to approach staging models, however, and the decisions made will have an impact on the types of regions that appear as salient to each stage. Michelle Farrell in our group at MGH has been working on a similar approach in the Harvard Aging Brain Study, however, rather than grouping individuals via their baseline abnormality, she is approaching the staging model from a purely longitudinal perspective. Her findings, presented at AAIC this year, showed that the earliest regions that were affected were precuneus, isthmus CC, rostral ACC, medial OFC, and middle and inferior temporal lobe. Farrell was able to validate her staging model in a longitudinal study (ADNI), while the Mattsson team was only able to validate in the cross-sectional study (BIOFINDER). 

    One important thing to consider is for whom these staging models are being developed. Mattsson and colleagues have used all individuals, from cognitively healthy to AD, to define their staging approach. The advantage here is that a wide dynamic range in longitudinal change can be harnessed to identify early and late-accumulating regions. I do wonder, however, whether we should also be honing our earliest stages on a cognitively healthy group alone in this way. I wonder if it might be possible to get a more fine-grained “early accumulator” stage based on the earliest hints of change.

    Finally, I think an important caveat to consider here is the clinical meaningfulness of these staging approaches that utilize longitudinal PET. While it will be of great importance to understand the natural history of amyloid accumulation from a scientific perspective, it will be impossible to implement this on a clinical level. One then has to ask the next question: How much do we lose by only using cross-sectional data to create stages? And how much do longitudinal and cross-sectional approaches align with each other? We are now starting to gather more evidence to answer these questions.


    . In vivo staging of regional amyloid deposition. Neurology. 2017 Nov 14;89(20):2031-2038. Epub 2017 Oct 18 PubMed.

  2. This is fantastic work using a unique approach to stage amyloid by combining information from cerebrospinal fluid (CSF) and positron emission tomography (PET) data to place individuals in four groups: CSF-/PET- (non-accumulators); CSF+/PET- (early accumulators); CSF-PET+ (discordant); and CSF+/PET+ (late accumulators). As amyloid accumulation affects regions at different time points along the disease, the authors compared the spatiotemporal dynamics of regional PET data among these four groups resulting in three regional composites (early, intermediate, and late). Progression through the stages was overall concordant with the staging model and only a very low proportion of the individuals ever reverted to a lower stage. Associations with changes in CSF, atrophy, and cognition were congruent with the expected temporal evolution of biomarker changes, showing initial decreases in CSF Aβ, followed by increases in CSF phospho-tau and cognitive decline, and finally accelerated atrophy. Importantly, these findings, generated using Alzheimer’s Disease Neuroimaging Initiative data, were validated in the independent cross-sectional, BioFINDER study dataset, which uses a different PET tracer and a different assay.

    This approach differed from postmortem studies as well as recent staging work (Grothe et al., 2017), where only cross-sectional data was used and where no regionally varying cutoffs were applied. Utilizing regional rates of change is more powerful because both the individual- as well as the group-level data are taken into account and it allows sequential changes to be determined. It is interesting to discover that the regional patterns of amyloid progression in this study are analogous to a recent study examining spatial amyloid patterns in autosomal-dominant Alzheimer’s disease participants (ADAD) using longitudinal PET data (Gordon et al., 2018). Participants with ADAD can be positioned along the disease trajectory according to their estimated years of onset. This removes the need to acquire decades of biomarker data and provides a unique perspective into the temporal patterns of amyloid accumulation. It is promising to notice that the patterns in the Mattsson and Gordon studies are similar, given that the pathologic hallmarks of ADAD are similar to sporadic Alzheimer’s disease.

    The subcortical regions ended up in none of the stages. In the work by Grothe and colleagues, only a small proportion of individuals showed amyloid positivity in the striatum. Striatal amyloid has important predictive value in advanced stages (Hanseeuw et al., 2018). Not incorporating these regions into the staging scheme might be related to greater noise in these smaller regions that would confound rates of change.

    This work potentially has two significant implications for the field: This elegant approach provides critical information on what “amyloid negativity” could mean and how we can discern individuals in the amyloid-negative spectrum into those who are likely to progress to preclinical Alzheimer’s disease versus those who are less likely by combining CSF and regional PET information (Bischof and Jacobs, 2019). Participants in stages zero and 1 (early accumulators who were negative on the amyloid PET scan) exhibited at risk of 14.7 percent and 71.4 percent, respectively,  to progress to a higher stage. It would be interesting to examine the contribution of APOE-E4 carriers to these patterns of progression.

    The disappointing outcomes of several amyloid-focused trials indicated that we have to improve our ability to detect individuals in earlier stages who are likely to progress to accumulate significant amyloid. The results of this study convincingly demonstrate that we can gain sensitivity in predicting progression by using regional information instead of global composites (Farrell et al., 2018). This study again highlights the need for longitudinal data to understand and predict progression of amyloid pathology. In addition, these findings suggest that using a more granular approach of cross-sectional regional amyloid patterns could be promising to predict progression in the future, which will be valuable for clinical setting and can inform risk-specific enrolment strategies in clinical trials.


    . Subthreshold amyloid and its biological and clinical meaning: Long way ahead. Neurology. 2019 Jul 9;93(2):72-79. Epub 2019 Jun 5 PubMed.

    . Regional amyloid accumulation and cognitive decline in initially amyloid-negative adults. Neurology. 2018 Nov 6;91(19):e1809-e1821. Epub 2018 Oct 10 PubMed.

    . Spatial patterns of neuroimaging biomarker change in individuals from families with autosomal dominant Alzheimer's disease: a longitudinal study. Lancet Neurol. 2018 Mar;17(3):241-250. Epub 2018 Feb 1 PubMed.

    . In vivo staging of regional amyloid deposition. Neurology. 2017 Nov 14;89(20):2031-2038. Epub 2017 Oct 18 PubMed.

    . PET staging of amyloidosis using striatum. Alzheimers Dement. 2018 Oct;14(10):1281-1292. Epub 2018 May 21 PubMed.

    . Staging β-Amyloid Pathology With Amyloid Positron Emission Tomography. JAMA Neurol. 2019 Jul 17; PubMed.

  3. I read the paper and enjoyed it. Staging of AD using PET with both ligands for amyloid and tau has appeared in previous papers, as acknowledged by the authors. This has most convincingly been done in subjects with dominantly inherited AD where the clinical chronology of the disease is well established. This facilitates relating the known age of symptom onset to the imaging and CSF data. For the sporadic disease establishing this chronology is more difficult. Here the authors are able to establish the time course of amyloid plaque accumulation in a convincing manner. It helps solidify the relationship between the sporadic and inherited forms of the disease which extends to the transgenic models of the disease in rodents.

    The other thing that caught my attention was the striking anatomy of their “early” stage of amyloid deposition. Again, this topography has been noted before, but this is a most convincing demonstration. The question to be considered is, why this anatomy? What makes this topography (i.e., the brain’s default mode network) vulnerable? They offer some genetic information that focuses on the distribution of ion channels. Indulging in a bit of self-promotion, I would point out that this anatomy also houses a unique set of genes and a metabolic profile that are associated with plasticity (Goyal et al., 2014; Goyal et al., 2017) and is the least myelinated part of the human cerebral cortex (Glasser et al., 2014​).  All of these features deserve our attention as we seek to understand the root cause of AD.


    . Aerobic glycolysis in the human brain is associated with development and neotenous gene expression. Cell Metab. 2014 Jan 7;19(1):49-57. PubMed.

    . Loss of Brain Aerobic Glycolysis in Normal Human Aging. Cell Metab. 2017 Aug 1;26(2):353-360.e3. PubMed.

    . Trends and properties of human cerebral cortex: correlations with cortical myelin content. Neuroimage. 2014 Jun;93 Pt 2:165-75. Epub 2013 Apr 6 PubMed.

  4. Cross-sectional neuropathological studies suggested that amyloid accumulation follows a distinct pattern with 3 to 5 stages (Braak and Braak, 1991; Thal et al., 2002). With the availability of amyloid PET tracers, these patterns can now also be assessed in vivo. Mattsson and colleagues propose a three-stage model with start of aggregation in the basal frontal area and posterior cingulus and precuneus, then in the remaining neocortex and allocortex, and eventually in the primary sensory and motor cortices and basal occipital lobe.

    They first defined three groups based on the presence of abnormal CSF Aβ42 or amyloid PET with the assumption that those with abnormal CSF Aβ42 but normal amyloid PET are in an earlier stage (early accumulator) than those with both abnormal CSF Aβ42 and amyloid PET (late accumulator). Next, brain regions were identified that showed different amyloid aggregation rates over time between early and late accumulators and a control group with normal amyloid measures.

    The cross-sectional aggregation in areas that showed specific accumulation rates in early and late stages were used to allocate individuals to these stages. As almost all neocortical regions showed similar accumulation rates between early and late accumulators, the cross-sectional aggregation in these regions were used for allocation of individuals in a novel intermediate group. Using the same ADNI dataset but based on cross-sectional data only, Grothe et al. suggested a four-stage model.

    The CSF Aβ42 concentrations in the four Grothe stages (zero, 1, 2, 3, 4) were respectively 224, 209, 160, 138, 127 ng/L, while CSF Aβ42 in the Mattsson stages (zero, 1, 2, 3) were respectively 224, 178, 139, 137 ng/L . This could suggest that the Grothe system is a bit more refined. However, head-to-head comparisons between the models and with global amyloid SUVr are needed to determine which measure best reflects disease progression.


    . Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991;82(4):239-59. PubMed.

    . Phases of A beta-deposition in the human brain and its relevance for the development of AD. Neurology. 2002 Jun 25;58(12):1791-800. PubMed.

    . In vivo staging of regional amyloid deposition. Neurology. 2017 Nov 14;89(20):2031-2038. Epub 2017 Oct 18 PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Refining Models of Amyloid Accumulation in Alzheimer’s Disease
  2. Daydreaming Network Serves as Ground Zero for Aβ Deposition
  3. Aβ, Then Metabolism, Then Atrophy: In Familial AD, Cascade is Definitive
  4. PET Staging Charts Gradual Course of Amyloid Deposition in Alzheimer’s

Paper Citations

  1. . Cerebrospinal fluid analysis detects cerebral amyloid-β accumulation earlier than positron emission tomography. Brain. 2016 Apr;139(Pt 4):1226-36. Epub 2016 Mar 2 PubMed.
  2. . Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991;82(4):239-59. PubMed.

External Citations

  1. European Amyloid Imaging to Prevent Alzheimer’s Disease

Further Reading

No Available Further Reading