Multivariate binary classification of imbalanced datasets-A case study based on high-dimensional multiplex autoimmune assay data.

The classification of a population by a specific trait is a major task in medicine, for example when in a diagnostic setting groups of patients with specific diseases are identified, but also when in predictive medicine a group of patients is classified into specific disease severity classes that might profit from different treatments. When the sizes of those subgroups become small, for example in rare diseases, imbalances between the classes are more the rule than the exception and make statistical classification problematic when the error rate of the minority class is high. Many observations are classified as belonging to the majority class, while the error rate of the majority class is low. This case study aims to investigate class imbalance for Random Forests and Powered Partial Least Squares Discriminant Analysis (PPLS-DA) and to evaluate the performance of these classifiers when they are combined with methods to compensate imbalance (sampling methods, cost-sensitive learning approaches). We evaluate all approaches with a scoring system taking the classification results into consideration. This case study is based on one high-dimensional multiplex autoimmune assay dataset describing immune response to antigens and consisting of two classes of patients: Rheumatoid Arthritis (RA) and Systemic Lupus Erythemathodes (SLE). Datasets with varying degrees of imbalance are created by successively reducing the class of RA patients. Our results indicate possible benefit of cost-sensitive learning approaches for Random Forests. Although further research is needed to verify our findings by investigating other datasets or large-scale simulation studies, we claim that this work has the potential to increase awareness of practitioners to this problem of class imbalance and stresses the importance of considering methods to compensate class imbalance.

Serum-autoantibodies for discovery of prostate cancer specific biomarkers.

BACKGROUND: The currently used prostate cancer serum marker has a low cancer specificity and improved diagnostics are needed. Here we evaluated whether autoantibodies are present in sera of prostate cancer patients and whether they are useful diagnostic markers for prostate cancer. METHODS: Sera from 20 prostate cancer patients and 20 healthy controls were incubated on expression clone arrays containing more than 37,000 recombinant human proteins. Functional annotation clustering of the identified autoantigens was performed using the DAVID database. Autoantigens identified in the prostate cancer group were validated on microarrays using sera of 40 prostate cancer patients, 40 patients with elevated PSA levels but prostate cancer negative biopsies (benign disease), and 40 healthy controls. RESULTS: We detected autoantibodies against 408 different antigens in sera of prostate cancer patients. One hundred seventy-four of these were exclusively detected in the cancer group compared to the healthy control group. Functional annotation clustering revealed an enrichment of RNA-associated, cytoskeleton, and nuclear proteins. The autoantibody panel was validated in serum samples of independent prostate cancer patients. Autoantibody profiles discriminated between prostate cancer patients and benign disease patients with an ROC curve AUC of 0.71. TTLL12, a protein recently described to be over-expressed in prostate cancer, was the highest ranked discrimination autoantigen. CONCLUSION: A variety of autoantibodies were identified in sera of prostate cancer patients and provide a first step towards autoantibody diagnostics. Serum autoantibodies reflect the disease and represent valuable tools not only for prostate cancer, but also for other diseases affecting the immune response.

A systematic approach for the identification of novel, serologically reactive recombinant Varicella-Zoster Virus (VZV) antigens.

BACKGROUND: Varicella-Zoster virus causes chickenpox upon primary infection and shingles after reactivation. Currently available serological tests to detect VZV-specific antibodies are exclusively based on antigens derived from VZV-infected cells. RESULTS: We present a systematic approach for the identification of novel, serologically reactive VZV antigens. Therefore, all VZV open reading frames were cloned into a bacterial expression vector and checked for small scale recombinant protein expression. Serum profiling experiments using purified VZV proteins and clinically defined sera in a microarray revealed 5 putative antigens (ORFs 1, 4, 14, 49, and 68). These were rearranged in line format and validated with pre-characterized sera. CONCLUSIONS: The line assay confirmed the seroreactivity of the identified antigens and revealed its suitability for VZV serodiagnostics comparable to commercially available VZV-ELISA. Recombinant ORF68 (gE) proved to be an antigen for high-confidence determination of VZV serostatus. Furthermore, our data suggest that a serological differentiation between chickenpox and herpes zoster may be possible by analysis of the IgM-portfolio against individual viral antigens.

Profiling of IgG antibodies targeting unmodified and corresponding citrullinated autoantigens in a multicenter national cohort of early arthritis in Germany.

OBJECTIVE: To assess the diagnostic potential of IgG antibodies to citrullinated and corresponding native autoantigens in early arthritis. METHODS: IgG autoantibodies to 390 distinct unmodified and corresponding in vitro citrullinated recombinant proteins were measured by a multiplex assay in baseline blood samples from a German multicenter national cohort of 411 early arthritis patients (56.5 ± 14.6 years, 62.8% female). The cohort was randomly split into a training cohort (n = 329, 28.6% ACPA positive) and a validation cohort (n = 82, 32.9% ACPA pos.). The diagnostic properties of candidate antibodies to predict a subsequent diagnosis of rheumatoid arthritis (RA) as opposed to a non-RA diagnosis were assessed by receiver operating characteristics analysis and generalized linear modeling (GLM) with Bonferroni correction in comparison to clinically determined IgM rheumatoid factor (RF) and citrullinated peptide antibody (ACPA) status. RESULTS: Of 411 patients, 309 (75.2%) were classified as RA. Detection rates of antibody responses to citrullinated and uncitrullinated forms of the proteins were weakly correlated (Spearman's r = 0.13 (95% CI 0.029-0.22), p = 0.01). The concentration of 34 autoantibodies (32 to citrullinated and 2 to uncitrullinated antigens) was increased at least 2-fold in RA patients and further assessed. In the training cohort, a significant association of citrullinated "transformer 2 beta homolog" (cTRA2B)-IgG with RA was observed (OR 5.3 × 103, 95% CI 0.8 × 103-3.0 × 106, p = 0.047). Sensitivity and specificity of cTRA2B-IgG (51.0%/82.9%) were comparable to RF (30.8%/91.6%) or ACPA (32.1%/94.7%). Similar results were obtained in the validation cohort. The addition of cTRA2B-IgG to ACPA improved the diagnostic performance over ACPA alone (p = 0.026 by likelihood ratio test). CONCLUSIONS: cTRA2B-IgG has the potential to improve RA diagnosis in conjunction with RF and ACPA in early arthritis.

Determination and validation of off-target activities of anti-CD44 variant 6 antibodies using protein biochips and tissue microarrays.

We describe a novel procedure for determination and validation of off-target activities of anti-cluster designation antigen identity 44 (CD44) variant 6 recombinant antibodies by combining two complementary technology platforms; namely UNIchip AV-400, containing a printed serial dilution of CD44 variant 6 and approximately 400 different human proteins, and TISSOMICS, enabling human tissue microarray analysis in high-throughput mode. We have analyzed the performance of two human monoclonal recombinant antibodies directed against CD44 variant 6 protein, BMS 116 and BMS 125, in a blinded study. The antibodies exhibit a clear differentiation with regard to their binding profiles in the two systems. BMS 116 shows a low degree of specificity in the normal human Food and Drug Administration (FDA) tissue panel, which was confirmed by binding to more than 206 proteins on the protein biochip. In contrast, BMS 125 gives a highly selective membranous staining on selected human epithelial tissue components with no off-target activities observed on the protein biochip. Additionally, antibody BMS 125 shows a higher sensitivity to its antigen CD44 variant 6 than antibody BMS 116 as determined by the protein biochip.

Off-target activity of TNF-alpha inhibitors characterized by protein biochips.

Tumor necrosis factor-alpha inhibitors are widely and successfully used to treat rheumatic diseases. However, significant side effects have been reported. To detect the potential off-target activities of such inhibitors we characterized two therapeutic antibodies (adalimumab, infliximab) and one receptor fusion protein (etanercept) on protein biochips (UNIchip AV-400) containing a printed serial dilution of tumor necrosis factor-alpha and about 384 different human proteins. Etanercept binds to ten proteins (affinity: 20-33% of tumor necrosis factor-alpha recognition), and six of these proteins are related to ribosomal proteins. Interestingly, adalimumab binds to the same six proteins related to ribosomal proteins (affinity: 12-18%) as well as to four proteins crucially involved in ribosomal protein synthesis. Alignment of protein sequences indicates no significant sequence homology between these ten proteins bound by the biological drugs with the highest off-target activities. Taken together, our in vitro results demonstrate that a significant number of proteins are recognized by tumor necrosis factor-alpha inhibitors and are related to ribosome biogenesis.

Serum Autoantibodies in Chronic Prostate Inflammation in Prostate Cancer Patients.

BACKGROUND: Chronic inflammation is frequently observed on histological analysis of malignant and non-malignant prostate specimens. It is a suspected supporting factor for prostate diseases and their progression and a main cause of false positive PSA tests in cancer screening. We hypothesized that inflammation induces autoantibodies, which may be useful biomarkers. We aimed to identify and validate prostate inflammation associated serum autoantibodies in prostate cancer patients and evaluate the expression of corresponding autoantigens. METHODS: Radical prostatectomy specimens of prostate cancer patients (N = 70) were classified into high and low inflammation groups according to the amount of tissue infiltrating lymphocytes. The corresponding pre-surgery blood serum samples were scrutinized for autoantibodies using a low-density protein array. Selected autoantigens were identified in prostate tissue and their expression pattern analyzed by immunohistochemistry and qPCR. The identified autoantibody profile was cross-checked in an independent sample set (N = 63) using the Luminex-bead protein array technology. RESULTS: Protein array screening identified 165 autoantibodies differentially abundant in the serum of high compared to low inflammation patients. The expression pattern of three corresponding antigens were established in benign and cancer tissue by immunohistochemistry and qPCR: SPAST (Spastin), STX18 (Syntaxin 18) and SPOP (speckle-type POZ protein). Of these, SPAST was significantly increased in prostate tissue with high inflammation. All three autoantigens were differentially expressed in primary and/or castration resistant prostate tumors when analyzed in an inflammation-independent tissue microarray. Cross-validation of the inflammation autoantibody profile on an independent sample set using a Luminex-bead protein array, retrieved 51 of the significantly discriminating autoantibodies. Three autoantibodies were significantly upregulated in both screens, MUT, RAB11B and CSRP2 (p>0.05), two, SPOP and ZNF671, close to statistical significance (p = 0.051 and 0.076). CONCLUSIONS: We provide evidence of an inflammation-specific autoantibody profile and confirm the expression of corresponding autoantigens in prostate tissue. This supports evaluation of autoantibodies as non-invasive markers for prostate inflammation.

Sequential high-content profiling of the IgG-autoantibody repertoire reveals novel antigens in rheumatoid arthritis.

BACKGROUND: The aim was to identify novel diagnostic autoantibody candidates for rheumatoid arthritis (RA) by comprehensive screening for autoreactivity. METHOD: We incubated 5892 recombinant proteins coupled to fluorescent beads, with patients' sera for the detection of IgG-autoantibodies in three independent patient cohorts: A (n = 72 patients with established RA); B/B- (n = 116 patients with early RA (B) and n = 51 CCP-negative patients with early RA from B (B-)); and C (n = 184 patients with early seronegative RA), in comparison to matched healthy controls. Intersects of significantly increased autoantibodies as determined by the Mann-Whitney test were sought. RESULT: Screening of 5892 antigens in RA cohorts A and B, or the seronegative cohorts B- and C revealed intersects of 23 and 13 significantly increased autoantibodies, respectively. Reactivity to three antigens was increased in all cohorts tested: N-acetylglucosamine-1-phosphate transferase, gamma subunit (GNPTG), heterogeneous nuclear ribonucleoprotein A1-like 2 (HNRNPA1), and insulin-like growth factor binding protein 2 (IGFBP2). CONCLUSIONS: Comprehensive sequential screening for autoantibodies reveals novel candidates for diagnostic markers in both seropositive and seronegative RA and suggests new fields of research into the pathogenesis of RA.

Protein biochips: A new and versatile platform technology for molecular medicine.

The human genome has been sequenced and the challenges of understanding the function of the newly discovered genes have been addressed. High-throughput technologies such as DNA microarrays have been developed for the profiling of gene expression patterns in whole organisms or tissues. Protein arrays are emerging to follow DNA chips as possible screening tools. Here, we review the generation and application of microarray technology to obtain more information on the regulation of proteins, their biochemical functions and their potential interaction partners. Already, a large variety of assays based on antibody-antigen interactions exists. In addition, the medical relevance of protein arrays will be discussed.

High-throughput protein arrays: prospects for molecular diagnostics.

High-throughput protein arrays allow the miniaturized and parallel analysis of large numbers of diagnostic markers in complex samples. Using automated colony picking and gridding, cDNA or antibody libraries can be expressed and screened as clone arrays. Protein microarrays are constructed from recombinantly expressed, purified, and yet functional proteins, entailing a range of optimized expression systems. Antibody microarrays are becoming a robust format for expression profiling of whole genomes. Alternative systems, such as aptamer, PROfusion, nano- and microfluidic arrays are all at proof-of-concept stage. Differential protein profiles have been used as molecular diagnostics for cancer and autoimmune diseases and might ultimately be applied to screening of high-risk and general populations.

Aggregation of the protein TRIOBP-1 and its potential relevance to schizophrenia.

We have previously proposed that specific proteins may form insoluble aggregates as a response to an illness-specific proteostatic dysbalance in a subset of brains from individuals with mental illness, as is the case for other chronic brain conditions. So far, established risk factors DISC1 and dysbindin were seen to specifically aggregate in a subset of such patients, as was a novel schizophrenia-related protein, CRMP1, identified through a condition-specific epitope discovery approach. In this process, antibodies are raised against the pooled insoluble protein fractions (aggregomes) of post mortem brain samples from schizophrenia patients, followed by epitope identification and confirmation using additional techniques. Pursuing this epitope discovery paradigm further, we reveal TRIO binding protein (TRIOBP) to be a major substrate of a monoclonal antibody with a high specificity to brain aggregomes from patients with chronic mental illness. TRIOBP is a gene previously associated with deafness which encodes for several distinct protein species, each involved in actin cytoskeletal dynamics. The 3' splice variant TRIOBP-1 is found to be the antibody substrate and has a high aggregation propensity when over-expressed in neuroblastoma cells, while the major 5' splice variant, TRIOBP-4, does not. Endogenous TRIOBP-1 can also spontaneously aggregate, doing so to a greater extent in cell cultures which are post-mitotic, consistent with aggregated TRIOBP-1 being able to accumulate in the differentiated neurons of the brain. Finally, upon expression in Neuroscreen-1 cells, aggregated TRIOBP-1 affects cell morphology, indicating that TRIOBP-1 aggregates may directly affect cell development, as opposed to simply being a by-product of other processes involved in major mental illness. While further experiments in clinical samples are required to clarify their relevance to chronic mental illness in the general population, TRIOBP-1 aggregates are thus implicated for the first time as a biological element of the neuropathology of a subset of chronic mental illness.

Investigation of autoantibody profiles for cerebrospinal fluid biomarker discovery in patients with relapsing-remitting multiple sclerosis.

Using the UNIarray® marker technology platform, cerebrospinal fluid immunoglobulin G reactivities of 15 controls and 17 RRMS patients against human recombinant proteins were investigated. Patient cerebrospinal fluids were oligoclonal band positive and reactivities were compared to that of sex- and age-matched controls. We hereby aimed at the characterization of autoreactivity in patients with RRMS. Differences in autoreactivities between control and RRMS samples were identified comprising autoantigens identified in this study only and previously reported autoantigens as well. A combination of the 10-15 most significant proteins may be investigated further as autoantigens for diagnostic purposes. Additional investigations may include minimizing the number of proteins used in such diagnostic tests.

New analytical tools for studying autoimmune diseases.

Protein microarrays with immobilised proteins on their surface are new analytical tools to overcome the current limits with respect to sample volume and throughput. They have a great potential as well with respect to multiplexing of complex samples, as a research tool and in diagnostics. Based on recent advances in this technology, new applications for protein microarrays in studying autoimmune diseases were described. Required tools for bioinformatical analysis of protein microarrays concerning normalisation, clustering and classification methods are discussed. The huge potential of this technology as well as future requirements such as protein microarray based diagnostics are presented.

Profiling humoral autoimmune repertoire of dilated cardiomyopathy (DCM) patients and development of a disease-associated protein chip.

Dilated cardiomyopathy (DCM) is a myocardial disease characterized by progressive depression of myocardial contractile function and ventricular dilatation. Thirty percent of DCM patients belong to the inherited genetic form; the rest may be idiopathic, viral, autoimmune, or immune-mediated associated with a viral infection. Disturbances in humoral and cellular immunity have been described in cases of myocarditis and DCM. A number of autoantibodies against cardiac cell proteins have been identified in DCM. In this study, we have profiled the autoantibody repertoire of plasma from DCM patients against a human protein array consisting of 37,200 redundant, recombinant human proteins and performed qualitative and quantitative validation of these putative autoantigens on protein microarrays to identify novel putative DCM specific autoantigens. In addition to analyzing the whole IgG autoantibody repertoire, we have also analyzed the IgG3 antibody repertoire in the plasma samples to study the characteristics of IgG3 subclass antibodies. By combining screening of a protein expression library with protein microarray technology, we have detected 26 proteins identified by the IgG antibody repertoire and 6 proteins bound by the IgG3 subclass. Several of these autoantibodies found in plasma of DCM patients, such as the autoantibody against the Kv channel-interacting protein, are associated with heart failure.

Mouse protein arrays from a TH1 cell cDNA library for antibody screening and serum profiling.

The mouse is the premier genetic model organism for the study of disease and development. We describe the establishment of a mouse T helper cell type 1 (T(H)1) protein expression library that provides direct access to thousands of recombinant mouse proteins, in particular those associated with immune responses. The advantage of a system based on the combination of large cDNA expression libraries with microarray technology is the direct connection of the DNA sequence information from a particular clone to its recombinant, expressed protein. We have generated a mouse T(H)1 expression cDNA library and used protein arrays of this library to characterize the specificity and cross-reactivity of antibodies. Additionally, we have profiled the autoantibody repertoire in serum of a mouse model for systemic lupus erythematosus on these protein arrays and validated the putative autoantigens on highly sensitive protein microarrays.

Profiling of alopecia areata autoantigens based on protein microarray technology.

Protein biochips have a great potential in future parallel processing of complex samples as a research tool and in diagnostics. For the generation of protein biochips, highly automated technologies have been developed for cDNA expression library production, high throughput protein expression, large scale analysis of proteins, and protein microarray generation. Using this technology, we present here a strategy to identify potential autoantigens involved in the pathogenesis of alopecia areata, an often chronic disease leading to the rapid loss of scalp hair. Only little is known about the putative autoantigen(s) involved in this process. By combining protein microarray technology with the use of large cDNA expression libraries, we profiled the autoantibody repertoire of sera from alopecia areata patients against a human protein array consisting of 37,200 redundant, recombinant human proteins. The data sets obtained from incubations with patient sera were compared with control sera from clinically healthy persons and to background incubations with anti-human IgG antibodies. From these results, a smaller protein subset was generated and subjected to qualitative and quantitative validation on highly sensitive protein microarrays to identify novel alopecia areata-associated autoantigens. Eight autoantigens were identified by protein chip technology and were successfully confirmed by Western blot analysis. These autoantigens were arrayed on protein microarrays to generate a disease-associated protein chip. To confirm the specificity of the results obtained, sera from patients with psoriasis or hand and foot eczema as well as skin allergy were additionally examined on the disease-associated protein chip. By using alopecia areata as a model for an autoimmune disease, our investigations show that the protein microarray technology has potential for the identification and evaluation of autoantigens as well as in diagnosis such as to differentiate alopecia areata from other skin diseases.

Cell-free protein expression and functional assay in nanowell chip format.

The expression and characterization of large protein libraries requires high-throughput tools for rapid and cost-effective expression and screening. A promising tool to meet these requirements is miniaturized high-density plates in chip format, consisting of an array of wells with submicroliter volumes. Here, we show the combination of nanowell chip technology and cell-free transcription and translation of proteins. Using piezoelectric dispensers, we transferred proteins into nanowells down to volumes of 100 nL and successfully detected fluorescence using confocal laser scanning. Moreover, we showed cell-free expression of proteins on a nanoliter scale using commercially available coupled transcription and translation systems. To reduce costs, we demonstrated the feasibility of diluting the coupled in vitro transcription and translation mix prior to expression. Additionally, we present an enzymatic inhibition assay in nanowells to anticipate further applications, such as the high-throughput screening of drug candidates or the identification of novel enzymes for biotechnology.

Protein array technology: the tool to bridge genomics and proteomics.

The generation of protein chips requires much more efforts than DNA microchips. While DNA is DNA and a variety of different DNA molecules behave stable in a hybridisation experiment, proteins are much more difficult to produce and to handle. Outside of a narrow range of environmental conditions, proteins will denature, lose their three-dimensional structure and a lot of their specificity and activity. The chapter describes the pitfalls and challenges in Protein Microarray technology to produce native and functional proteins and store them in a native and special environment for every single spot on an array, making applications like antibody profiling and serum screening possible not only on denatured arrays but also on native protein arrays.

A nonredundant human protein chip for antibody screening and serum profiling.

There is burgeoning interest in protein microarrays, but a source of thousands of nonredundant, purified proteins was not previously available. Here we show a glass chip containing 2413 nonredundant purified human fusion proteins on a polymer surface, where densities up to 1600 proteins/cm(2) on a microscope slide can be realized. In addition, the polymer coating of the glass slide enables screening of protein interactions under nondenaturing conditions. Such screenings require only 200-microl sample volumes, illustrating their potential for high-throughput applications. Here we demonstrate two applications: the characterization of antibody binding, specificity, and cross-reactivity; and profiling the antibody repertoire in body fluids, such as serum from patients with autoimmune diseases. For the first application, we have incubated these protein chips with anti-RGSHis(6), anti-GAPDH, and anti-HSP90beta antibodies. In an initial proof of principle study for the second application, we have screened serum from alopecia and arthritis patients. With analysis of large sample numbers, identification of disease-associated proteins to generate novel diagnostic markers may be possible.