A surge of interest in defining molecular biomarkers of health and disease from human plasma has recently emerged with the recent launch of the pilot Plasma Proteome Project (PPP) by the Human Proteome Organization (HUPO) [1]. The easy clinical access and processing of plasma samples, and the abundance of proteins as well as metabolites that may collectively define a person's health status, have made human plasma the top choice among bio-fluids for future clinical molecular diagnostic applications. The fluctuating nature of blood from different individuals, huge dynamic protein concentration ranges (up to 10¹²), and the protein detection limits of most MS platforms, have made the plasma proteome elusive to define. Many proteomics researchers even believe that the current "plasma proteome" observed by a single shotgun MS experiment is analogous to a stochastic sampling of the human proteome, with low run-to-run consistencies and inherent detection biases peculiar to each type of MS platform [2]. Even when used for healthy individual plasma, with multi-dimensional separations and advanced bioinformatics search software tools, proteins identified in different shotgun MS/MS plasma proteomics experiments are often inconsistent with each other except for the most abundant proteins. To overcome the poor coverage, potential bias, and complementary nature of each experimental measurement of the human plasma proteome, it is necessary for biomedical researchers to collect and assess all reliable publicly-available plasma protein data sets generated from different MS analytical and computational platforms for healthy individuals. A comprehensive integrated resource of the human plasma proteins for healthy individuals, currently missing in the field of clinical proteomics, would enable researchers to understand the basic constituents, diversity, and variability of the human plasma proteome. Such a resource would provide a high amount of comparative power for interpreting proteomics profile changes in patients' plasma, and may supplement or compensate for limitations and biases associated with the set of controls for a given study. It would also improve the ability for finding protein biomarkers that are known to occur in healthy human plasma for instances where a protein is differentially expressed in a patient sample related to the quantities observed in the study control.

Although multiple projects to profile the human plasma proteome have been attempted, including PeptideAtlas, GPMO, HUPO PPP, and several recent publications [3–6], an unaddressed need has been for a compiled, central repository structured to enable the stable retrieval, comparison and querying of results. The existing sources vary widely in terms of data set size, available user interface, experimental protocol or sample details, choices of protein identifiers, linking to peptide evidence from MS experiments, MS search software used, and extent of data annotation. This information needs to be compiled further and assembled for end users before they can consider incorporating human plasma proteome data into their studies. The largest single source of data is from an independently conducted experimental study that utilized a ion-mobility spectrometry (IMS) platform to chart the existence of 9,087 proteins based on 37,842 unique inferred peptide sequences, of which 2,928 proteins are high-confidence [6]. There is not however a web-based interface or other online resource for making this data widely available. The other sources of data we examined have some form of online presentation (aside from publication) and range in size from less than one thousand to over several thousand identified proteins. The Plasma Proteome Database (PPD) provides a web interface and is geared for providing detailed functional annotations of 3,778 distinct proteins based on data extracted from the literature, yet the PPD provides information on neither experimental protocol or associated MS-detected peptides used for protein identification [7]. HUPO PPP information consists of 3,020 proteins and 47,950 peptides along with experimental protocol information and is available to the public online, but the data is only accessible as flat files [8]. The Institute of Systems Biology (ISB) has surveyed and analyzed a comparably smaller set of data produced by 28 human plasma proteomics experiments, and has reported an approximate count of 960 proteins based on the 6,929 distinct observed peptides in their web-interfaced PeptideAtlas database [3]. Another resource, providing evidence for human proteomics based mainly on data from HUPO PPP, is hosted by the Global Proteome Machine Organization (GPMO). An important feature of the GPMO database is that it provides annotated information to assist with the difficult process of validating peptide MS/MS spectra and patterns of protein coverage [4]. GPMO also includes data from non-human organisms such as cats, guinea pigs, rabbits, unicellular eukaryotes like yeasts, as well as a number of prokaryotic organisms.

By gathering the protein and peptide data used to characterize the proteome of a healthy individual, we made an attempt to develop a resource that presents a comparative baseline of plasma proteomics results against which proteomic data from patients with diseases such as cancer, neurodegenerative diseases, metabolic diseases, and other genetic disorders may be studied. In this effort, we define "healthy" or "normal" as human adults without major known life-threatening diseases, genetic diseases, HIV, or inflammation at the time of blood drawing (a slightly more stringent variation than the HUPO definition in Omenn et al. [5]). On these premises, we developed an integrated database HIP² (Healthy Human Individual's Integrated Plasma Proteome) by compiling all of the existing experimental data performed on healthy individual samples, and creating a web-based interface to aid the many upcoming projects of protein biomarker studies of health and disease [9]. With HIP², clinical samplings of patient plasma may be better compared to random or non-random aspects of overlap with the reported set of healthy human plasma proteins.

The HIP² database provides protein biologists and clinical biomedical researchers with a new gateway for exploring proteins from the human proteome with peptide-level evidence found in plasma. The basic questions that the HIP² database helps biological researchers answer includes "whether a protein may be found in human plasma" and "how likely or easily it is for a protein to be observed in healthy human plasma with mass spectrometry." The HIP² database allows its users to assess the confidence of identifying plasma proteins in "normal" plasma MS proteomics experiments by examining such evidence as the number of matched peptide hits, data sources covered, MS experiments observed, types of MS platforms, and search software used. The protein and peptide sequence information also enables the user to examine peptide evidence that may be mapped to different gene splice variants and protein isoforms. Partial protein trypsin digestions can be evaluated based on multiple peptide to protein alignment information presented in the database. The overall quality of digested peptide mapped to proteins can be further used by mass spectrometry data analysts to assess different performance of MS proteomics platforms or samples. A typical example in the HIP² database of a protein-peptide sequence comparison is how the peptide sequence 'KQSAGLVLWGAILFVAWNALLLLFFWTRPAPGRPPSVSALDGDPASLTR' is present in two proteins, IPI00000138 and IPI00179044, and aligns with the same sites of trypsin cleavage (as shown in Fig. 6). In the case of protein IPI00000138, evidence for the protein was found in three data sources, three MS experiments, three MS platforms, two MS search software and six mapped peptide sequences, whereas in the case of protein 'IPI00179044', there are not any experimentally proven peptides from MS results.

The HIP² database has been mapped to cover a significant portion of the human proteome, which can be retrieved by user queries of plasma proteins or peptides, with full MS experimental context and detailed peptide-to-protein mapping relationships. As of the latest version (Feb 2008), the HIP² database aggregates information from 14 different protein separation/MS analytical platforms, 63 different samples, and 6 different types of MS search software. The HIP² database provides associated identifications of 12,787 protein entries representing 11,588 unique proteins covering 17% of all 67,511 proteins in the human IPI database [11]. Over 77% of the proteins included are identified from two or more different peptide sequences as shown in Fig. 3, where identification is defined by both peptide detection and a protein sequence match. We observed that almost 24% of the proteins included have been identified from two or more experimental platforms as shown in Fig. 4, while only 106 proteins (<1%) are identified by all four sources, implicating substantial variation in protein detections caused by a number of factors including experimental parameters and subject-to-subject variation (Fig. 5). Finding detailed instances of experimental detection of proteins by experimentally proven peptides is necessary to evaluate future options for experimental detection. For biostatisticians, the HIP² database can be a supplemental resource for assessing the likelihood of observing different peptides of the same plasma proteins from different platforms. Our resource is also useful to computational proteomics researchers, since we enable navigation among various ranges of association between a peptide sequence and a candidate protein sequence; this information may be valuable for assigning probabilistic confidence of all human proteins, whether or not they have been experimentally observed in human plasma [16]. The study of how search software is used in the experimental protocol, especially for how sequence alignment and locations of potential tryptic cleavage sites (highlighted by the HIP² interface) influence peptide detection, can now be directly addressed by the proteomics community. This online HIP² database can be a valuable source of information to biomedical researchers as they interpret proteomics profile changes in patients' plasma and pursue biomarker discovery.

The primary goal of the HIP² database is to support future clinical proteomics research, especially the discovery of biomarkers through plasma proteomics profiling. For biomedical researchers interested in MS-based plasma biomarker studies, HIP² can be the first database to search against a list of candidate proteins/genes or peptides before choices of prioritized biomarker candidates are made. As the database grows, additional annotation information of human plasma proteins such as relative abundance, normal range of variability, detectability, peptidomic patterns associative with cleavage, mutation and putative sites of glycosylation will be added. HIP² helps provide an integrated interface where database curators and data contributors can work together to collect ongoing published data from healthy human plasma proteomics experiments, foster community-based assessment of presence and absence of proteins in healthy human plasma, and provide a centralized data repository for subsequent bioinformatics analysis of the consistency and biases of each MS proteomics platform or search software. We expect this database to become an essential clinical proteomics resource, helping link together the community of biomedical researchers engaged in biomarker studies and the community of mass spectrometry researchers developing sensitive analytical solutions.

The online content of HIP² is freely available to all WWW users. The database infrastructure and software tools used to develop the database are subject to the intellectual property protection terms of Indiana University.

Project name: Database for Healthy Human Individual's Integrated Plasma Proteome

Project home page: HIP² website [17]

Browser requirements: Modern browsers (e.g., Firefox or Microsoft Explorer) will function satisfactorily.