The development of global proteomic profiling in the mid-1990 s raised the expectations for quick discovery of new biomarkers [1]. More importantly, it was expected that profiling of body fluids using high throughput, sensitive and specific methods would result in bringing new and approved diagnostic and therapeutic biomarkers from bench to bedside in a fast track manner [2]. However, soon after the first large profiling experiments were performed, researchers observed several major problems: (i) very high dynamic range of the expression of proteins in the body fluids can reach 10¹² orders of magnitude, thereby excluding the possibility to quantitate both low and high abundance proteins without additional sample fractionation(s) [3]; (ii) range of concentration for any given protein varies from individual to individual in general population as well as in cohorts of patients; (iii) standard operating procedures - including sample preparation, mass spectrometers used, and bioinformatic database searching - varied between proteomic labs, resulting in variability and only partial overlap of results [4]; and (iv) orthogonal validation of biomarkers in body fluids is essential following discovery phase, however these methods often fail to confirm initial results [5].

Of all the issues listed above, several are beyond our control and others require more technological development; validation of quantitative proteomics data is one such issue requiring advancement [6, 7]. Examples of orthogonal validation techniques for MS-based proteomics include Enzyme Linked ImmunoSorbent Assay (ELISA) [8–10] and Western blot [11, 12]. In comparison, examples of parallel validation techniques include Stable Isotope Standards and Capture by Anti-Peptide Antibodies (SISCAPA) [13, 14] and Multiple Reaction Monitoring (MRM) [15, 16]. Each technique has advantages and drawbacks for the validation of potential biomarkers. For example, orthogonal validation using Western blot or ELISA requires the use of antibodies; some of which are not well characterized and when used, may result in misleading or skewed data.

Proteomic studies from our laboratory have shown that gelsolin is differentially expressed in the plasma and Cerebrospinal Fluid (CSF) of Human Immunodeficiency Virus (HIV)-infected individuals with and without dementia [17–19]. Likewise, gelsolin circulating in the plasma of monkeys infected with simian immunodeficiency virus (SIV) is also differentially expressed between pre-infection, acute and chronic infection [19]. We have also found that monocyte derived macrophage (MDM) activated by HIV infection in vitro produce and secrete gelsolin (Ciborowski, P.; Kraft-Terry, S. both unpublished). Taking this together, we postulated that if gelsolin is validated, it may become a candidate as a diagnostic biomarker and be justified to move to experiments using larger cohorts of patients. However, validation of the differential expression of gelsolin in body fluids occurred to be a challenging task, as quantitative Western blot did not confirm differential expression unambiguously. As further studies indicated this was caused by two major reasons. First, high variability in the immunoreactivity of commercially available antibodies and the variability in recognition of gelsolin originating from CSF or plasma resulted in ambiguity. Second, immunoaffinity purification of gelsolin followed by MS/MS revealed that although the gelsolin circulating in the plasma and CSF was the secreted form of gelsolin (plasma gelsolin; pGSN), several other forms in addition to the full-length molecule (86kDa) were also in circulation ranging in molecular weight from 10 kDa to 188 kDa [20]. Based on these prior studies and observations, this study focused on problems with validation of gelsolin using antibody based orthogonal assays.

Previously published MS-based proteomic studies have shown that plasma gelsolin (pGSN) is differentially expressed in HIV infected humans, SIV infected monkeys and in vitro HIV infected MDM [18–20, 23]. Prior to conducting further studies using larger cohorts of samples from patients, we attempted to validate its expression using a smaller number of samples. Our validation effort using quantitative Western blot analysis gave ambiguous results and indicated that differences in validation strongly depend on which antibody was used. Therefore, the initial goal of our study was to select an anti-gelsolin antibody that when used for quantitative Western blot analysis would most closely reflect the results of proteomic profiling.

Specificity of anti-gelsolin antibodies

Subsequent experiments brought to light new information concerning our previous results of Western blot validations [25] in which we observed a single band corresponding to the full-length gelsolin molecule. Concurrent experiments of immunoaffinity purification from the same samples showed multiple forms of gelsolin. This discrepancy prompted us to further explore the specificity of anti-hGSN antibodies to explain if Western blot validation of proteomic profiling might be biased and do not reflect real levels of intact and processed forms of gelsolin. From more than 20 commercially available anti-hGSN antibodies, we selected three: mouse monoclonal, goat polyclonal and rabbit polyclonal; all were raised against an epitope from the C-terminal portion of gelsolin. A sheep polyclonal anti-hGSN antibody was also tested, but results could not be obtained due to high background (data not shown). Figure 1A is a schematic diagram of the gelsolin molecule along with the location of the epitope corresponding to each antibody. Figure 1B summarizes the specificity of those three anti-hGSN antibodies used for Western blot assay against the gelsolin found various samples. It was unexpected that that monoclonal antibody raised to a synthetic peptide (located in the C-terminal end of gelsolin) reacted only with a single band of approximately 86 kDa, corresponding to the full length and intact gelsolin molecule. This indicated the conformational dependence of the antibody; moreover, it suggested that the conformation of the entire molecule - not only its C-terminal portion - is necessary for its immunoreactivity. Furthermore, Western blot analysis using either goat or rabbit pAb recognized additional forms of gelsolin with both larger and smaller molecular weights. It is likely that the smaller molecular weight forms are truncated forms of the full-length molecule based on trends observed by LC/ESI-MS/MS (Table 1). Further analysis using LC/ESI-MS/MS revealed that the larger molecular weight form (Figure 2A, asterisked band) contains fibronectin, which co-immunopurified with gelsolin; this comes as no surprise as it is well known that fibronectin binds gelsolin [26].

Gel band	M.W.*	Theoretical minimum M.W. **	Peptide sequence	Peptide position in pGSN
Band 1	86 kDa	65816.00 kDa	EVQGFESATFLGYFK	121 - 135
			HVVPNEVVVQR	151 - 161
			PALPAGTEDTAKEDAANR	251 - 268
			QTQVSVLPEGGETPLFK	347 - 363
			DPDQTDGLGLSYLSSHIANVER	371 - 392
			AGALNSNDAFVLK	558 - 570
			TPSAAYLWVGTGASEAEK	571 - 588
			AQPVQVAEGSEPDGFWEALGGK	600 - 621
			DSQEEEKTEALTSAK	687 - 701
			RYIETDPANR	702 - 711
			RTPITVVK	714 - 721
Band 2	83 kDa	50126.19 kDa	PALPAGTEDTAK	251 - 262
			QTQVSVLPEGGETPLFK	347 - 363
			DPDQTDGLGLSYLSSHIANVER.V	371 - 393
			AQPVQVAEGSEPDGFWEALGGK.A	600 - 621
			DSQEEEKTEALTSAK	687 - 701
			YIETDPANR	703 - 711
Band 3	64 kDa	64805.80 kDa	R.EVQGFESATFLGYFK.S	120 - 136
			K.PALPAGTEDTAK.E	250 - 263
			K.QTQVSVLPEGGETPLFK.Q	346 - 364
			K.DSQEEEKTEALTSAK.R	686 - 702
			R.YIETDPANR.D	702 - 712
			-.YIETDPANR.-	703 - 711
Band 4	60 kDa	49194.23 kDa	K.PALPAGTEDTAK.E	250 - 263
			R.DPDQTDGLGLSYLSSHIANVER.V	370 - 393
			K.AGALNSNDAFVLK.T	557 - 571
			R.AQPVQVAEGSEPDGFWEALGGK.A	599 - 622
			K.DSQEEEKTEALTSAK.R	686 - 702
Band 5	54 kDa	41077.02 kDa	K.QTQVSVLPEGGETPLFK.Q	346 - 364
			R.DPDQTDGLGLSYLSSHIANVER.V	370 - 393
			K.VPVDPATYGQFYGGDSYIILYNYR.H	430 - 455
			K.AGALNSNDAFVLK.T	557 - 571
			-.TGAQELLR.-	589 - 596
			R.AQPVQVAEGSEPDGFWEALGGK.A	599 - 622
			K.DSQEEEKTEALTSAK.R	686 - 702
			-.RTPITVVK.-	714 - 721
Band 6	45 kDa	36044.42 kDa	R.VPFDAATLHTSTAMAAQHGMDDDGTGQK.Q	392 - 421
			K.VPVDPATYGQFYGGDSYIILYNYR.H	430 - 455
			K.AGALNSNDAFVLK.T	557 - 571
			K.TPSAAYLWVGTGASEAEK.T	570 - 589
			R.AQPVQVAEGSEPDGFWEALGGK.A	599 - 622
			K.DSQEEEKTEALTSAK.R	686 - 702
			R.RYIETDPANR.D	701 - 712
			R.YIETDPANR.D	702 - 712
			R.RTPITVVK.Q	713 - 722
Band 7	27 kDa	30525.25 kDa	VPVDPATYGQFYGGDSYIILYNYR	431 - 454
			AGALNSNDAFVLK	558 - 570
			RYIETDPANR	702 - 711
Band 8	19 kDa	29308.93 kDa	VPVDPATYGQFYGGDSYIILYNYR	431 - 454
			DSQEEEKTEALTSAK	687 - 701

Included is the band identification (corresponding to extracted bands in Figure 1, column A), molecular weight based on electrophoretic mobility, theoretical minimum molecular weight as calculated using ExPASy Compute pI/Mw tool, identified peptides, and peptide location in secreted pGSN. The peptide -DSQEEKTEALTSAK- was the most commonly identified peptide (in 7 of 8 bands). * - Molecular weight (M.W.) is approximate based on electrophoretic mobility in 1DE SDS-PAGE. ** - Theoretical Molecular weight was approximated using the ExPASy Compute pI/Mw tool and was calculated using the first peptide position through the last peptide position as determined using MS/MS.

It was most unexpected that goat and rabbit pAb showed such great differences in immunoreactivity within samples. For example, goat pAb reacted weakly with only one form of gelsolin in the CSF whereas rabbit pAb recognized strongly two forms of gelsolin in that same sample. None of these antibodies recognized a form of gelsolin in any sample with a molecular weight lower than 45 kDa with the exception of goat pAb against commercially available human pGSN (Sigma Aldrich), which was able to recognize an additional band with a molecular weight of less than 38 kDa. In addition to the lower molecular weight forms, there were several samples in which higher molecular weight forms were detected; goat anti-hGSN pAb was able to detect these higher molecular weight forms in both human plasma and commercial gelsolin samples. Protein purification and concentration can often cause proteins to aggregate, which is a potential explanation for these higher molecular weight bands. These higher molecular weight bands warranted further investigation and using LC/ESI-MS/MS on bands excised from recombinant gelsolin, we were able to positively identify only gelsolin.

Immunoreactivity of gelsolin forms

Figure 3 includes a Western blot using goat anti-hGSN antibodies against a linear dilution of recombinant gelsolin. In the most concentrated sample (2 μg) of GSN, a total of 3 clear and distinguishable bands were detected: >188, 60 and 54 kDa both representing different forms of pGSN. Also detected were two bands at 166 kDa and 86 kDa; these bands however were not clear and distinguishable, but instead were oversaturated and unquantifiable. A 7-fold dilution (0.016 μg) of GSN resulted in only one clear, distinguishable and quantifiable band at 86 kDa. It was determined that immunodetection using goat anti-hGSN is dependent on the concentration of each form present in the sample.

Based on the banding pattern observed, peptides recognized and their location, molecular weight observed for each band and the calculated theoretical minimum molecular weight, a schematic for each band was created (Figure 4).

Biomarker discovery and validation - or even the complete characterization - of the plasma and/or proximal fluids (including CSF) has been a daunting task. New biomarkers have not emerged as expected, despite the effort put forth experimentally by both small, single laboratories with limited clinical samples [3, 13, 27] as well as large, research organizations like the Human Plasma Proteome Project (HPPP) [28]. A conundrum has emerged with respect to validation of biomarkers following the discovery phase; is a lack of validation due to the assumptions that have been made about how a particular disease progresses or is it that the tools and reagents used are not adequate to the task? Accordingly, we postulate that a better understanding of the molecular mechanisms underlying diseases will help us to understand observed changes at the protein level and will also result in the validation of already discovered as well as new biomarker candidates.

For the majority of studies validation is based entirely or in part on immunoreactivity of specific antibodies. In terms of ELISA, this approach has been proven as very useful and accurate in many instances, e.g. measurement of bacterial proteins/toxins etc. However, when the dynamic changes of human proteins are measured or taken into consideration (including alternative splicing, post translational modification, regulated processing or degradation), immunoreactivity based assays become quite inaccurate. There has been and will continue to be attempts to improve these assays and thereby promote the acceptance of scientifically sound biomarkers. For example, Rifai et al. proposed a biomarker "pipeline" including discovery, qualification, verification, assay optimization, validation and commercialization to help aid in the discovery of better protein biomarkers [29]. Several of steps in this pipeline require the use of antibodies - from immunoaffinity purification in the qualification and verification stages to immunoassays like Western blot and ELISA in validation.

Experimental data that we present in this study helps to understand why in many cases validation based on immunoreactivity may lead to inconclusive or even misleading results. Moreover, we also conclude that other methods such as MRM may provide inaccurate results in the validation of biomarkers. For example, quantitative methods requiring the use of antibodies such as ELISA and quantitative Western blot will vary depending on which antibody is used (Figure 1 and Figure 2). Additionally, it was shown that the concentration of the biomarker in question might also adversely affect the results of quantitation using Western blot (Figure 3). With respect to quantitation, the different immunoaffinity purification methods used in sample preparation must also be considered. To date, our laboratory has yet to witness non-specific removal of gelsolin forms, however it has been realized that using different methods of immunoaffinity purification may adversely affect the quantitation of protein. Therefore using different methods of immunoaffinity purification should not be used in quantitative studies, but may be included in qualitative based studies. Furthermore, based on the results of this study, we reaffirmed that the conformation of the antigen does plays a key role for immunodetection; this was seen with the Western blots using mouse anti-hGSN antibody in Figures 1 and 2 being only able to detect a single gelsolin form. Additionally, goat anti-hGSN antibody was able to pull down various forms of gelsolin in human plasma during immunoaffinity purification; however it was unable to detect all the forms following 1DE, transfer and Western blot.

Our results reported here and those reported in previous papers [30–32] have a much broader implication to which method should be used for validation and eventually which potential biomarker candidate will be used or rejected from testing on larger cohorts of clinical samples.

Although ELISA has long been considered both reliable and high throughput and it is a technique that utilizes conjugated antibodies to quantify the targeted proteins, it also has limitations in its ability to differentiate between protein forms. In comparison, the Western blot - which is able to detect expression changes in the various forms of any given protein addressing the limitation of ELISA - is not a high-throughput technique and its reliability is often questioned because of saturation of chemiluminescent signal measured with X-ray films. Standardization of Western blot is much more difficult even if a fluorescently tagged secondary antibody is used. Protein microarrays, which address the issue of high-throughput, is also based on antigen-antibody interaction and must be performed using very well characterized antibodies. If an antibody used for microarrays recognizes only one or two forms, only a fragment of information about the differential expression of any protein will be received, similar to Western blot assay. Therefore, the 2D-differential in-gel electrophoresis (DIGE) profiling method - which separates full-length forms from fragments (resulting from processing or degradation) - appears to be an attractive alternative method. In our previous profiling studies using 2D-DIGE, we were able to show that the best indicator of changes of complement C3 in CSF, which is processed by multi-step well-defined mechanism, is a "residual" α-40 chain [18]. However, lack of good antibody to this fragment of C3 made orthogonal Western blot validation impossible at that time.

A novel approach known as Stable Isotope Standards and Capture by Anti-Peptide Antibodies (SISCAPA) was developed to allow for the enrichment of targeted proteins in complex samples [33] and thereby could facilitate biomarker validation. This method is based on peptide quantitation in complex mixtures such as the total tryptic digest of plasma samples. The SISCAPA method first combines immunoaffinity purified native peptide using anti-peptide antibodies immobilized on 100-nanoliter column and spiked stable-isotope-labeled internal standard peptide of the same sequence. Next, both peptides are measured by ESI-MS/MS and quantity is calculated based on the ratio of heavy (standard) to light (native) peptide, much like multiple reaction monitoring (MRM, below). Although the SISCAPA method may lead to increased sensitivity, it is a technology utilizing antibodies and therefore the same concerns with respect to ELISA and Western blot are applicable.

The MRM approach for validation is based on the comparison of abundance of selected peptides originating from a sample and spiked standard [16, 34]; usually the peptides that are well ionized are selected for MRM quantitation. However, ambiguous results may occur depending on the peptide(s) chosen. For example, if one peptide is selected from N-terminal end of gelsolin and one from C-terminal region for MRM quantitation (or SISCAPA), the results of quantitative comparisons can be very different, as shown in Figure 5. More importantly, each peptide reflects a different situation; the N-terminal peptide will indicate quantity of full-length molecule and multimers whereas the C-terminal peptide will reflect the degree of gelsolin processing and/or degradation. Such an ambiguous result, if not further explained, may result in the rejection of a putative biomarker from further studies. Additionally, if a third peptide from the mid-region is selected and the quantity is averaged, the end result may not be different than the control sample and potential biomarker will also be rejected from further studies.

Gelsolin is a candidate biomarker for several neurocognitive diseases but before it can be integrated into the "biomarker pipeline" [29], further steps must be made to improve the immunoreactivity of anti-gelsolin antibodies. Without further antibody development, characterization and optimization, candidate biomarkers such as gelsolin will lack quantitative validation and thereby be unable to enter clinical assay development.

Validation is one of the critical steps in bringing new biomarkers from bench to bedside in translational research. Our data presented here using gelsolin as an example, highlights a set of specific problems associated with antibody based validation methods. We also briefly describe how each of the current widely accepted methods of validation has inherent weaknesses yet each are strong enough that if used alone may lead to ambiguous or even false results. Hence, conclusions based on our experimental data have a broad application as to how we should approach validation methodologically and partially explains lack of real progress in the translation of biomarkers from bench to bedside.