Studies of familial and sporadic Alzheimer’s disease paint a remarkably consistent picture of how biomarkers change in preclinical stages. Now, new cross-sectional biomarker data from the Alzheimer’s Prevention Initiative’s Colombian kindred further solidify these models. Published in the January 12 JAMA Neurology, the data in their broad strokes resemble findings from early onset families in the Dominantly Inherited Alzheimer Network (DIAN). A few intriguing discrepancies emerged as well, although it was unclear if these represented differences in the underlying pathology, or an artifact of how biomarkers are measured. “The next question is whether these data will translate to what is happening longitudinally,” noted Adam Fleisher, who led the study while at Banner Alzheimer’s Institute, Phoenix. He now works at Eli Lilly.

“The results are remarkably consistent with what we have found in DIAN,” Randall Bateman at Washington University in St. Louis wrote to Alzforum. “This further supports a consistent finding of Alzheimer’s biomarker changes 15 to 20 years before symptom onset, and suggests an order of events leading up to cognitive decline. The authors should be congratulated on a heroic effort and analysis—it is a tour de force accomplishment.”

Previously, API reported amyloid imaging data from the same Colombian cohort, comprised of 32 mutation carriers and 22 non-carriers between 20 and 60 years old. Seven of the carriers were diagnosed with mild cognitive impairment, and five with AD, with the remainder cognitively normal. In carriers, fibrillar amyloid became abnormal around 28 years of age, or about 16 years before the expected age of cognitive impairment, and plateaued 10 years later, the researchers found (see Nov 2012 news). In a separate 2012 paper, Banner scientists also described mixed biomarker data from a younger group in the Colombian kindred. In these 18- to 26-year-olds, the researchers saw signs of shrunken brain volume and abnormal brain-activation patterns in the absence of detectable amyloid pathology (see also Dec 2011 conference newsMar 2012 conference news).

API Biomarker Curves. Biomarkers become progressively more abnormal in mutation carriers; circles represent the age at which each marker deviates significantly from levels in non-carriers. [Copyright © 2015 American Medical Association. All rights reserved.]

The new paper focuses on the 20- to 60-year-old cohort, comparing previous amyloid PET results to structural and functional imaging and fluid biomarkers (see image above). The first biomarker to move was Aβ42 in cerebrospinal fluid (CSF), which deviated from normal levels around 24 years of age, or 20 years before expected cognitive impairment. Amyloid imaging followed at 16 years before symptom onset, then waning brain metabolism was seen with FDG PET at 15 years beforehand. At the same time, levels of CSF total tau shot up, while phosphorylated tau reached abnormal levels shortly thereafter, at 13 years out. Hippocampal volume was the last to budge, becoming abnormal about six years before expected symptom onset. The first subtle cognitive deficits in tests of word recall also appeared at this time.

The findings parallel those from DIAN, which likewise reported CSF Aβ42 moving first, followed by abnormal amyloid imaging and elevated CSF tau, both occurring around 15 years before symptom onset (see Jul 2012 newsAug 2012 conference news). “We were intrigued to see changes in spinal fluid Aβ before changes in amyloid PET, as the DIAN study did. It was nice to confirm that,” Fleisher told Alzforum. 

Nonetheless, some differences cropped up as well. In the DIAN cohort, abnormal brain glucose metabolism came later, at 10 years before symptom onset, while hippocampal volume deviated sooner, as early as 15 years prior. In addition, DIAN participants showed the first signs of cognitive slippage on a logical memory test at 10 years out. “I am rather surprised by the differences,” Gaël Chételat at INSERM-EPHE-University of Caen, France, wrote to Alzforum. “These might partly be due to methodological differences in the way the biomarkers are measured, as this is known to have a considerable impact on the findings. … Of course, this might also reflect differences between the genetic variants.” The Colombian kindred carries a single E280A presenilin 1 mutation, whereas the DIAN cohort includes 51 different mutations in APP or presenilin.

For the metabolic differences, Fleisher agreed that either explanation might fit. However, he believes methodological differences explain other discrepancies. For example, both the API and DIAN datasets record the first changes in hippocampal volume around 15 years before symptom onset, but due to differences in the statistical analysis, the DIAN study reports significance earlier than API, he said. In addition, the two studies calculate age of onset differently, with DIAN using the age at which people first notice memory problems, and API using the age at which doctors formally diagnose mild cognitive impairment. This may also lead to some differences in biomarker staging, Fleisher said.

Fleisher pointed out that although the API cohort is smaller than the DIAN group, it has the advantage of including people only from the same ethnic group and environment and having only a single presenilin 1 mutation, presumably lessening variability in the pathological presentation. “The pathology has a well-described natural history of how it progresses clinically over time. Therefore, what we are seeing in cross-section analysis likely maps to what we would see longitudinally,” Fleisher predicted. Banner collects longitudinal data from this cohort as well, but has not published on this yet. The first longitudinal data from DIAN revealed a surprise, with CSF tau dropping after symptoms appeared (see Mar 2014 news). To date, Banner has not seen this in the API cohort, Fleisher said. “We need to duplicate the finding in other datasets, and gain a better understanding of whether it happens in sporadic AD as well.”

So far, data from both familial studies echo those from longitudinal studies of sporadic AD (see Apr 2013 conference news), and conform well to models of biomarker progression put forth by Clifford Jack at the Mayo Clinic in Rochester, Minnesota, and others (see Feb 2013 conference news). API mainly diverges from the Jack model in showing a much earlier drop in brain metabolism. Otherwise, the data fit closely, including CSF Aβ42 changes preceding those in amyloid PET, Fleisher said. In the API dataset, some of the curves look linear rather than sigmoid, but Fleisher noted that this is likely to be an artifact of the limited data points and age range examined. Other researchers have noted that some biomarkers change in a roughly linear fashion during prodromal disease stages (see Dec 2014 conference news).—Madolyn Bowman Rogers


  1. This is paper represents a huge effort that provides a valuable and strong contribution to the field. The most impressive characteristic of this study is that the cohort of 32 mutation carriers (including 20 asymptomatic and 12 cognitively impaired individuals) has a single genetic variant and the same ethnicity, race, and general geographic location with similar cultural influences. This probably reduces the variability considerably compared, for instance, with the Dominantly Inherited Alzheimer's Network (DIAN) study, where the 88 mutation carriers had a combined total of 51 different mutation pedigrees. It would be interesting to get direct comparisons between both cohorts (in terms of variability but also as to the results per se), as it is difficult to evaluate these differences just by comparing this paper with those from DIAN.

    In any case, there is no doubt as to the relevance of this study. And yet I do not completely share the authors' interpretation. Indeed, instead of the similarities with the DIAN study (see Bateman et al., 2012),  I am rather surprised by the contrasts. In the Alzheimer's Prevention Initiative study, Fleisher et al. found that hippocampal atrophy differs between carriers and non-carriers six years before the kindred’s median age at MCI onset, whereas DIAN found it occurred approximately 15 years before expected symptom onset. Also, the plasma Aβ1-42 levels were consistently elevated from 15 years prior to symptom onset in mutation carriers in DIAN, whereas in the API study there was no significant separation between carrier and non-carrier levels, and in fact the values tend to overlap more as people age (see Fig. 1 in Fleisher et al.).

    This divergence might partly be due to methodological differences, notably in the way the biomarkers are measured, since this is known to have a considerable impact on the findings (see e.g. Frisoni et al., 2013). Also, as highlighted by the authors, the sensitivity in the measures might differ among biomarkers, which would influence the sequence in which they test positive. The optimization of the methods might differ across biomarkers as well, depending on  the quality of the acquired images (especially for MRI).

    Of course, biomarker differences might also reflect heterogeneity of the genetic variants. This raises the broader issue of the generalization of the findings, and notably relates to the question that always gets raised about autosomal dominant AD studies: How relevant are results from mutation carriers to those expected in sporadic AD? This question is especially relevant when attempting to characterize and compare the age at initial change across biomarkers, because mutations alter the age at onset compared with the sporadic form, and they modulate, specifically, Aβ-related changes.

    Apart from the comparison to DIAN, I also have a different take on the sequence of preclinical AD pathologies: In the discussion, the authors claim that the results are consistent with the biomarker timeline proposed by Jack, namely “CSF and PET measures of Aβ pathology are followed by CSF measures of tau pathology and regional CMRgl decline, followed by hippocampal atrophy and clinical progression.” However, it seems to me that, maybe surprisingly, the two PET markers (FDG and Florbetapir), as well as CSF tau, emerge at the same time, if one looks at the mean age the markers test positive (figure below). There is even no difference from CSF Aβ42 when looking at the 95 percent confidence interval. One would definitively need to use statistical comparisons (in the age of onset across biomarkers) to further understand how age at onset of these biomarkers differs significantly from each other.


    Illustration of the age at initial change (vertical bar) with the 95 percent confidence interval (data taken from Table 2 in Fleisher et al.).


    . Clinical and biomarker changes in dominantly inherited Alzheimer's disease. N Engl J Med. 2012 Aug 30;367(9):795-804. PubMed.

    . Imaging markers for Alzheimer disease: Which vs how. Neurology. 2013 Jul 30;81(5):487-500. PubMed.

    View all comments by Gael Chetelat
  2. We thank Dr. Chetelat for her kind remarks and thoughtful input. While there are some differences between the DIAN and API data, we continue to be impressed by the consistency of our findings, especially given the relatively small sample sizes, large confidence intervals, and differences in mutations, image analysis techniques employed, and the ways in which biomarker trajectories and ages at clinical onset were characterized.

    Regarding ages at clinical onset, DIAN estimated each mutation carrier’s age at clinical onset based on the estimated age of symptom onset in his or her affected parent, whereas API estimated the PSEN1 E280A mutation carriers’ age at clinical onset based on the median age at which carriers in the kindred met clinical criteria for MCI. There are also different ways to measure the onset of progressive biomarker changes, and they impact estimated ages of progressive biomarker onset. Indeed, our article discusses three complementary methods for calculating the onset of progressive biomarker changes.

    For estimated ages at the onset of hippocampal shrinkage, initial slope declines began to deviate approximately 15 years prior to clinical onset in DIAN and six years prior to clinical onset in API. On the other hand, hippocampal volumes began to significantly differ between mutation carriers and non-carriers roughly three years prior to clinical onset in DIAN and two years prior to clinical onset in API. Given the small sample sizes, large confidence intervals, and similarities in the age-related hippocampal changes in DIAN and API, we would be hesitant to suggest that the cohorts differ significantly in their hippocampal volume trajectories. 

    We previously reported that 18- to 26-year-old PSEN1 E280A mutation carriers had elevated (not reduced) CSF Aβ42 levels more than two decades before the kindred’s estimated median age at clinical onset, consistent with Aβ42 overproduction in this form of ADAD (Reiman et al.,  2012). While we did not have a sufficient number of very young adult subjects in the present study to confirm that finding, we continue to believe that CSF Aβ42 levels are initially elevated in autosomal dominant AD prior to beginning the sequestration of Aβ42 in amyloid plaques.

    Regarding plasma Aβ42 levels, DIAN and API findings each suggest that these levels are elevated throughout the preclinical and clinical stages of autosomal dominant AD, and neither study demonstrated significant associations with age. While it is conceivable that significant direct or inverse associations with age might be detected in larger samples, we currently postulate that the levels are unchanged throughout the natural history of autosomal dominant AD. 

    As we discussed in our article and as Dr. Chetelat astutely observes, we could not statistically distinguish ages of onset between the different biomarkers (with the exception of CSF Aβ42 and hippocampal volumes), due at least in part to our relatively small sample sizes and large confidence intervals. Nonetheless, we believe that our age associations are consistent with the hypothetical curves by Cliff Jack (Jack et al., 2013). 

    Future studies promise to refine the onset and trajectories of different biomarker, cognitive, and clinical changes in autosomal dominant AD, which typically overproduce Aβ, and determine the extent to which those findings are generalizability to late-onset AD, a disease with reduced Aβ clearance.


    . Brain imaging and fluid biomarker analysis in young adults at genetic risk for autosomal dominant Alzheimer's disease in the presenilin 1 E280A kindred: a case-control study. Lancet Neurol. 2012 Dec;11(12):1048-56. PubMed.

    . Tracking pathophysiological processes in Alzheimer's disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurol. 2013 Feb;12(2):207-16. PubMed.

    View all comments by Eric M. Reiman

Make a Comment

To make a comment you must login or register.


News Citations

  1. API Echoes DIAN: Biomarker Changes Precede Symptoms by 20 Years
  2. Reeling In Biomarker Data in Young Carriers, API Rocks Staging Boat
  3. Miami: When Does Amyloid Deposition Start in Familial Alzheimer’s?
  4. Paper Alert: DIAN Biomarker Data Show Changes Decades Before AD
  5. In Big Picture, Familial AD’s Biomarker Data Resemble LOAD
  6. DIAN Longitudinal Data Surprises With Late Drop in Tau
  7. From Natural History, A "Renaissance" for Amyloid Hypothesis
  8. HAI—Sharper Curves: Revamping a Biomarker Staging Model
  9. Large Studies Agree: Brain Amyloid Accelerates Cognitive Decline

Mutations Citations

  1. PSEN1 E280A (Paisa)

Further Reading

Primary Papers

  1. . Associations between biomarkers and age in the presenilin 1 E280A autosomal dominant Alzheimer disease kindred: a cross-sectional study. JAMA Neurol. 2015 Mar;72(3):316-24. PubMed.