MicroRNA Profiling in Cancer

A Bioinformatics Perspective

Yuriy Gusev, Yuriy Gusev

• Publisher: Jenny Stanford Publishing
• Year: 2019
• ISBN:
  9780429533723
• Edition: 1
• Topic: Medicine
• Subtopic: Oncology

This book presents current advances in the emerging interdisciplinary field of microRNA research of human cancers from a unique perspective of quantitative sciences: bioinformatics, computational and systems biology, and mathematical modeling. This volume contains adaptations and critical reviews of recent state-of-the-art studies, ranging from technological advances in microRNA detection and profiling, clinically oriented microRNA profiling in several human cancers, to a systems biology analysis of global patterns of microRNA regulation of signaling and metabolic pathways. Interactions with transcription factor regulatory networks and mathematical modeling of microRNA regulation are also discussed.

CHAPTER 1

MEASUREMENT AND INTERPRETATION OF MICRORNA EXPRESSION PROFILES

Bo Curry and Robert Ach

Agilent Laboratories, Agilent Technologies 5301 Stevens Creek Blvd., Santa Clara, CA 95051

The growing recognition of the importance of miRNAs in cellular regulation has given rise to considerable interest in measuring miRNA levels in biological samples. We discuss here some of the important factors to be considered to ensure accurate, sensitive, and reproducible measurements of miRNA levels, with particular attention to the Agilent miRNA microarray system. We compare and contrast microarray with qPCR measurements, showing that there is in general excellent agreement between Agilent microarray and TaqMan qPCR miRNA measurements, though with some particular exceptions. We discuss methods of assessing RNA quality, and the importance of standardizing sample prep methods to increase reproducibility and avoid some sources of systematic variation. When choosing techniques and transforms to apply to the data, it is important to understand the sensitivity and dynamic range of the measurement. We show examples of dose-response curves for array and qPCR assays. We then discuss some potential sources of measurement bias, and propose some methods of normalizing the raw data which attempt to correct for bias from different sources. Finally, we give a brief overview of methods for analyzing miRNA expression profiling data and some caveats that must be considered.

1.Background

Many studies are revealing the complexity of the roles miRNAs play in normal and abnormal animal cell development, differentiation, and regulation (for examples of recent reviews, see^{1, 2, 3, 4 and 5}). Many of the recent findings involve measurements of miRNA expression levels. It is important for such miRNA measurements to be reproducible and quantitatively comparable from sample to sample. In a typical experimental design, cells from blood, tissue, or cell culture are isolated, their RNA is extracted and quantitated, and the amount of each targeted miRNA in the RNA sample is measured. The measured amount of each miRNA is then normalized to some measure of the original sample amount, often to a proxy for the cell count of the sample⁶. Statistical techniques are then applied to the data, often in conjunction with clinical or other expression data, to answer the research questions addressed in the experiment. Each of these steps adds statistical noise and potential bias to the results, and it is important to understand these sources of error in order to choose appropriate statistics for evaluating observed correlations and variations.

In this chapter, we discuss some of the important considerations for making accurate, reproducible miRNA measurements. We compare microarray and qPCR measurements, and discuss the importance of sample prep methods and RNA quality. We then discuss normalization methods. Appropriate normalization of raw microarray data can reduce bias in comparisons among different samples. Once measurement noise has been carefully characterized and RNA quality is well controlled, most residual variation in miRNA profiles is either due to biological noise or to the biological phenomenon that is of interest in the study.

2.Methods for Measuring miRNAs

Several methods for miRNA profiling are currently in common use. Three of the most commonly used methods are microarrays (reviewed in ^{7, 8 and 9}), quantitative RT-PCR (qRT-PCR, or qPCR)^{10, 11, 12, 13 and 14}, and high-throughput sequencing.¹⁵ These techniques have different strengths and weaknesses, and thus are often used at different stages of the same study. Microarrays are convenient for profiling large numbers of previously characterized miRNAs. qPCR methods are well-suited to measuring smaller numbers of targets, especially in large numbers of samples. High-throughput sequencing is most suited to discovery, since it is the only method which can detect previously unknown miRNAs. We discuss below the characteristics of microarray measurements and of qPCR measurements, and a cross-platform comparison of microarrays and qPCR methods using Agilent miRNA microarrays and Applied Biosystems TaqMan qPCR.

2.1.Microarray measurements of miRNAs

A number of different microarray platforms have been used to profile miRNAs (for examples, see^{16, 17, 18, 19, 20, 21, 22, 23 and 24}). These platforms differ in probe design strategies and RNA labeling methods. The microarray data discussed here was acquired on the Agilent miRNA platform, following the manufacturer’s recommended protocols.²⁵ The Agilent microarray platform features the direct end-labeling and profiling of mature miRNAs directly from total RNA without the use of size fractionation or RNA amplification.^{16, 26, 27} The labeling reaction involves ligation of one labeled cytosine to the 3’ end of each RNA sequence in the sample. The labeling is performed under denaturing conditions, ensuring a high labeling yield, minimal sequence bias,¹⁶ and consistently reproducible efficiency for every miRNA sequence.^{26 27}

Agilent microarray probe design features base-pairing with the additional nucleotide incorporated at the 3’-end of the miRNA during labeling.¹⁶ Probes are designed using empirical melting point-determination, making the platform capable of single-nucleotide discrimination in the miRNA sequences.¹⁶ Hairpin structures incorporated at the 5’-end of the probes allow the binding of the mature miRNAs while discouraging the binding of longer RNAs in the total RNA sample.^16,26,27 The labeled sample is hybridized to the microarray under conditions that approach equilibrium, with a substantial and reproducible fraction of the labeled targets hybridized to the array.¹⁶

Agilent arrays comprise up to four different sequences probing each target miRNA, each replicated four or more times. They also include a large number of negative controls, which are designed to not hybridize to any labeled sequences, and are used for background subtraction and estimation of background noise. After hybridization, washing, and scanning, the image is processed as follows.²⁸ First the array features are located, and the mean signal for each feature is computed as the robust average of the counts per pixel reported for central pixels of the feature. Background fluorescence is estimated by a robust RMS fit of a surface through the negative control features, along with other features reporting weak signals not significantly higher than the average of the negative controls. The height of this fitted surface at each feature location is used to estimate the background for that feature, and the standard deviation of the residuals from the surface fit is used to estimate the noise in the background. This background is then subtracted from the mean signal of each feature, producing the background-subtracted signal (BGSubSignal). Since a large fraction of the miRNA genes are not expressed in a typical sample, the distribution of background-subtracted signals is expected to appear as a roughly Gaussian distribution centered near zero, and tailing towards the high end, where genes expressed at low levels contribute to the distribution (Figure 1).

Figure 1. Distribution of weak signals after background subtraction of a typical array measurement. The width of this distribution is a measure of the background noise on the array, which determines the detection limit.

The BGSubSignals of replicated features with the same probe sequence are then summed, after applying a statistical test to reject outliers (at p < .05). The sum of BGSubSignals for each probe sequence is multiplied by the number of pixels in each feature and by a scaling constant, and the product is reported as the TotalProbeSignal for that probe. Next, the TotalProbeSignals for all the different probes targeting a given miRNA gene are summed to produce the TotalGeneSignal for that gene, which is reported in t...