Real-time polymerase chain reaction-based exponential sample amplification for microarray gene expression profiling

Diagnostic biases in translational bioinformatics

BACKGROUND

With the surge of translational medicine and computational omics research, complex disease diagnosis is more and more relying on massive omics data-driven molecular signature detection. However, how to detect and prevent possible diagnostic biases in translational bioinformatics remains an unsolved problem despite its importance in the coming era of personalized medicine.

METHODS

In this study, we comprehensively investigate the diagnostic bias problem by analyzing benchmark gene array, protein array, RNA-Seq and miRNA-Seq data under the framework of support vector machines for different model selection methods. We further categorize the diagnostic biases into different types by conducting rigorous kernel matrix analysis and provide effective machine learning methods to conquer the diagnostic biases.

RESULTS

In this study, we comprehensively investigate the diagnostic bias problem by analyzing benchmark gene array, protein array, RNA-Seq and miRNA-Seq data under the framework of support vector machines. We have found that the diagnostic biases happen for data with different distributions and SVM with different kernels. Moreover, we identify total three types of diagnostic biases: overfitting bias, label skewness bias, and underfitting bias in SVM diagnostics, and present corresponding reasons through rigorous analysis. Compared with the overfitting and underfitting biases, the label skewness bias is more challenging to detect and conquer because it can be easily confused as a normal diagnostic case from its deceptive accuracy. To tackle this problem, we propose a derivative component analysis based support vector machines to conquer the label skewness bias by achieving the rivaling clinical diagnostic results.

CONCLUSIONS

Our studies demonstrate that the diagnostic biases are mainly caused by the three major factors, i.e. kernel selection, signal amplification mechanism in high-throughput profiling, and training data label distribution. Moreover, the proposed DCA-SVM diagnosis provides a generic solution for the label skewness bias overcome due to the powerful feature extraction capability from derivative component analysis. Our work identifies and solves an important but less addressed problem in translational research. It also has a positive impact on machine learning for adding new results to kernel-based learning for omics data.

Overcome support vector machine diagnosis overfitting

Support vector machines (SVMs) are widely employed in molecular diagnosis of disease for their efficiency and robustness. However, there is no previous research to analyze their overfitting in high-dimensional omics data based disease diagnosis, which is essential to avoid deceptive diagnostic results and enhance clinical decision making. In this work, we comprehensively investigate this problem from both theoretical and practical standpoints to unveil the special characteristics of SVM overfitting. We found that disease diagnosis under an SVM classifier would inevitably encounter overfitting under a Gaussian kernel because of the large data variations generated from high-throughput profiling technologies. Furthermore, we propose a novel sparse-coding kernel approach to overcome SVM overfitting in disease diagnosis. Unlike traditional ad-hoc parametric tuning approaches, it not only robustly conquers the overfitting problem, but also achieves good diagnostic accuracy. To our knowledge, it is the first rigorous method proposed to overcome SVM overfitting. Finally, we propose a novel biomarker discovery algorithm: Gene-Switch-Marker (GSM) to capture meaningful biomarkers by taking advantage of SVM overfitting on single genes.

Stochastic profiling of transcriptional regulatory heterogeneities in tissues, tumors and cultured cells

Single-cell variations in gene and protein expression are important during development and disease. Such cell-to-cell heterogeneities can be directly inspected one cell at a time, but global methods are usually not sensitive enough to work with the starting material of a single cell. Here we provide a detailed protocol for stochastic profiling, a method that infers single-cell regulatory heterogeneities by repeatedly sampling small collections of cells selected at random. Repeated stochastic sampling is performed by laser-capture microdissection or limiting dilution, followed by careful exponential cDNA amplification, hybridization to microarrays and statistical analysis. Stochastic profiling surveys the transcriptome for programs that are heterogeneously regulated among cellular subpopulations in their native tissue context. The protocol is readily optimized for specific biological applications and takes about 1 week to complete.

Influence of RNA labeling on expression profiling of microRNAs

Although a number of technical parameters are now being examined to optimize microRNA profiling experiments, it is unknown whether reagent or component changes to the labeling step affect starting RNA requirements or microarray performance. Human brain/lung samples were each labeled in duplicate, at 1.0, 0.5, 0.2, and 0.1 μg of total RNA, by means of two kits that use the same labeling procedure but differ in the reagent composition used to label microRNAs. Statistical measures of reliability and validity were used to evaluate microarray data. Cross-platform confirmation was accomplished using TaqMan microRNA assays. Synthetic microRNA spike-in experiments were also performed to establish the microarray signal dynamic range using the ligation-modified kit. Technical replicate correlations of signal intensity values were high using both kits, but improved with the ligation-modified assay. The drop in detection call sensitivity and miRNA gene list correlations, when using reduced amounts of standard-labeled RNA, was considerably improved with the ligation-modified kit. Microarray signal dynamic range was found to be linear across three orders of magnitude from 4.88 to 5000 attomoles. Thus, optimization of the microRNA labeling reagent can result in at least a 10-fold decrease in microarray total RNA requirements with little compromise to data quality. Clinical investigations bottlenecked by the amount of starting material may use a ligation mix modification strategy to reduce total RNA requirements.

Identifying single-cell molecular programs by stochastic profiling

Cells within tissues can be morphologically indistinguishable yet show molecular expression patterns that are remarkably heterogeneous. Here, we describe an approach for comprehensively identifying coregulated, heterogeneously expressed genes among cells that otherwise appear identical. The technique, called “stochastic profiling”, involves the repeated, random selection of very-small cell populations via laser-capture microdissection, followed by a customized single-cell amplification procedure and transcriptional profiling. Fluctuations in the resulting gene-expression measurements are then analyzed statistically to identify transcripts that are heterogeneously co-expressed. We stochastically profiled matrix-attached human epithelial cells in a three-dimensional culture model of mammary-acinar morphogenesis. Of 4,557 transcripts, we identified 547 genes with strong cell-to-cell expression differences. Clustering of this heterogeneous subset revealed several molecular “programs” implicated in protein biosynthesis, oxidative-stress responses, and nuclear factor-κB signaling, which were independently confirmed by RNA fluorescence in situ hybridization. Thus, stochastic profiling can reveal single-cell heterogeneities without measuring individual cells explicitly.

Human airway epithelial cell culture to identify new respiratory viruses: coronavirus NL63 as a model

Propagation of new human respiratory virus pathogens in established cell lines is hampered by a lack of predictability regarding cell line permissivity and by availability of suitable antibody reagents to detect infection in cell lines that do not exhibit significant cytopathic effect. Recently, molecular methods have been used to amplify and identify novel nucleic acid sequences directly from clinical samples, but these methods may be hampered by the quantity of virus present in respiratory secretions at different time points following the onset of infection. Human airway epithelial (HAE) cultures, which effectively mimic the human bronchial environment, allow for cultivation of a wide variety of human respiratory viral pathogens. The goal of the experiments described here was to determine if propagation and identification of a human respiratory virus may be achieved through inoculation of HAE cultures followed by whole transcriptome amplification (WTA) and sequence analysis. To establish proof-of-principle human coronavirus NL63 (HCoV-NL63) was evaluated, and the first visualization of HCoV-NL63 virus by transmission electron microscopy (TEM) is reported. Initial propagation of human respiratory secretions onto HAE cultures followed by TEM and WTA of culture supernatant may be a useful approach for visualization and detection of new human respiratory pathogens that have eluded identification by traditional approaches.

Gene expression in lung and basal forebrain during influenza infection in mice

Inbred mice develop strain-dependent changes in sleep during the first few days after inoculation with influenza virus. To identify genes with the potential to differentially modulate sleep under this condition, we performed complementary DNA microarray analysis of both lung and basal forebrain (BF) of infected (I) and uninfected BALB/cByJ (C) and C57BL/6J (B6) mice. This analysis showed significant variation in the expression pattern of 667 and 1217 of the surveyed genes in BF and lung, respectively (P < 0.01). Applying the additional criterion of an effect size >or=2, 495 genes differed in expression in lung compared with 204 in BF. In BF, more genes were differentially expressed as a function of mouse strain, whereas in lung, more genes were differentially expressed as a function of health status. Significant alterations in expression after infection were more numerous and robust in BALB/cByJ vs. C57BL/6J mice. Some genes showed significant variation in both tissues as a function of strain or condition, but the changes in general were not parallel. Genes that showed significant and robust variation as a function of strain, health status or tissue included those related to immune function, metabolism, signal transduction, cell cycle regulation, apoptosis and other miscellaneous categories. Different patterns of gene expression in BF of uninfected mice suggest the possibility of fundamental mechanistic differences in pathways that modulate vigilance in these strains, whereas differences in expression of lung of infected mice suggests different peripherally generated sleep-modulatory stimuli in the two strains.

Amplification biases: possible differences among deviating gene expressions

Background

Gene expression profiling has become a tool of choice to study pathological or developmental questions but in most cases the material is scarce and requires sample amplification. Two main procedures have been used: in vitro transcription (IVT) and polymerase chain reaction (PCR), the former known as linear and the latter as exponential. Previous reports identified enzymatic pitfalls in PCR and IVT protocols; however the possible differences between the sequences affected by these amplification defaults were only rarely explored.

Results

Screening a bovine cDNA array dedicated to embryonic stages with embryonic (n = 3) and somatic tissues (n = 2), we proceeded to moderate amplifications starting from 1 μg of total RNA (global PCR or IVT one round). Whatever the tissue, 16% of the probes were involved in deviating gene expressions due to amplification defaults. These distortions were likely due to the molecular features of the affected sequences (position within a gene, GC content, hairpin number) but also to the relative abundance of these transcripts within the tissues. These deviating genes mainly encoded housekeeping genes from physiological or cellular processes (70%) and constituted 2 subsets which did not overlap (molecular features, signal intensities, gene ID). However, the differential expressions identified between embryonic stages were both reliable (minor intersect with biased expressions) and relevant (biologically validated). In addition, the relative expression levels of those genes were biologically similar between amplified and unamplified samples.

Conclusion

Conversely to the most recent reports which challenged the use of intense amplification procedures on minute amounts of RNA, we chose moderate PCR and IVT amplifications for our gene profiling study. Conclusively, it appeared that systematic biases arose even with moderate amplification procedures, independently of (i) the sample used: brain, ovary or embryos, (ii) the enzymatic properties initially inferred (exponential or linear) and (iii) the preliminary optimization of the protocols. Moreover the use of an in-house developed array, small-sized but well suited to the tissues we worked with, was of real interest for the search of differential expressions.

Therapeutic proteasome inhibition in experimental acute pancreatitis

AIM: To establish the therapeutic potential of proteasome inhibition, we examined the therapeutic effects of MG132 (Z-Leu-Leu-Leu-aldehyde) in an experimental model of acute pancreatitis.

METHODS: Pancreatitis was induced in rats by two hourly intraperitoneal (ip) injections of cholecystokinin octapeptide (CCK; 2 × 100 μg/kg) and the proteasome inhibitor MG132 (10 mg/kg ip) was administered 30 min after the second CCK injection. Animals were sacrificed 4 h after the first injection of CCK.

RESULTS: Administering the proteasome inhibitor MG132 (at a dose of 10 mg/kg, ip) 90 min after the onset of pancreatic inflammation induced the expression of cell-protective 72 kDa heat shock protein (HSP72) and decreased DNA-binding of nuclear factor-κB (NF-κB).

Furthermore MG132 treatment resulted in milder inflammatory response and cellular damage, as revealed by improved laboratory and histological parameters of pancreatitis and associated oxidative stress.

CONCLUSION: Our findings suggest that proteasome inhibition might be beneficial not only for the prevention, but also for the therapy of acute pancreatitis.

Evaluation of small sample cDNA amplification for microdissected airway expression profiling in COPD

Small airway obstruction and emphysematous destruction account for the airflow limitation that defines chronic obstructive pulmonary disease (COPD). While laser capture microdissection (LCM) allows gene expression studies in small airways separately from the surrounding parenchyma, tissue size limits the number of genes examined. The present study evaluates the Clontech SMART amplification to test the hypothesis that this amplification provides RNA in sufficient quantity and quality to evaluate large numbers of genes in airways < 2 mm diameter obtained by LCM. Commercial reference RNA was amplified 200-fold and the expression levels of 51 genes relative to the unamplified RNA had a correlation coefficient of 0.84. For two pairs of RNA preparations (commercial placenta versus commercial lung; lung sections prepared for LCM from GOLD 0 (at risk for COPD) versus GOLD 2 (moderate disease) patients linear regression of Delta Ct's (delta cycle thresholds) of unamplified versus amplified RNA gave correlation coefficients of R = 0.95. In RNA from microdissected small airways, expression patterns in all GOLD classes of COPD severity were very similar between unamplified and amplified RNA. We conclude that SMART amplification provides cDNA sufficient for studying large numbers of genes even in laser-captured small airways and this cDNA maintains the relative expression found in corresponding unamplified RNAs.

Single molecule transcription profiling with AFM

Established techniques for global gene expression profiling, such as microarrays, face fundamental sensitivity constraints. Due to greatly increasing interest in examining minute samples from micro-dissected tissues, including single cells, unorthodox approaches, including molecular nanotechnologies, are being explored in this application. Here, we examine the use of single molecule, ordered restriction mapping, combined with AFM, to measure gene transcription levels from very low abundance samples. We frame the problem mathematically, using coding theory, and present an analysis of the critical error sources that may serve as a guide to designing future studies. We follow with experiments detailing the construction of high density, single molecule, ordered restriction maps from plasmids and from cDNA molecules, using two different enzymes, a result not previously reported. We discuss these results in the context of our calculations.

Two-dimensional strandness-dependent electrophoresis

Two-dimensional strandness-dependent electrophoresis (2D-SDE) separates nucleic acids in complex samples according to strandness, conformation and length. Under the non-denaturing conditions of the first electrophoretic step, single-stranded DNA, double-stranded DNA and RNA.DNA hybrids of similar length migrate at different rates. The second electrophoretic step is performed under denaturing conditions (7 mol l(-1) urea, 55 degrees C) so that all the molecules are single-stranded and separate according to length only. 2D-SDE is useful for revealing important characteristics of complex nucleic acid samples in manipulations such as amplification, renaturation, cDNA synthesis and microarray hybridization. It can also be used to identify mispaired, nicked or damaged fragments in double-stranded DNA. The protocol takes approximately 2 h and requires only basic skills, equipment and reagents.

Evaluation of two methods for generating cRNA for microarray experiments from nanogram amounts of total RNA

Several methods have been developed for amplification of RNA, making it possible to use cDNA microarrays for analysis of samples limited in amount of total RNA. The most widely used amplification protocol, the Eberwine method, amplifies RNA in a linear manner through in vitro transcription (IVT). However, when starting material is limited to nanogram amounts of total RNA, several rounds of amplification are necessary, making this method both expensive and labor-intensive. Amplification by PCR is robust and is able to amplify extremely limiting material. However, it is possible that the nonlinear nature of PCR could result in reduced reproducibility of the amplification compared with IVT. We have evaluated two methods that use a combination of PCR and IVT for amplification of nanogram amounts of total RNA. We have compared microarray results obtained by these methods with results obtained by two established methods: indirect labeling of 20 microg total RNA and Eberwine amplification of 1 microg total RNA. Starting from as little as 5 ng of total RNA, both methods yielded results in concordance with the Eberwine method.

Improved DOP-PCR-based representational whole-genome amplification using quantitative real-time PCR

In many cases, only a minute amount of partially degraded genomic DNA can be extracted from archived clinical samples. Diverse whole-genome amplification methods are applied to provide sufficient amount of DNA for comparative genome hybridization, single-nucleotide polymorphism, and microsatellite analyses. In these applications, the reliability of the amplification techniques is particularly important. In PCR-based approaches, the plateau effect can seriously alter the original relative copy number of certain chromosomal regions. To eliminate this distorting effect, we improved the standard degenerate oligonucleotide-primed PCR (DOP-PCR) technique by following the amplification status with quantitative real-time PCR (QRT-PCR). With real-time detection of the products, we could eliminate DNA overamplification. Probes were prepared from 10 different tumor samples: primary and metastatic melanoma tissues, epidermoid and bronchioloalveolar lung carcinomas, 2 renal cell carcinomas, 2 colorectal carcinomas, and a Conn and Cushing adenoma. Probes were generated by using nonamplified and amplified genomic DNA with DOP-PCR and DOP-PCR combined with QRT-PCR. To demonstrate the reliability of the QRT-PCR based amplification protocol, altogether 152 relative copy number changes of 44 regions were determined. There was 85.6% concordance in copy number alterations between the QRT-PCR protocol and the nonamplified samples, whereas this value was only 63.8% for the traditional DOP-PCR. Our results demonstrate that our protocol preserves the original copy number of different chromosomal regions in amplified genomic DNA than standard DOP-PCR techniques more accurately.

Options available for profiling small samples: a review of sample amplification technology when combined with microarray profiling

The possibility of performing microarray analysis on limited material has been demonstrated in a number of publications. In this review we approach the technical aspects of mRNA amplification and several important implicit consequences, for both linear and exponential procedures. Amplification efficiencies clearly allow profiling of extremely small samples. The conservation of transcript abundance is the most important issue regarding the use of sample amplification in combination with microarray analysis, and this aspect has generally been found to be acceptable, although demonstrated to decrease in highly diluted samples. The fact that variability and discrepancies in microarray profiles increase with minute sample sizes has been clearly documented, but for many studies this does appear to have affected the biological conclusions. We suggest that this is due to the data analysis approach applied, and the consequence is the chance of presenting misleading results. We discuss the issue of amplification sensitivity limits in the light of reports on fidelity, published data from reviewed articles and data analysis approaches. These are important considerations to be reflected in the design of future studies and when evaluating biological conclusions from published microarray studies based on extremely low input RNA quantities.

The proteasome inhibitor MG132 protects against acute pancreatitis

The cell-permeant MG132 tripeptide (Z-Leu-Leu-Leu-aldehyde) is a peptide aldehyde proteasome inhibitor that also inhibits other proteases, including calpains and cathepsins. By blocking the proteasome, this tripeptide has been shown to induce the expression of cell-protective heat shock proteins (HSPs) in vitro. Effects of MG132 were studied in an in vivo model of acute pancreatitis. Pancreatitis was induced in male Wistar rats by injecting 2 x 100 microug/kg cholecystokinin octapeptide intraperitoneally (ip) at an interval of 1 h. Pretreating the animals with 10 mg/kg MG132 ip before the induction of pancreatitis significantly inhibited IkappaB degradation and subsequent activation of nuclear factor-kappaB (NF-kappaB). MG132 also increased HSP72 expression. Induction of HSP72 and inhibition of NF-kappaB improved parameters of acute pancreatitis. Thus MG132 significantly decreased serum amylase, pancreatic weight/body weight ratio, pancreatic myeloperoxidase activity, proinflammatory cytokine concentrations, and the expression of pancreatitis-associated protein. Parameters of oxidative stress (GSH, MDA, SOD, etc.) were improved in both the serum and the pancreas. Histopathological examinations revealed that pancreatic specimens of animals pretreated with the peptide demonstrated milder edema, cellular damage, and inflammatory activity. Our findings show that simultaneous inhibition of calpains, cathepsins, and the proteasome with MG132 prevents the onset of acute pancreatitis.