Having established the first 42 amino acids in amyloid-β as a valid biomarker for Alzheimer’s disease, researchers are now turning their attention to all the other fragments of amyloid precursor protein (APP) in the cerebrospinal fluid (CSF). Two pilot studies published in June suggest that other APP peptides could also serve as biomarkers, and although it is too early to be sure, they might even work better than amyloid-β(1-42).

“Aβ42 has been our workhorse,” said Anne Fagan of Washington University in St. Louis, Missouri. “But that is not to say that there might not be more things that are even more informative,” added Fagan, who was not involved in either study.

As a biomarker, “The problem with Aβ itself is that it actually goes down; we think it is aggregating in the brain so Aβ42 levels in the CSF go down,” said Michael Wolfe of Harvard University. He explained that Aβ42 exists in two states—insoluble in the brain and soluble in the CSF, making it impossible to measure total Aβ42 with a spinal tap. But a marker that is always soluble would be fully accessible from CSF, perhaps yielding more accurate results. Wolfe was also not an author in the current works.

First authors Adnan Halim and Gunnar Brinkmalm, along with senior authors Göran Larson and Jonas Nilsson, led one study at the University of Gothenburg, Sweden. As they write in this week’s Proceedings of the National Academy of Sciences USA online, the group happened upon a rare, small, glycosylated, amino-terminal fragment of Aβ. The researchers were surprised to discover that the sugar group had set down not on a threonine or serine amino acid—as sugar groups generally do—but on a tyrosine. Studying this unexpected peptide in CSF samples, the researchers discovered that it was more abundant in fluid from people who had Alzheimer’s than in those who were cognitively normal.

Robert Perneczky of Technical University in Munich, Germany, led a study published in the June 22 Neurology online. He and his colleagues examined another potential biomarker, soluble APP (sAPP). Cleavage of APP, by α- or β-secretase, produces sAPP from the amino end of the precursor. Although other researchers have explored sAPP as a potential biomarker (Lewczuk et al., 2010), Perneczky’s work is the first to take a longitudinal approach, with an average follow-up time of three years. According to his statistical analysis, Perneczky said, sAPP was a better predictor of progression to AD than the standard Aβ42.

What Is That Sugar Doing There?
Nilsson and Larson recently developed a method to capture and identify glycoproteins (Nilsson et al., 2009). Halim was experimenting with the method in a CSF sample when he discovered a set of short, glycosylated Aβ fragments, 15-20 amino acids in length. To acquire enough of these bits to study in detail, the researchers used an antibody to pull down all forms of Aβ from CSF. They saw dozens of different-sized fragments, both glycosylated and not, but the team chose to focus on the small, glycosylated peptides because they were novel. “They are short and most likely not aggregation-prone,” said coauthor Henrik Zetterberg. These particular peptides made up less than 1 percent of the total Aβ species in the sample.

Using mass spectrometry, the researchers discovered that their fragments of interest came in six different lengths, always starting with amino acid 1 in the Aβ sequence and ending with any of amino acids 15-20. Since sugar groups normally attach to serines and threonines, they suspected serine-8 would be glycosylated. Instead, analysis showed it was on the tyrosine-10.

“Many people did not think such glycosylations existed,” Zetterberg said. There is no enzyme known to glycosylate tyrosine, and only a couple of examples of proteins glycosylated at this amino acid (Smythe et al., 1988; Aon and Curtino, 1985; Zarschler et al., 2010).

The authors were unable to determine the exact structure of the sugar group, which could be based on glucose, galactose, or mannitol, they write. Another unanswered question, Zetterberg said, is whether this glycosylation of Aβ is rare or commonplace, and whether it occurs only in the nervous system or in other cell types as well.

The scientists did not find Aβ42 glycosylated on tyrosine-10, suggesting to Zetterberg that the modification takes place before APP has been processed by γ-secretase. “Most likely the glycosylation occurs before the cleavage,” Zetterberg said, because glycosylation is an intercellular event that would have to occur before APP reaches the plasma membrane, where the first cleavage occurs.

Zetterberg hypothesizes that APP, once glycosylated at the cell surface, could be prevented from reentering the cell and encountering the γ-secretase that would normally cleave it. Alternatively, the glycosylation could conceivably alter the protein’s shape, so that after it reenters the cell, γ-secretase cuts it not at amino acid 42, but in the teens. Glycosylation has been shown to regulate proteolysis in a handful of systems (Schjoldager et al., 2010; Semenov et al., 2010; May et al., 2003; Maryon et al., 2007). Thus, the glycosylation could block or alter the pathway leading to Aβ42, explaining why those were never found glycosylated.

The authors wondered if the glycosylated peptides could correlate with Alzheimer’s disease. Using samples they had on hand, they measured the amounts of the fragments in CSF from six people with AD and seven healthy controls. They discovered that the relative abundance of the tyrosine-10 glycosylated fragments was increased by a factor of 1.1-2.5, in the AD samples than in the control fluids.

“I think this is a really critical modification of the protein,” Larson said. But he does not know if the glycosylation is pathogenic or compensatory.

Wolfe speculated that some of the small fragments could be products of both α- and β-secretase cleavage together. Normally, only one or the other cleaves APP, although Zetterberg and Kaj Blennow, another coauthor of the PNAS study, recently found that they could cleave the same protein (see ARF related news story on Portelius et al., 2010).

If the glycosylation does prevent Aβ42 production, Zetterberg acknowledges, then one might predict the glycosylated fragments would be downregulated in AD in favor of the larger, classical amyloid-β fragment. In fact, the scientists found the opposite. More studies, he hopes, will clarify the situation as well as help determine if glycosylated Aβ fragments could serve as a biomarker for AD. To use this fragment as a diagnostic or prognostic test, Zetterberg noted, doctors would need a less “cumbersome” method than antibody enrichment and mass spectrometry.

Begin at the Beginning
Aβ42 and tau levels in CSF are a commonly used biomarker for AD. But in looking at Aβ42, researchers are examining the very last product of APP cleavage. Instead, Perneczky suggests in his work, it could be more predictive to examine the very first step, the production of sAPP (which can be sAPPα or sAPPβ, depending on which secretase cleaves it from APP). Perneczky found similar trends for both sAPPα and sAPP-β, but the latter was more predictive. Cleavage at the α position prevents Aβ generation because it destroys the Aβ sequence.

The Munich team examined a small cohort of 58 people who had mild cognitive impairment (MCI). They followed them for an average of three years, assessing whether they progressed to Alzheimer’s, remained at the MCI stage, or returned to normal cognition.

Of those who went on to full-blown AD, their CSF samples had more sAPP-β—approximately 1,200 ng/ml—than their counterparts who had did not (900 ng/ml). “The biomarker, in combination with tau, was superior to the established combination of tau plus Aβ42,” Perneczky said. Considering sAPPβ, tau, and age, the researchers predicted who would or would not progress to AD with 80 percent sensitivity and 81 percent specificity. The team is now examining whether blood sAPP-β levels correlate with AD.

Looking at sAPP is a “worthwhile” pursuit, Larson said. However, Zetterberg noted that the researchers included no subjects who were cognitively healthy to start with. The rise and fall of sAPP could be due to many reasons, he suggested. Perneczky, now working to replicate the results, is including a cognitively normal group, he said.—Amber Dance


  1. This is an interesting article to read, and the supporting documents contain a wealth of information about the reported findings. Tyrosine O-glycosylation is unusual, but the data demonstrated clearly that it actually occurred in this instance. The authors characterized the glycosylation sites of amyloid precursor proteins (APP) and amyloid β (Aβ) for the first time, and the data they provided demonstrated that clearly. It is interesting that Tyr10 O-glycosylation was not identified in APP, but was identified in shorter N-terminal Aβ1-X, with X equal to or less than 20.

    Meanwhile, the same Tyr10 O-glycosylation was not identified in all of the full-length Aβ isoforms. I wish the authors had showed how they came up with a scale of the heat map for the spectral relative intensities of peptide glycosylation. I was wondering if the scale was done with calibration peptide standards, or if it was based solely on the response of the mass spectrometer used in the study.

    Anyhow, the heat map seemed to illustrate the difference in intensities of peptides with Tyr10 O-glycosylation compared with those without it. The finding does suggest that Tyr10 O-glycosylation occurred primarily in samples obtained in Alzheimer's disease (AD) patients; however, data from at least one of the non-AD patients also showed increased Tyr10 O-glycosylation. The borderline of who has AD, and who has not, was not clear according to the findings. Relying on the 530 pg/ml Aβ1-42 concentration was insufficient to separate the patient population as stated in this article or the 500 pg/ml that was reported elsewhere (1). Unfortunately, the shortcoming of this study is that the sample size was limited, and a deviation of one out of six non-AD patient results becomes statistically significant.


    . Human apoE isoforms differentially regulate brain amyloid-β peptide clearance. Sci Transl Med. 2011 Jun 29;3(89):89ra57. PubMed.

    View all comments by Kwasi Mawuenyega
  2. Perneczky and colleagues report that CSF measurement of sAPPβ may be useful and superior to Aβ1-42 in the early and differential diagnosis of incipient (preclinical) Alzheimer's dementia. These data validate our earlier observations (Lewczuk et al., 2010), which also revealed a significant elevation of sAPPβ in incipient AD that was as pronounced as in early Alzheimer's dementia (AD). In both studies, other dementias did not show elevated CSF sAPPβ. Whereas in our study, the clinical diagnosis and stratification of mild cognitive impairment (MCI) patients was cross-validated by other CSF dementia biomarkers (total tau; phospho-tau181; Aβ1-42), Perneczky and colleagues used the clinical course of the patients (conversion to AD/MCI-AD; stable MCI; or conversion to cognitive normal/non-AD, i.e., MCI-NAD) to stratify the different MCI subgroups at baseline. Moreover, both groups applied different quantitative immunoassays to measure sAPPα and sAPPβ, whereas the same assays were used to measure CSF total tau and Aβ1-42.

    The observations of Perneczky and colleagues are of high clinical relevance; however, some critical issues have to be addressed.

    Though the MCI-AD and MCI-NAD groups show a significant age difference, age was included in the statistical discriminator model (sAPPβ, total tau, age). The authors do not report if overall sAPPβ and age are significantly correlated. Surprisingly, Aβ1-42 was not significantly lowered in MCI-AD as compared to MCI-NAD or patients with frontotemporal dementia. The available literature clearly indicates that CSF Aβ1-42 is already decreased in incipient AD (high-risk MCI). Since no further control group is provided, these data are hard to interpret but may question the conclusion that sAPPβ is superior to Aβ1-42 in the early and differential diagnosis of incipient AD. In this respect, and in general, an independent statistical analysis of the MCI-AD subgroup of patients who converted back to cognitive normal (n = 8) may have helped to clarify this issue. Since the mean clinical follow-up time was comparatively short (33.1 months), slow converters (MCI to AD) may be hidden within the MCI-NAD cohort. Perneczky et al. did not report data on CSF values of phosphorylated tau (e.g., phospho-tau181). Together with total tau, these data could have been helpful to define a subgroup of slow MCI to AD converters.

    A puzzling difference between the data reported by the two groups is that we also found significantly elevated CSF sAPPα in MCI-AD as compared to MCI-NAD. This may be explained by the two different phenotyping strategies applied to stratify the patient subgroups at baseline. However, this may also be due to the different assays which were applied in the two independent studies. Both groups reported a strong correlation of sAPPβ and sAPPα; however, we measured much higher values of sAPPα relative to sAPPβ (Meso Scale Discovery, Gaithersburg, USA; multiplex assay) as compared to Perneczky and colleagues (IBL Gunma, Japan; monoplex ELISA). This assay-dependent difference is unlikely to be due to cross-reactivity within the multiplex assay, since we controlled for cross-reactivity by independent Western blot analysis.

    In summary, the data of Perneczky and coworkers are very promising, but they need further validation regarding the claim of a diagnostic superiority of sAPPβ relative to Aβ1-42 and the performance of sAPPβ relative to sAPPα. Moreover, to approximate the true predictive value of sAPPβ for incipient AD, longer clinical follow-up times should be applied for age-matched MCI subgroups. Together with CSF data on elevated BACE protein content and enzyme activity in preclinical AD (Zhong et al., 2007; Zetterberg et al., 2008), it is tempting to speculate that β-secretase activity is also increased in early (preclinical) multigenetic AD, as known for genetic AD (Bateman et al., 2011). This finding would reinforce β-secretase as a potential therapeutic target for the preventive drug treatment of AD in high-risk MCI patients. However, total CSF Aβ—unlike sAPPβ—is not consistently reported to be elevated in preclinical AD as a consequence of elevated β-secretase activity. In the event that increased Aβ peptide generation is initially paralleled by increased oligomerization, resulting in epitope masking (soluble oligomers) and subsequent precipitation into insoluble Aβ, this may not be a contradiction.


    . Soluble amyloid precursor proteins in the cerebrospinal fluid as novel potential biomarkers of Alzheimer's disease: a multicenter study. Mol Psychiatry. 2010 Feb;15(2):138-45. PubMed.

    . Levels of beta-secretase (BACE1) in cerebrospinal fluid as a predictor of risk in mild cognitive impairment. Arch Gen Psychiatry. 2007 Jun;64(6):718-26. PubMed.

    . Elevated cerebrospinal fluid BACE1 activity in incipient Alzheimer disease. Arch Neurol. 2008 Aug;65(8):1102-7. PubMed.

    View all comments by Jens Wiltfang

Make a Comment

To make a comment you must login or register.


News Citations

  1. Sweet 16: Novel APP Processing Pathway and a New Biomarker?

Paper Citations

  1. . Soluble amyloid precursor proteins in the cerebrospinal fluid as novel potential biomarkers of Alzheimer's disease: a multicenter study. Mol Psychiatry. 2010 Feb;15(2):138-45. PubMed.
  2. . Enrichment of glycopeptides for glycan structure and attachment site identification. Nat Methods. 2009 Nov;6(11):809-11. PubMed.
  3. . Isolation and structural analysis of a peptide containing the novel tyrosyl-glucose linkage in glycogenin. EMBO J. 1988 Sep;7(9):2681-6. PubMed.
  4. . Protein-bound glycogen is linked to tyrosine residues. Biochem J. 1985 Jul 1;229(1):269-72. PubMed.
  5. . Protein tyrosine O-glycosylation--a rather unexplored prokaryotic glycosylation system. Glycobiology. 2010 Jun;20(6):787-98. PubMed.
  6. . O-glycosylation modulates proprotein convertase activation of angiopoietin-like protein 3: possible role of polypeptide GalNAc-transferase-2 in regulation of concentrations of plasma lipids. J Biol Chem. 2010 Nov 19;285(47):36293-303. PubMed.
  7. . Processing of pro-B-type natriuretic peptide: furin and corin as candidate convertases. Clin Chem. 2010 Jul;56(7):1166-76. PubMed.
  8. . Differential glycosylation regulates processing of lipoprotein receptors by gamma-secretase. J Biol Chem. 2003 Sep 26;278(39):37386-92. PubMed.
  9. . O-linked glycosylation at threonine 27 protects the copper transporter hCTR1 from proteolytic cleavage in mammalian cells. J Biol Chem. 2007 Jul 13;282(28):20376-87. PubMed.
  10. . A novel Abeta isoform pattern in CSF reflects gamma-secretase inhibition in Alzheimer disease. Alzheimers Res Ther. 2010;2(2):7. PubMed.

Further Reading


  1. . Gene expression profiling of peripheral blood leukocytes identifies and validates ABCB1 as a novel biomarker for Alzheimer's disease. Neurobiol Dis. 2011 Sep;43(3):698-705. PubMed.
  2. . Longitudinal change of biomarkers in cognitive decline. Arch Neurol. 2011 Oct;68(10):1257-66. PubMed.
  3. . Long-term stability of Alzheimer's disease biomarker proteins in cerebrospinal fluid. J Alzheimers Dis. 2011;26(2):255-62. PubMed.
  4. . Amyloid imaging as a surrogate marker in clinical trials in Alzheimer's disease. Q J Nucl Med Mol Imaging. 2011 Jun;55(3):265-79. PubMed.

Primary Papers

  1. . CSF soluble amyloid precursor proteins in the diagnosis of incipient Alzheimer disease. Neurology. 2011 Jul 5;77(1):35-8. Epub 2011 Jun 22 PubMed.
  2. . Site-specific characterization of threonine, serine, and tyrosine glycosylations of amyloid precursor protein/amyloid beta-peptides in human cerebrospinal fluid. Proc Natl Acad Sci U S A. 2011 Jul 19;108(29):11848-53. PubMed.