We have described here a technology with which the entire complement of serum IgG antibodies can be screened against a peptoid library in order to identify complexes of disease-specific antibodies and individual peptoids. This approach to the discovery of serum antibody biomarkers differs from other efforts of which we are aware in that it makes no attempt to directly identify the native antigen or a close mimic thereof by screening libraries of peptides, lipids, proteins, nucleic acids or other naturally occurring molecules. Instead, the idea behind this approach is that it should be possible to use high-throughput screening to identify a synthetic, unnatural molecule that happens to have the right shape and chemical functionality to bind the antigen recognition pocket of the antibody of interest well enough to pull it out of serum, even if said molecule does not bind as well as the native antigen. While peptoids and peptides share an α-amino acid backbone, they are otherwise quite different in shape and chemical properties. For example, the side chain in peptoids protrudes from the main chain nitrogen, which is sp2
hybridized, whereas the peptide α-carbon is sp3
-hybridized. In addition, peptoids lack the N-H group in the main chain, which is often a contributor to hydrogen bond interactions that stabilize either peptide secondary structures or interactions with a partner binding protein. Finally, many of the side chains in the library of peptoids used in this study did not resemble the side chains of the 20 common amino acids (see Supp. Fig. 1
). Thus, this approach is quite distinct from previously reported screens of peptide libraries (Robinson et al., 2002b
; Wang et al., 2005
), which aim to identify a native epitope or at least a close relative. The peptoid molecules cannot possibly mimic a native peptide antigen closely.
As shown schematically in , this type of screening was done using a microarray format that allowed comparison of the binding of antibodies from case and control serum samples to thousands of peptoids. Molecules that retained far more IgG antibody from the case samples were considered candidate capture agents for IgG antibodies highly enriched in the disease state of interest. This protocol was first employed to test if peptoids could be identified that capture antibodies that distinguish between healthy mice and animals with EAE. The same analysis was applied to mice immunized with a peptide derived from Ovalbumin. In each case, three peptoids, called AMogP1-3 and AOvaP1-3, respectively, were identified that captured high levels of IgG antibody from the serum of the immunized animals, but not the control animals. As must have been the case (see above) the peptoids bear no obvious resemblance to the Mog and Ova peptide antigens, respectively (see Supp. Fig. S2
), even though it was demonstrated that these molecules indeed bind the antibodies raised against the peptide antigens ( and Supp. Fig. S4
). Subsequent validation studies with samples obtained from mice not used in the training set validated these peptoids as excellent capture agents for antibodies unique to the EAE and Ovalbumin peptide-immunized mice, respectively (Figs. and ).
Of course, the more important question is whether this approach is relevant to the discovery of peptoid-antibody complexes that might be of utility in medical diagnostics. While the mouse work was encouraging and proved the principle of using libraries of unnatural molecules to search for autoantibody ligands, this study employed relatively homogeneous laboratory mice and simple, single antigen immunization models. The greater immunological diversity between different people than between different laboratory mice might complicate the application of this technology to the discovery of biomarkers for human disease. To address this important question, we carried out a preliminary study of serum samples collected from patients with Alzheimer's Disease (AD). It has been reported that AD patients have lower levels of serum anti-amyloid antibodies than healthy individuals (Weksler et al., 2002
). While this difference is not sufficient to act as the basis of a diagnostic test, it does suggest the possibility of finding more useful antibody markers of the disease. Peptoid libraries were screened using serum samples from six AD patients (three autopsy-confirmed), six matched control individuals and six Parkinson's Disease (PD) patients. Three peptoids were identified that captured at least three-fold higher levels of IgG antibodies from all six of the AD patients than any of the controls or PD patients (Figs. and ). The structures of these peptoids are shown in . Depletion of ADP1-binding antibodies from the serum of AD samples demonstrated that ADP1 and ADP3 bind the same IgG antibodies, while ADP2 binds different antibodies (). Thus, we have discovered at least two candidate autoantibody biomarkers for AD.
Subsequent blinded studies were conducted using samples from more AD patients, controls (see Supplementary Table 1
) and patients with a different disease (lupus) to test the utility of the peptoids identified in the original screen. These “validation samples” were not employed in the training set and thus constitute a fair and critical test of the utility of the peptoid-antibody complexes as biomarkers (Ransohoff, 2005
). Once unblinded, the data (see Figs. and ) showed that these peptoid antibody complexes are indeed highly promising biomarkers for the diagnosis of AD (see Supp. Fig. S5
and Supp. Tables S2-6
for the results of a comprehensive statistical analysis).
Two of the control individuals displayed a relatively high level of the AD antibodies (NC31 and 41), similar to that seen in the AD patient with the lowest levels of antibodies (AD15)(see ). The individual from whom the NC31 sample was collected is a 75 year-old female with a mini-mental state examination (MMSE) score (McKhann et al., 1984
) of 29 out of a possible 30 and without obvious clinical signs of AD (Supplementary Table 1
). The NC41 sample was from a 65 year-old female with the same MMSE score. These may represent examples of false positives, for example due to the cross-reaction of non-AD associated antibodies with these peptoids, or could represent pre-symptomatic detection of developing disease. Since these samples contained high levels of antibodies that bind to ADP2 as well as antibodies that bind to ADP1 and ADP3, we favor the latter hypothesis, but this cannot be concluded with certainty.
The development of a simple blood test for AD is an important unrealized goal (Blennow et al., 2010
). This preliminary study is promising in that it represents a high level of diagnostic sensitivity and specificity (Saah and Hoover, 1997
), at least within the relatively limited range of samples analyzed. However, it is important to point out that more work will be required before it is clear whether the peptoids ADP1-3 will be useful reagents for the clinical diagnosis of AD. First, the analysis of a larger number of patient samples derived from a more diverse population will be required. Second, it will be important to test samples collected from patients with mild cognitive impairment (MCI) that have gone on to develop confirmed AD, since early detection is an important clinical goal. Third, all of the measurements done in this study were conducted on a microarray platform that may not be easily employed in a clinical setting, so optimized conditions for using the peptoids on other analytical platforms will have to be developed. Fourth, if these biomarkers are indeed validated then it will be of great interest to identify the native antigens that they recognize. Studies to address all of these issues are in progress. The experiments in this paper were designed solely to address the issue of whether this technology is applicable to the discovery of biomarker candidates for human disease. We conclude that this is indeed the case.
In summary, we have developed and validated a technology based on parallel screens of synthetic combinatorial libraries for the discovery of IgG biomarkers and simple, synthetic capture agents capable of retaining them from serum. We believe that this technology will have a significant impact on the development of diagnostic tests for a variety of important diseases.