When looking in PubMed, one sees that there is an exponential increase in the number of studies on biomarkers for AD, and the opportunities should be excellent for making meta-analyses on the topic. However, if one sets out to curate the literature, it will rapidly become clear that there is no standardized way of reporting how tests were performed, on what patients, how samples were handled, how cut-points were determined, etc. This is a well-known problem in other fields of medicine, which stimulated the creation of the STARD (STAndards for the Reporting of Diagnostic accuracy studies) criteria some years ago.

The work now to adapt the STARD criteria to studies on neurodegenerative diseases is very important. In my mind, the first draft of the STARDdem document was rather naïve in respect to what is important to know for fluid biomarkers, which is my specialty. For imaging biomarkers and other measures to help in making the diagnosis, the situation may be the same. I think that it is very important that specialists respond to the invitation by the STARDdem authors to read...  Read more

When looking in PubMed, one sees that there is an exponential increase in the number of studies on biomarkers for AD, and the opportunities should be excellent for making meta-analyses on the topic. However, if one sets out to curate the literature, it will rapidly become clear that there is no standardized way of reporting how tests were performed, on what patients, how samples were handled, how cut-points were determined, etc. This is a well-known problem in other fields of medicine, which stimulated the creation of the STARD (STAndards for the Reporting of Diagnostic accuracy studies) criteria some years ago.

The work now to adapt the STARD criteria to studies on neurodegenerative diseases is very important. In my mind, the first draft of the STARDdem document was rather naïve in respect to what is important to know for fluid biomarkers, which is my specialty. For imaging biomarkers and other measures to help in making the diagnosis, the situation may be the same. I think that it is very important that specialists respond to the invitation by the STARDdem authors to read and comment on the draft to make the first public version as good as possible.

View all comments by Henrik Zetterberg

This is an exciting and timely project that will benefit scientific research and clinical practice in the long run.

One difficulty I can foresee is that dementia is a long, pathological process rather than an event. So it would be very hard to define a "standard" point of "conversion" from "MCI" or "pre-AD" to dementia. The distinction between MCI and dementia depends on an individual's "function" compared to a previous level. The comparison is often subject to clinical judgment, and, in certain contexts, making a good judgment can be very difficult.

"MCI" also suffers from lack of a standard "operational" definition. For example, one person could be labeled as amnestic MCI, non-amnestic MCI/single- or multiple-domain MCI, based on poor performance on neuropsychological tests, but in reality, it is almost impossible to standardize the number of tests, types of tests, and which norm to be compared to in defining MCI.

View all comments by Lei Feng

There is a similar open project entitled Use of Biomarkers for Neurodegenerative Disorders in the Clinical Setting. It is devoted to all neurodegenerative diseases, including those with dementia. It is funded by the Neurosciences Foundation, a Spanish not-for-profit organization. Anyone with interest in the field is invited to join.

View all comments by Carlos Vázquez

Our article (Ray et al., 2007) gained a lot of attention, but it was very early days and we had to work with what was available. Our samples were from multiple centers, and the cases and controls were not perfectly matched for each. There was also a difference in age between cases and controls, and the analytical platform we had used was a somewhat moving target, because the manufacturer (RayBiotech) made several changes to the array during the time we used it. Nevertheless, I think several of the markers we identified have biological relevance in AD and brain aging, and we are pursuing some of them successfully (e.g., MCSF). I would also draw attention to work from our lab that has been overlooked (Britschgi et al., 2011). We used an independent set of samples, a different analytical platform, and an innovative new approach to predict pathological parameters in AD using plasma markers as variables. Several models we developed reproduced six proteins out of the 18-protein Ray signature....  Read more

Our article (Ray et al., 2007) gained a lot of attention, but it was very early days and we had to work with what was available. Our samples were from multiple centers, and the cases and controls were not perfectly matched for each. There was also a difference in age between cases and controls, and the analytical platform we had used was a somewhat moving target, because the manufacturer (RayBiotech) made several changes to the array during the time we used it. Nevertheless, I think several of the markers we identified have biological relevance in AD and brain aging, and we are pursuing some of them successfully (e.g., MCSF). I would also draw attention to work from our lab that has been overlooked (Britschgi et al., 2011). We used an independent set of samples, a different analytical platform, and an innovative new approach to predict pathological parameters in AD using plasma markers as variables. Several models we developed reproduced six proteins out of the 18-protein Ray signature. Most consistently, we found changes in MCSF, GCSF and IL-3.

Still, I think even now there are major challenges to find markers that will hold up in multiple studies across different centers and become clinical tools. It took maybe 10 years to achieve clinical utility with CSF Aβ and tau ELISAs, and I think it will take as long with any other protein-based assay (one at a time).

The main problem is that protein measurements are extremely difficult to standardize, and multiplex assays are notoriously inexact. Major problems with current assays are that the reagents (antibodies, standards) are "research use only" and not clinical grade. They may, therefore, change from batch to batch, leading to variations in sensitivity and absolute concentrations for a given protein. Another, more trivial problem is that assays (e.g., ELISAs, Luminex) from different manufacturers may detect different isoforms of the same protein, active versus pre-proteins, or post-translationally modified proteins versus unmodified, leading sometimes to completely opposite results between groups.

I think we are at a similar stage in this field as genetics was with SNP studies 10 years ago. Geneticists produced lists of more than 100 genes with linkage to AD, of which most did not hold up in the much larger GWAS. This showed that sample size is key. However, even if thousands of blood samples will be analyzed, we will still have the problem that the protein assays and sample collection are not standardized.

Our lab continues to develop and use protein screens, and we currently measure more than 600 proteins in blood plasma or CSF using antibody-based microarrays. We know these arrays produce false-positive and -negative results, but we run several hundred samples in one batch to reduce variability. We have identified several interesting new proteins and pathways that we are now validating in biological assays and animal models of AD. I think we will have to go this hard way and link biology to any of the proteins that come out of screens before they are worth the effort to produce a clinical-grade assay.

View all comments by Tony Wyss-Coray