Testing a widely used cerebrospinal fluid Aβ42 immunoassay, scientists in Cambridge, Massachusetts, proposed modifications that they believe improve its precision and accuracy enough to pick up subtle changes within an AD cohort that would indicate potential therapeutic efficacy in drug trials. The modifications could also help it pass muster with regulatory agencies, the scientists claim. “We have performed the first Good Laboratory Practice (GLP)-equivalent validation of a CSF Aβ42 assay. Our results show that this immunoassay underestimates the amount of Aβ in CSF samples, and that modifications we have introduced improve accuracy,” said Valerie Cullen, who led the study with co-first author Ross Fredenburg and senior investigator Michael Solomon. All three were previously at the biotech company Link Medicine; Cullen is now consulting for Neurophage Pharmaceuticals, Fredenburg works at AstraZeneca, and Solomon has moved to Ember Therapeutics, all in the Boston, Massachusetts, area. Their paper appeared as Open Access May 3 in the American Association of Pharmaceutical Scientists (AAPS) Journal.

CSF Aβ42 drops prior to symptomatic Alzheimer’s dementia, and this decline predicts progression in people with mild cognitive impairment. One of the most extensively used commercial assays for measuring CSF Aβ42 is the INNOTEST Aβ42 ELISA from Innogenetics NV—a Belgian company bought in 2010 by Fujirebio of Tokyo. On the market since 2000, the INNOTEST kit discriminates AD from normal aging and other neurologic conditions (Hulstaert et al., 1999). It has been validated on autopsy-confirmed patient samples (Engelborghs et al., 2008) and has Europe’s CE mark of quality for diagnostic applications. Yet the assay suffers considerable intercenter and interlab variation (Mattsson et al., 2010). So do other CSF biomarker immunodetection methods, and this hinders their use in multisite clinical trials. Furthermore, different assay batches supplied by the manufacturer can give different readings for a given CSF sample (see ARF related news story on Fagan et al., 2011).

Such problems crop up in part because Aβ42 loves to aggregate. Indeed, the same pesky propensity that drives plaque formation within the brain also makes the peptide stick to ELISA collection tubes. Furthermore, Aβ42’s stickiness contributes to matrix effects—a property of the sample that causes faulty measurements for a particular analyte, depending on what else floats in the sample.

In their paper, Cullen, Fredenburg, and colleagues report ways to get around these two issues in the INNOTEST Aβ42 ELISA. To reduce Aβ adherence, the researchers spiked CSF samples with the detergent Tween-20. To deal with matrix interference, they diluted samples sixfold before loading them onto the ELISA plate. The manufacturer's instructions say to use undiluted CSF. However, by doing so, “you actually underestimate the amount of Aβ42 by at least twofold,” Cullen told Alzforum. The researchers determined this by running the ELISA on normal, MCI, and AD CSF samples at concentrations ranging from undiluted to 1:16 dilution.

“This advanced validation of a method for improving the quantification of Aβ1-42 in human CSF is extremely helpful to strengthen the use of CSF biomarkers in future clinical trials and in routine diagnostic settings,” Brit Mollenhauer of Paracelsus-Elena Clinic, Kassel, Germany, wrote in an e-mail to Alzforum (see full comment). “I hope that other assays (e.g., for total tau protein) and platforms for CSF proteins get the same attention and thorough validation.” Mollenhauer and Cullen formerly worked as postdocs in Michael Schlossmacher’s lab at Brigham & Women’s Hospital in Boston.

The authors optimized the ELISA to support a Phase 1b AD trial completed last year. Designed to assess safety and tolerability of LKN-754, a compound related to autophagy developed by Link, the trial monitored CSF Aβ42 pharmacodynamics in response to treatment, Cullen said. She and Fredenburg declined to comment on the trial outcome, but said they would be confident using the optimized methods to stratify patients into study cohorts or to monitor pharmacodynamics of other compounds.

Several scientists expressed concern about the optimizations proposed in the paper. On adding detergent to the CSF samples, Anne Fagan of Washington University School of Medicine, St. Louis, Missouri, wonders if this might affect Aβ42’s native structure in human fluid. “It has been proposed that oligomeric Aβ resides in CSF,” said Fagan, who directs the biomarker core for the international, multicenter Dominantly Inherited Alzheimer Network (DIAN) study. “Could Tween-20 disrupt the oligomeric species, and maybe that’s why you see elevated Aβ42 levels compared to samples without detergent?”

Another question is whether the detergent might influence Aβ42’s immunoreactivity, scientists said. “Detergent molecules will likely associate themselves with what is being measured,” suggested Leslie Shaw of University of Pennsylvania School of Medicine, Philadelphia. “They certainly will be in the matrix of what is being measured. What is the effect of this modified matrix on the immunologic reaction taking place?” Along with UPenn colleague John Trojanowski, Shaw co-directs the Alzheimer’s Disease Neuroimaging Initiative (ADNI) biomarker core.

ADNI uses Innogenetics’ bead-based multiplex AlzBio3 immunoassay—not the plate-based ELISA that Cullen and colleagues optimized. The AlzBio3 system is more suitable for high-throughput analysis because it allows measurement of all three CSF AD biomarkers (Aβ42, total tau, and phospho-tau) in a single well. Prior work has shown diagnostic equivalence between the AlzBio3 and ELISA methods for detecting AD (Olsson et al., 2005; Reijn et al., 2007; Fagan et al., 2011). Cullen and colleagues initially considered validating the AlzBio3 assay; however, when preliminary runs hinted at matrix interference issues for Aβ—which has different chemical properties from tau or phospho-tau—they opted against trying to optimize all three analytes in the same kit. Independent of the current paper, the scientists did examine the Meso Scale Discovery platform, which measures Aβ38, Aβ40, and Aβ42 simultaneously, and hope to publish that data shortly, Cullen said.

Fagan’s group uses both the INNOTEST ELISA and AlzBio3 kits for measuring CSF Aβ42, but has more experience with the former. She and Shaw discussed the possibility of doing a collaborative project testing the reported dilution and/or detergent modifications on samples from people with corresponding amyloid imaging data and clinical follow-up. “Whether we could consider modifying the protocol would depend on the results of that study,” Fagan said. “In our hands, the INNOTEST ELISA as it is currently sold has performed quite well in identifying PIB-positive folks and those who will progress clinically. However, there remain quality control challenges such as lot-to-lot variability in commercial kits, so if we can get a better assay, then we’d certainly consider it, at least for certain projects.”

Sebastiaan Engelborghs, who leads a reference lab for biomarker analysis at the University of Antwerp, Belgium, called the new study “a step forward.” However, he noted that the analysis did not address lot-to-lot and interlab variability. Both are nagging problems under exploration in U.S. and European qualification efforts to render AD biomarker assays more broadly useful and consistent across labs (see ARF related news story and ARF news story). “I would not apply the (modified method) until the findings are reproduced in another independent study,” Engelborghs told Alzforum. Douglas Galasko of the University of California, San Diego, expressed similar concern. “It will be of interest to see whether applying this new methodology changes assay reproducibility (especially between labs), and sensitivity and specificity for AD, compared to the ELISA that follows the Innogenetics instructions,” he wrote in an e-mail to ARF (see full comment below).

The current paper highlights the need for a gold standard method to measure CSF Aβ42, as well as certified reference material for calibrating the various assays for these measurements, suggested Henrik Zetterberg of the University of Gothenburg, Sweden, in an e-mail to Alzforum.

Hugo Vanderstichele, formerly of Innogenetics, helped develop the INNOTEST Aβ42 ELISA. He also emphasized the need for international reference materials, as well as for “regulatory documents describing all necessary requirements for an immunoassay.” Along with three coworkers, Vanderstichele last year left Innogenetics to co-found a new biotech company called ADx Neurosciences, which focuses on the development of assays for novel biomarkers for dementia diagnosis in Alzheimer’s, Parkinson’s, and others diseases. “The AD diagnostic market is at a critical point in the sense that clear performance requirements for the assays need to be described for in-vitro diagnostics, as well as for pharmaceutical companies,” Vanderstichele wrote (see full comment below). He noted that validating an assay for pharma use is not the same as validating it as an in-vitro diagnostic.

Robert Dean, who oversees Aβ marker methods for Eli Lilly’s AD team, expressed a similar view. “If you are using an assay in a clinical research setting that is more academic, then you don’t have to make a regulatory claim,” he told Alzforum. “But when a pharmaceutical company runs a trial and uses an assay, we are typically expected to validate the assay’s performance—and in the process define sources of pre-analytical, analytical, and post-analytic variability—as well as determine whether the assay is fit for intended use. I think this paper does a nice job addressing these issues.”—Esther Landhuis.

Cullen VC, Fredenburg RA, Evans C, Conliffe PR, Solomon ME. Development and Advanced Validation of an Optimized Method for the Quantitation of Aβ42 in Human Cerebrospinal Fluid. AAPS J. 3 May 2012. Abstract


Make a Comment

To make a comment you must login or register.

Comments on News and Primary Papers

  1. This is a very important and useful manuscript by Valerie Cullen and coworkers from Link Medicine, who very thoroughly validated and improved an established ELISA platform for the measurement of Aβ1-42 (Innotest® Aβ42 ELISA). The decrease of CSF Aβ1-42 has been (in combination with total tau protein) helpful for the early and differential diagnosis of Alzheimer’s disease, and was even included in clinical research criteria for Alzheimer’s disease (1). Recently, it was shown to be useful for other neurodegenerative diseases, such as dementia with Lewy bodies and Parkinson’s disease. This ELISA has been routinely used in investigations in Europe, but inter-center variation (by pre-analytical, analytical, and post-analytical factors), and inter-laboratory variation (caused by analytical factors), especially for CSF Aβ1-42, lowered the utility also for clinical multicenter trials (2). This has been addressed by a large, multicenter investigation of various assays across 40 laboratories (3).

    The team at Link Medicine validated and improved the Innotest Aβ42 ELISA, according to Good Laboratory Practice level, by minimizing the required CSF volume by dilution, thereby minimizing matrix effects; they changed the calibration curve, introduced quality control samples, and tested different polypropylene vial types and freeze/thaw cycles. This advanced validation of a method for the improvement of quantification of Aβ1-42 in human CSF is extremely helpful to strengthen the use of CSF biomarkers in future clinical trials and in routine analysis. I do hope that other assays (e.g., for total tau protein) and platforms for CSF proteins get the same attention and thorough validation.

    The validation issue will also be addressed by a recent EU initiative (BIOMARKAPD, coordinated by Bengt Winblad) to troubleshoot and improve assay systems. The aim of BIOMARKAPD is the harmonization and standardization of biomarker in cerebrospinal fluid (CSF) for Alzheimer’s disease (Aβ1-42 and tau protein) and Parkinson’s disease (α-synuclein) (see website).

    View all comments by Brit Mollenhauer
  2. This is an interesting paper. It revisits aspects of standardization of Innogenetics Aβ assays (both ELISA and X-MAP) that have received attention before (e.g., by the Blennow and Shaw laboratories). Previous attempts have defined excellent within-laboratory standardization, but inter-laboratory variability for levels of Aβ42 have persistently varied by about 15-30 percent in the international QC program. The main new factors to emerge from this study are:

    1. Addition of Tween when cryotubes are thawed to decrease Aβ adsorption to the polypropylene (which may vary by manufacturer).

    Use of frozen CSF samples that have been stored for three months or longer is an important practical aspect for many laboratories, which have to wait until they have collected enough CSF samples to fill an ELISA plate. And for many research studies, running samples in batches has been used as a way to decrease variability—again pointing to attention to sample collection and storage.

    2. Dilution of 1:6 of CSF, which eliminated a matrix effect and suggested that CSF levels of Aβ42 are higher than typically reported using these assays.

    It will be of interest to see whether applying this new assay methodology changes assay reproducibility (especially between laboratories), and sensitivity and specificity for AD, compared to the ELISA that follows the Innogenetics instructions.

    View all comments by Douglas Galasko
  3. The INNOTEST β amyloid(1-42) was developed many years ago using know-how available at that time (1). Notwithstanding an imperfect accuracy (recovery from samples was an issue), the user-friendly assay format, as well as the clinical value of the assay, was documented in many studies (2) and validated using autopsy-confirmed patient samples (3). The assay already obtained a CE label for diagnostic applications. Assay validation data are also described in the kit Insert of the assay.

    More recently, after integration of the assay in AD therapeutic trials, the issue on matrix interference became more apparent. The accuracy of an assay is linked to recovery, linearity, and parallelism. The group of Valerie Cullen used all critical raw materials and components already available in the commercial kit and modified the test instructions to solve the problem of accuracy by the inclusion of a dilution for the CSF and the addition of detergent to limit the absorption to recipients. However, according to our present understanding of the performance of these types of assays, improvements are especially linked to accuracy, and not necessarily to the other analytical performance characteristics (e.g., precision).

    • An accurate value in a sample can only be obtained after extensive standardization at the level of the assay, the sample, and the operator in the lab (4). It is an integrated approach.
    • A better understanding of what is being measured (monomers, oligomers) will help to evaluate the impact of assay modification on diagnostic applications.
    • The addition of a detergent is important for labs using recipients with a high Aβ absorption rate or to reduce differences in absorption rates, as described previously by Perret-Liaudet (5). The dilution of CSF will reduce the matrix interference and, more importantly, will result in different analyte concentrations in the CSF as compared to previous clinical studies done with the same assay, following test instructions as provided in the kit insert. This is the result of the release of Aβ from the binding proteins present in the CSF (equals higher amounts of detectable Aβ). The qualification of this new, promising approach has only been qualified on a few CSF samples.
    • The results of the new assay format will only be comparable with previous data if a conversion factor can be determined between the old and the new protocol, which is constant over the whole concentration range and independent of the diagnostic groups.
    • In order to integrate this new approach, a full validation, including round-robin studies as done in the Alzheimer's Association Quality Control program (led by Kaj Blennow, Sweden) (6) is required, as well as a confirmation of its clinical value using samples available from worldwide consortia (e.g., ADNI). This will help us understand whether or not the modification will also improve other aspects of the assay, such as precision in testing among labs.
    • The robustness of the CSF dilution protocol will have to be verified.
    • The integration of this new approach, if proven to be valuable, requires an investment from the vendor of the kit.
    • At present, there is a lack of international reference materials, as well as regulatory documents describing all necessary requirements for an immunoassay, either for using in in-vitro diagnostics or pharma. There is a need for consensus on this. The focus on validation of the assay for pharma might be different from validation for in-vitro diagnostics.

    The AD diagnostic market is at a critical point in the sense that clear performance requirements for the assays need to be described for IVD, as well as for pharma. If needed, modifications of the assays have to be implemented, or assays have to be designed on new technology platforms which do not have these matrix issues. The availability of a reference method and/or reference materials will further help to standardize output among the different assays and protocols. In any case, the clinical performance of the assay will have to be confirmed.

    View all comments by Hugo Vanderstichele
  4. Much prior effort in this field has already gone toward validating this assay. As I see it, the authors of this paper have reached a point where we all have been for a decade; in other words, the “modifications” reported here achieved precision and repeatability that every well-functioning laboratory can achieve by following the manufacturer’s SOPs. My lab's numbers are, for example, 12-15 percent for interassay and about 5-7 percent for intra-assay imprecision. Moreover, it also seems to me that the modifications done by the authors have resulted in the increasing of the lowest detection limit.

    It is not fully accurate to state that this assay has not been validated. After all, it has the CE certificate and is routinely used for human diagnostics. In the interest of full disclosure, I have consulted for Innogenetics for some years, and have seen how much effort its assay scientists have invested in validating the assay.

    I have co-coordinated large national/international projects for inter-laboratory QC, and Neurochemical Dementia Diagnostics in my lab is accredited according to ISO 15189. I agree there are imprecision problems around these diagnostics, not only Aβ1-42 but also Tau, pTau181, and Aβ1-40 (see Zimmermann et al, 2011). However, I respectfully submit that the solution is not modifications to get back to 15 percent imprecision. Rather, the starting point should be simply that all laboratories follow what is recommended in the manufacturer’s protocol and standardize pre-analytical sample handling procedures.

    View all comments by Piotr Lewczuk
  5. 1. Minimal required dilutions (MRD): It is a fact that the CSF matrix interferes with the Aβ42 peptides/oligomers. When dilutions are performed with the CSF, a better linearity is seen in this paper. In this way, dilutions seem to be the solution, and I agree on that. However, one problem arises: Due to the dilutions, the Lower Limit Of Quantitation increases to LLOQ = 375 pg/ml. This is not a problem per se, since all results increase, but as a consequence, a new reference range has to be set and a transformation formula for historically collected results has to be established. To set reference ranges and cut-offs, there must be linearity between the diluted CSF (new method) and the undiluted CSF (old method) in a large set of samples.

    2. Aβ42 adsorbs significantly; I agree on that as well. Aβ42 is sticky, and we have to take care of this problem. Everything that decreases the adherence of Aβ42 (without interfering with the assay) has to be implemented, and the effects of adding Tween-20 after thawing are striking and convincing in the current paper. This finding needs replication using other tubes that are currently in use (see Perret-Liaudet et al., 2012).

    3. I am interested in how many analysts performed the different assays in this paper. Is there any difference in the results if different people perform the same assay? In Amsterdam, we organized an international workshop to test this. During this workshop, analysts from 17 international centers performed the same assay under the same conditions, and at least 23 differences between analysts in performing the assay were noted (Teunissen et al., 2010). This indicates that it is very important to follow the exact procedures.

    4. What are the lot numbers of the assays that are used in this study? In our lab, we have experienced differences in results using the INNOTEST assay with different lot numbers. It is something that we have to take into consideration. Ideally, we have to have an assay (plates, diluents, standard, and QC samples) that never changes over time (Mulder et al., 2010; Verwey et al., 2008).

    5. One of the questions we have been asking ourselves for the last five years is, "What is the exact composition of the 'INNOTEST sample diluent'?” And how is the Aβ42 (standard included in the INNOTEST) processed? How is this Aβ42 synthesized, and by whom? I have experienced a lot of differences in synthetic Aβ42 made by different companies. Also, within the same company, different batches of Aβ42 can react differently when dissolved or used to spike. This is the same for Aβ40, although this peptide is more hydrophilic. Maybe this is one of the most important points. How exactly is the INNOTEST made?

    6. One of the big problems the authors touch upon is that the INNOTEST does not deliver QC samples. It would be a great step forward for quality if QC samples are being delivered together with the assay. Now we are forced to make our own QC samples to garantee quality over time. To implement an assay worldwide, QC samples are a necessity. This is the focus of the Global Biomarkers Standardization Consortium (GBSC), initiated by the Alzheimer's Association (Mattsson et al., 2010; Mattsson et al., 2011; Verwey et al., 2009).

    In summary, Aβ42 is influenced by the CSF matrix, and Aβ42 is a sticky peptide. At this moment, the suggestions of this paper to implement a minimal dilution and to add Tween-20 are interesting ones that need to be validated by independent studies before implementing in clinical practice. We believe that it is time to modify the INNOTEST assay to reach the goal of a solid assay that can be reliably used and yields uniform results worldwide.

    View all comments by Niek Verwey


News Citations

  1. Little by Little—Standardizing, Validating Those Biomarkers
  2. Worldwide Quality Control Set to Tame Biomarker Variation
  3. DC: Biomarkers, Parkinson’s—CAMD Needs All Hands on Deck

Paper Citations

  1. . Improved discrimination of AD patients using beta-amyloid(1-42) and tau levels in CSF. Neurology. 1999 May 12;52(8):1555-62. PubMed.
  2. . Diagnostic performance of a CSF-biomarker panel in autopsy-confirmed dementia. Neurobiol Aging. 2008 Aug;29(8):1143-59. Epub 2007 Apr 10 PubMed.
  3. . Inter-laboratory variation in cerebrospinal fluid biomarkers for Alzheimer's disease: united we stand, divided we fall. Clin Chem Lab Med. 2010 May;48(5):603-7. PubMed.
  4. . Comparison of analytical platforms for cerebrospinal fluid measures of β-amyloid 1-42, total tau, and p-tau181 for identifying Alzheimer disease amyloid plaque pathology. Arch Neurol. 2011 Sep;68(9):1137-44. PubMed.
  5. . Simultaneous measurement of beta-amyloid(1-42), total tau, and phosphorylated tau (Thr181) in cerebrospinal fluid by the xMAP technology. Clin Chem. 2005 Feb;51(2):336-45. Epub 2004 Nov 24 PubMed.
  6. . Diagnostic accuracy of ELISA and xMAP technology for analysis of amyloid beta(42) and tau proteins. Clin Chem. 2007 May;53(5):859-65. PubMed.
  7. . Development and advanced validation of an optimized method for the quantitation of aβ(42) in human cerebrospinal fluid. AAPS J. 2012 Sep;14(3):510-8. PubMed.

External Citations

  1. CE mark of quality
  2. Phase 1b AD trial
  3. Dominantly Inherited Alzheimer Network
  4. Alzheimer’s Disease Neuroimaging Initiative

Further Reading


  1. . Development and advanced validation of an optimized method for the quantitation of aβ(42) in human cerebrospinal fluid. AAPS J. 2012 Sep;14(3):510-8. PubMed.
  2. . Comparison of analytical platforms for cerebrospinal fluid measures of β-amyloid 1-42, total tau, and p-tau181 for identifying Alzheimer disease amyloid plaque pathology. Arch Neurol. 2011 Sep;68(9):1137-44. PubMed.
  3. . Standardization of preanalytical aspects of cerebrospinal fluid biomarker testing for Alzheimer's disease diagnosis: a consensus paper from the Alzheimer's Biomarkers Standardization Initiative. Alzheimers Dement. 2012 Jan;8(1):65-73. PubMed.

Primary Papers

  1. . Development and advanced validation of an optimized method for the quantitation of aβ(42) in human cerebrospinal fluid. AAPS J. 2012 Sep;14(3):510-8. PubMed.