In clinical research, a first-time finding may rouse intrigue, but it’s generally independent confirmations of the data that give them potential to leave an indelible mark. That’s why a study of cerebrospinal fluid (CSF) biomarkers of Alzheimer disease in this month’s Archives of Neurology should carry all the more weight—enough to warrant coverage in The New York Times. Several key findings—the presence of an AD signature in more than a third of cognitively normal seniors, and the actual figures that define this pathological read—confirm previously published data. In the newly published study, Hugo Vanderstichele, Innogenetics, Gent, Belgium, and colleagues not only identified an AD-like CSF signature in seniors enrolled in the Alzheimer’s Disease Neuroimaging Initiative (ADNI), but also validated these data in two independent European populations. (Innogenetics is a Belgian company that manufactures CSF test kits and stands to gain from their widespread use.) In a separate study reported last month at the International Conference on Alzheimer’s Disease in Honolulu, Hawaii, Steven Edland, University of California, San Diego, and colleagues derived a similar CSF signature, analyzing participants of wide-ranging age from five AD research centers in the U.S.

Prior research in Europe and the U.S. established low CSF Aβ1-42 and elevated CSF tau and phosphorylated tau-181P (p-tau) as early indicators of AD pathogenesis (see, e.g., ARF related news story on Fagan et al., 2007). More recently, the ADNI fluid biochemistry study, led by Leslie Shaw, University of Pennsylvania, Philadelphia, defined threshold concentrations of spinal fluid Aβ42 and tau that associated with disease (Shaw et al., 2009). Shaw and colleagues identified an Aβ42 cut point of 192 pg/ml or lower as diagnostic for AD, by analyzing pre-mortem CSF samples from autopsy-confirmed AD patients and age-matched cognitively normal research participants at the university’s AD research center. Applied to the ADNI dataset, these CSF criteria correctly identified AD patients more than 96 percent of the time.

In the current paper, first author Geert De Meyer, Ghent University, Belgium, and colleagues, including Shaw, chose a different strategy to tackle the same problem of identifying CSF thresholds that discriminate groups of people with and without AD. They capitalized on an uncanny pattern in the data from Shaw’s 2009 study: the normals fell into two fairly distinct subgroups—one with an AD-like CSF profile, the other without. This prompted De Meyer and colleagues to use a statistical approach known as mixture modeling to identify naturally occurring AD signatures, or cut points, based on CSF Aβ42 and p-tau181, in more than 400 ADNI participants, aged 55 to 90, with AD, mild cognitive impairment (MCI), or normal cognition. The novelty here is that De Meyer and colleagues determined their diagnostic cut points without any clinical or diagnostic information on the subjects. They did this to avoid bias-related problems with more commonly used methods for assessing biomarkers.

To validate the CSF threshold values coming from this clinically blind approach, the researchers analyzed two separate populations outside of ADNI: an autopsy-proven cohort in Belgium, and a subset of patients from a different European longitudinal study. In the Belgian cohort, 65 of 73 people had an AD diagnosis at autopsy, and the CSF criteria correctly classified 94 percent of these patients. In the European study (Hansson et al., 2006), 57 of 175 MCI patients converted to AD within five years, and the CSF threshold values identified 100 percent of these converts. Furthermore, 36 percent of cognitively normal ADNI seniors had AD-like CSF reads, which “underscores the presence of AD pathology before the onset of symptoms,” the authors write.

Perhaps as impressive was the fact that the study by De Meyer et al. arrived at a CSF cut point of 188 pg/ml, virtually identical to the 192 pg/ml coming from Shaw’s earlier work, which used a different study design and statistical methods. “What had been a huge issue is that numerical values (for CSF thresholds) varied all over the place,” Shaw told ARF. “We're starting to see better replication of the quantitative results, and that's very important.” Both studies used the same Innogenetics platform for their immunoassays, and each found the CSF data falling neatly into two bins, those with and without an AD signature, even within the normal groups.

At ICAD, Edland reported strikingly similar findings from a CSF biomarkers analysis done in collaboration with Elaine Peskind, University of Washington, Seattle, who served as a panelist in a recent ARF Webinar on the value of CSF analysis (see ARF Live Discussion). Edland and colleagues also used the Innogenetics platform, even applying the same unbiased statistical methods, to analyze 303 cognitively normal volunteers recruited at five AD research centers. These are not ADNI participants. Their CSF Aβ42 cut-off value came out between 190 and 200 pg/ml, Edland told ARF, which is in close proximity to the thresholds reported in the papers by De Meyer et al. and Shaw et al.

Edland’s study was unique in that it included younger adults. Of his volunteers, 125 were between the ages of 20 and 55. “The importance of having all those young normals is that it really nails down what a ‘normal’ CSF profile is,” Edland said in a phone interview. This set the stage for the “dramatic” finding in the older subgroup. “Many might guess that you've got normal CSF Aβ levels as young adults, and as you acquire disease, the [CSF readout] would drift to an AD-like level,” he said. However, CSF Aβ42 levels in the 56-and-older group were clearly bimodal, confirming the results of Shaw’s 2009 paper, which first described this CSF Aβ42 distribution in normals. “One mode looks just like the young normals, and the other looks just like AD,” said Edland. “The fascinating thing about the bimodal distribution is that it suggests the transition between normal and AD CSF profiles happens very quickly. It's not a smear of data. People jump from one distribution to the other. I think that is the most important message here.”

When they considered ApoE genotype, the researchers found that, among cognitively normal E4 carriers ages 70 and up, two-thirds have an AD-like CSF profile, Edland said. By contrast, only a fourth of similarly aged E4 non-carriers had the AD signature.

Taken together, the recent findings should fuel a growing movement to support CSF analysis, which has for years gotten short shrift because of some doctors’ reluctance to perform spinal taps (aka “lumbar punctures”) and patients’ unwillingness to receive the invasive procedure. On this point, an Archives of Neurology commentary on the paper by De Meyer et al. notes that spinal taps are “no more invasive than other outpatient procedures such as endoscopies that millions of Americans tolerate each year.” Moreover, the cost of CSF Aβ and tau readouts pales in comparison to “the consulting physician’s bill, the charge for neuropsychological testing, and the cost of a magnetic resonance brain scan” at most centers, write A. Zara Herskovits of Brigham and Women’s Hospital, and John Growdon of Massachusetts General Hospital, both in Boston.

Routine clinical use is still a way off, partly due to quality control and manufacturing issues. However, experts have launched worldwide initiatives to address these challenges (see ARF related news story), and included CSF measures among a set of biomarkers in a draft revision of AD diagnostic criteria (see ARF related news story).

As biomarker studies in milder populations proceed apace (see ARF coverage of presentations at this year’s International Conference on Alzheimer’s Disease), some think CSF assays warrant more attention in some research settings. “I feel the tests are ready for prime time for inclusion in treatment trials, for example,” Shaw told ARF, noting that CSF biomarkers could help identify participants at high or low disease risk.

Furthermore, “gazing into the future when there are neuroprotective medications for AD, we can envision a recommendation that CSF analyses be implemented as a screening test to identify clinically healthy individuals at risk for MCI and AD,” Herskovits and Growdon write. “The information gained would enable early application of treatments to delay onset of symptoms or slow progression of cognitive impairments.”—Esther Landhuis

Comments

  1. The Stanford Neurology Department held a journal club meeting on this paper, conducted by Geoffery Kerchner, M.D. This meeting brought out some concern that several conclusions appear faulty, controls are missing, and the comparison datasets are not defined or controlled. The analysis of ApoE genotype was only for “ε4 carrier status,” not by specific genotype, and the paper even reported that the carrier status indeed seemed to account for a substantial portion of the results.

    This journal club provided an opportunity for me to present and discuss an analysis of the exact same ADNI dataset CSF protein values reported in this paper (114 normal individuals, 200 MCI, 102 mild AD) that we conducted at the Stanford/VA Aging Clinical Research Center (with Art Noda, Beatriz Hernandez, and Jerome Yesavage). In our separate analysis, a major issue we see is that ApoE genotype may explain most of the CSF Aβ variation that occurs in the ADNI data reported in this paper. (See table.) As per our analysis, the ApoE genotype, not diagnosis, explains the CSF-Aβ divergences seen in the ADNI study. The abundance of the ApoE4/4 individuals in the AD group (22/102 subjects = 22 percent) versus their infrequency in the normal group (2/102 subjects = 2 percent), as well as similar variation in the same direction for ApoE3/4 (which is typical for both normal and Alzheimer’s groups), accounts for most of reported variance in this paper. Specifically, in our analysis, there is large significance for the comparisons for ε3/3 versus ε3/4 and ε3/3 versus ε4/4. (See table: normal versus mild AD) However, when changes for specific ApoE genotypes are examined independently, the differences between diagnostic groups for CSF Aβ is not significant (p >0.14 for all comparisons).

    In the paper, the absence of a control group description for the MCI subjects from Ghent (first non-ADNI comparison group) makes the interpretation of that dataset problematic, especially since their distribution was even further separated from the normal circles relative to the ADNI group. Further, the autopsy group (second comparison group) samples were collected within a year of death; thus, this replication group is probably a severely demented group, which may have had different levels of CSF proteins. The paper provides no ApoE genotype information to see how this critical factor could have affected the analyses of these two comparison groups.

    Our analysis of the ADNI data with respect to specific genotypes implies that it is not clear that CSF Aβ contributes any more information to diagnosis than do ApoE genotype and age. The relationship between a person’s ApoE genotype (not simply “carrier status”) and CSF Aβ must be studied more carefully, with respect to age and Alzheimer disease diagnosis, before any clinician should consider use of CSF Aβ as a diagnostic indicator. What is really missing in studies that have been conducted to date is how CSF Aβ levels change with respect to age in normal individuals according to ApoE genotype, particularly in subjects with the ApoE4/4 genotype. Peskind et al. (Peskind et al., 2006) have already shown that CSF Aβ42 shows a significantly and substantially greater decrease with age in subjects with an ApoE4 allele, and this work needs to be extended to a large population of subjects with the ApoE4/4 genotype across a broad age range.

    Another issue is CSF tau. This measure is highly related to cognitive impairment along the continuum from normal through MCI to mild dementia. The analysis of the ADNI data shows this relationship. (See table: normal versus mild AD) Of note, CSF tau appears to peak in mild dementia but may decline as neurons degenerate. The question then is whether CSF tau contributes any more information to estimation of clinical status than a proper medical history or cognitive testing.

    More generally, the attitude toward genotyping is an important issue in the medical community. The ApoE4/4 individuals seem to represent a group with extremely significant Alzheimer’s-related morbidity that is not getting proper attention. A consideration for research into preventive agents for AD is to focus on the ApoE4/4 individuals, who have a relatively more specific variety of AD that would likely facilitate study in this area greatly.

    References:

    . Age and apolipoprotein E*4 allele effects on cerebrospinal fluid beta-amyloid 42 in adults with normal cognition. Arch Neurol. 2006 Jul;63(7):936-9. PubMed.

    View all comments by John (Wes) Ashford
  2. Reply to comment by Wes Ashford
    Although it is true that ApoE genotype is associated with differences in CSF biomarker levels (e.g., Morris et al., 2010), our previous and more recent work has shown that it is not the most important measure for determining the likelihood of someone who is cognitively normal (clinical dementia rating, CDR, of 0) progressing to the earliest stages of AD dementia (CDR 0.5), at least over a several-year period of time (Fagan et al., 2007). In one part of our 2007 study, 61 cognitively normal elders were clinically followed for an average of three to four years after CSF collection. Thirteen of these individuals progressed to CDR >0 in that time span, while 48 remained clinically normal. There was no difference in the percentage of E4+ individuals in the two groups (each group comprised 31 percent E4+). Cox proportional hazard models revealed that education and the CSF tau/Aβ42 and ptau181/Aβ42 ratios, but not ApoE genotype, significantly predicted conversion from CDR 0 to CDR >0. In fact, after adjusting for the demographic variables (age, education, gender, and ApoE genotype), the hazard ratios of the two CSF measures actually increased (HR of tau/Aβ42 increased from 2.42 [95 percent CI: 1.15-5.08] to 5.21 [95 percent CI: 1.58-17.22]; HR of ptau181/Aβ42 increased from 1.78 [95 percent CI: 1.00-3.16] to 4.39 ([95 percent CI: 1.62-11.86]). This initial study was not suitably powered to test the effect of the various allele combinations.

    We are currently investigating this issue in a much larger cohort of clinically normal individuals with longitudinal follow-up. In these recent studies, we have found very similar results to the 2007 study, indicating that the tau/Aβ42 ratio is very predictive of which cognitively normal, CDR 0, individuals, will go on to progress to CDR >0. However, now the number of CDR 0 individuals we have followed is 164 instead of 61. Again, ApoE genotype was not predictive of who would progress over a three- to four-year period of time. In another study, we found that ApoE genotype did not contribute significantly in persons with elevated amyloid burden (as detected by amyloid imaging, PIB-PET) to the prediction of progression from cognitive normality to incident cognitive impairment and symptomatic AD (Morris et al., 2009). In our Adult Children Study (P01 AG026276), we are assessing the effect of parental family history of AD, as well as ApoE genotype, on the time course of fluid and imaging biomarkers of AD starting from middle age (age 45-74 years; 300 cognitively normal participants). We expect to learn much about the role ApoE genotype (and other factors) play in the initial appearance and subsequent development of AD neuropathology and its behavioral sequelae.

    References:

    . APOE predicts amyloid-beta but not tau Alzheimer pathology in cognitively normal aging. Ann Neurol. 2010 Jan;67(1):122-31. PubMed.

    . Cerebrospinal fluid tau/beta-amyloid(42) ratio as a prediction of cognitive decline in nondemented older adults. Arch Neurol. 2007 Mar;64(3):343-9. Epub 2007 Jan 8 PubMed.

    . Pittsburgh compound B imaging and prediction of progression from cognitive normality to symptomatic Alzheimer disease. Arch Neurol. 2009 Dec;66(12):1469-75. PubMed.

    View all comments by David Holtzman

Make a Comment

To make a comment you must login or register.

References

News Citations

  1. Biomarker Roundup: Collecting Clues from MRIs to RNAs
  2. Worldwide Quality Control Set to Tame Biomarker Variation
  3. Noisy Response Greets Revised Diagnostic Criteria for AD
  4. Honolulu: Biomarker Profiles That Spell Trouble for ‘Normals’?

Webinar Citations

  1. Untapped Resource? New Study to Boost Acceptance of CSF Analysis

Paper Citations

  1. . Cerebrospinal fluid tau/beta-amyloid(42) ratio as a prediction of cognitive decline in nondemented older adults. Arch Neurol. 2007 Mar;64(3):343-9. Epub 2007 Jan 8 PubMed.
  2. . Cerebrospinal fluid biomarker signature in Alzheimer's disease neuroimaging initiative subjects. Ann Neurol. 2009 Apr;65(4):403-13. PubMed.
  3. . Association between CSF biomarkers and incipient Alzheimer's disease in patients with mild cognitive impairment: a follow-up study. Lancet Neurol. 2006 Mar;5(3):228-34. PubMed.

External Citations

  1. The New York Times
  2. Alzheimer’s Disease Neuroimaging Initiative
  3. AD diagnostic criteria

Further Reading

Primary Papers

  1. . Diagnosis-independent Alzheimer disease biomarker signature in cognitively normal elderly people. Arch Neurol. 2010 Aug;67(8):949-56. PubMed.
  2. . Sharpen that needle. Arch Neurol. 2010 Aug;67(8):918-20. PubMed.