MicroRNA labeling methods directly influence the accuracy of microarray expression profiles

Joellyn M Enos, Jennifer L Duzeski, Paula L Roesch, James E Hagstrom & Mary-Anne V Watt Mirus Bio Corporation, Madison, Wisconsin 53719, USA.

ABSTRACT
MicroRNA (miRNA) expression profiling is commonly used to assess the relative levels of individual miRNAs in specific tissues. To achieve accurate results using microarray detection strategies, it is important that all sample miRNA species are representatively labeled prior to the hybridization step. In this study, we compared the expression profiles of miRNAs labeled using enzymatic and chemical methods. Using miRNA samples isolated from mouse brain and heart, we show that some specific miRNAs are not detected when enzymatic tailing methods are used; however they are detected with the direct chemical labeling method (Label IT^®). The presence of the discrepant miRNAs in the model mouse tissues was validated by northern blot analysis and qRT-PCR. Thus, miRNA expression profiles generated by enzymatic labeling methods (poly(A) polymerase tailing) may not represent all miRNAs present in a given sample. The Label IT technology offers a more sensitive and universal miRNA labeling method.

INTRODUCTION
MicroRNAs are phylogenetically conserved, small non-coding RNAs (~22 nucleotides) that modulate the expression of genes through post-transcriptional effects on target mRNA stability and translational efficiency [1]. Microarray expression profiling can be used to assess the relative expression of miRNAs when comparing different cell types, stages of development, and disease states [2]. To generate accurate and reproducible expression profiles using microarrays, it is critical that the labeling method effectively labels all miRNA species in the sample regardless of specific sequence or structure.

Due to the small size of miRNAs, specialized labeling methods are required. Prominent miRNA labeling methods are based on poly(A) polymerase I and involve the addition of nucleotides to the 3′ end of purified miRNAs. Fluorescent labels can be attached through enzymatic incorporation of labeled nucleotides (as in Ambion’s mirVana^™ miRNA Labeling Kit) or, by sequential enzymatic steps involving poly(A) polymerase I and T4 RNA ligase and subsequent hybridization with labeled dendrimers (Invitrogen’s NCode^™ miRNA Labeling System). In contrast, the Label IT chemical labeling method facilitates the covalent attachment of fluorescent labels directly to the miRNA molecules.

In this study, three different miRNA labeling methods were directly compared for their ability to detect differentially expressed miRNAs from two different tissue sources (mouse brain and heart). Quantitative real time (qRT) PCR and published northern blot data were used to validate the miRNA profiles.

MATERIALS and METHODS
miRNA-enriched RNA isolation
MicroRNA-enriched samples were prepared from heart and brain tissue from adult ICR mice (Harlan Laboratories, Indianapolis, IN, USA) using the mirVana miRNA Isolation Kit (Ambion, Austin, TX, USA).

miRNA labeling and hybridization
MicroRNA-enriched samples were chemically labeled via alkylation (Label IT miRNA Labeling Kit, Mirus Bio Corporation, Madison, WI, USA) or enzymatically labeled with the mirVana miRNA Labeling Kit (Ambion) or the NCode miRNA Labeling System (Invitrogen, Carlsbad, CA, USA) using the manufacturers’ recommended protocols. Labeled miRNAs were then hybridized to NCode Multi-Species miRNA microarray slides (Invitrogen). A “dye swap” setup was used with each hybridization experiment. Each brain and heart miRNA-enriched sample was labeled for detection with each of the provided fluorophores before determining relative expression (e.g. heart with Cy5, brain with Cy3, and vice versa).

Sequences of synthetic RNA oligonucleotides (IDT, Coralville, IA, USA) used for spike-in and labeling experiments were determined using the miRBase Sequence Database (http://microrna.sanger.ac.uk/sequences) [3-5]. The labeling density generated by the Label IT labeling method was estimated using spectrophotometric measurements. RNA molar concentration was determined using absorbance at 260 nm and the calculated molecular weight and extinction coefficient for each RNA oligonucleotide.

Image analysis and data processing
Microarray data was obtained using the Axon GenePix 4000B scanner (Molecular Devices, Sunnyvale, CA, USA) and associated GenePix Pro 5.0 software. Each of the data sets used was filtered using signal-to-noise criteria and the demonstration of differential expression for the given miRNA based on log2 calculations of heart: brain signal ratios. Quality data was defined to have signal greater than the average signal of the array negative controls in the relevant channel plus 1.5 standard deviations of the negative control signal. Measurements below this signal cutoff were recorded as “Not Detected”. Differentially expressed miRNAs were required to have a calculated log2(heart signal/brain signal) greater than 1 or less than –1 to represent greater than two fold differential expression between the two tissues. To fulfill the “dye swap” requirement, log transformed expression ratios were required to switch sign (e.g. from 3 to –3) when samples were detected using the opposite pairing.

RESULTS and DISCUSSION
MicroRNA-enriched brain and heart samples were labeled by chemical or enzymatic methods. The Label IT direct alkylation chemistry is based on an aromatic nitrogen mustard reactive group. This chemistry facilitates the direct covalent attachment of Cy3 or Cy5 labels onto miRNAs [6]. Both of the enzymatic methods used in this investigation are based on 3′ tailing of small RNAs using E.coli poly(A) polymerase I. The mirVana enzymatic method uses the polymerase and a mixture of unmodified and amine-modified nucleotides (uracils) to append a 20-50 polyuracil tail to the 3′ end of the miRNAs [7]. The 3′ extended miRNAs are then purified and chemically coupled with NHS ester-activated Cy3 or Cy5 labels, purified again and hybridized to the microarray. The NCode enzymatic method uses poly(A) polymerase to polyadenylate miRNAs followed by ligation of a capture sequence to the poly(A) tail [8]. After ligation, the sequence-tagged sample is purified and hybridized to the miRNA-specific microarray. The immobilized sequence-tagged miRNAs are then hybridized to labeled dendrimers containing sequence which is complementary to the capture sequence.. Notably, differences in miRNA expression profiles were observed with the different labeling methods when labeling the same miRNA-enriched samples. Ten miRNAs previously shown by northern analyses to be differentially expressed in these tissues were identified [9] and their relative abundance was determined (Figure 1). Eight of the ten miRNAs exhibited the expected tissue specific expression pattern [9]. However two of the ten miRNAs (miR-208 and miR-219) were only detected with the Label IT labeling method. Hybridization signal was not detected for either miR-208 or miR-219 in any of the replicates of either of the enzymatic labeling methods tested.

Figure 1. Discrepant microarray miRNA profiles obtained from chemical and enzymatic labeling methods. miRNA-enriched mouse heart and brain samples were hybridized to miRNA microarrays after labeling via Label IT alkylation (n=14), mirVana (n=3) or NCode (n=2) enzymatic methods. Positive relative expression values expressed as the log2 transformed ratio of heart/brain signal represent mouse miRNAs differentially expressed in heart, while negative values correspond to miRNAs differentially expressed in brain. The tissue specificity of each miRNA, as determined by northern blot analysis [9], is also presented.

To further investigate whether miR-208 and miR-219 could be detected following enzymatic labeling, we implemented a spike-in experiment using synthetic RNA oligonucleotides (Figure 2A). Two different spike-in miRNA sequences were used – control sequences representing miRNAs that were consistently detected by all three labeling methods, and test sequences representing the miRNAs that exhibited different profiles depending on the labeling method used. Heart specific miR-208 (test) and miR-1 (control) were each spiked into brain miRNA-enriched small RNA, and brain specific miR-219 (test) and miR-124a (control) were each added to the heart sample, each at a 10 fmol level. This spike-in amount is readily detectable by microarray analysis using each of the labeling methods (data not shown). Each spiked sample was labeled using the three different methods. Consistent with the original profile, spiked miR-208 and miR-219 were not detectable using either of the poly(A) polymerase-based labeling methods (Figure 2B). The enzymatic methods did detect the control spike-in miRNA sequences (miR-1 and miR-124a) in all samples, indicating the inability to detect miR-208 and miR-219 with enzymatic methods was not due to overall poor labeling of the sample. As previously observed, the Label IT chemical labeling method detected the miR-208 and miR-219 test spike-in sequences. Furthermore, qRT-PCR analysis was performed using the same mouse miRNA enriched RNA samples. The qRT-PCR results (data not shown) corroborate the published northern data which indicate that miR-1 and miR-208 are expressed in heart tissue, and miR-124a and miR-219 are expressed in brain tissue [9].

Figure 2. Specific miRNAs are not detected by enzymatic labeling methods. Synthetic RNA oligonucleotides representing control and test miRNAs were spiked into miRNA-enriched samples before chemical or enzymatic labeling and hybridization (Panel A). Control spike-in sequences represented miRNAs that were consistently detected on microarrays by all three labeling methods used. Test spike-in sequences represented the miRNAs that exhibited different profiles depending on the labeling method used. In each case, 10 fmol of each RNA oligonucleotide was spiked into the sample that did not endogenously contain the specific miRNA. Accordingly, miR-219 (test, 5′ UGAUUGUCCAAACGCAAUUCU 3′) and miR-124a (control, 5′ UAAGGCACGCGGUGAAUGCC 3′) were each spiked into heart RNA, and miR-208 (test, 5′ AUAAGACGAGCAAAAAGCUUGU 3′) and miR-1 (control, 5΄ UGGAAUGUAAAGAAGUAUGUA 3΄) were each spiked into brain RNA. Hybridization results are presented in Table B.

The Label IT reagent (Figure 3) contains an alkylating reactive group coupled with strong nucleic acid binding capability facilitated via electrostatic interaction. With this reagent, direct covalent modification of the RNA takes place during a one hour incubation at 37°C on any reactive heteroatom in the polynucleotide. Preferred heteroatoms are predicted to be N⁷ of guanine, N³ of adenine, and N³ of cytosine. To assess the efficiency and specificity of labeling, the Label IT labeling method was tested using synthetic RNA oligonucleotides representing four mammalian miRNA sequences each lacking a specific nucleotide. Consistent labeling was observed with each of the four miRNA sequences indicating the alkylation reaction was neither dependent on, nor preferential for, a specific nucleotide (Figure 4).

Figure 3. Chemical structure of the chemical labeling reagent. The Label IT labeling reagent consists of 3 parts: the label or fluorophore (for example, Cy3 or Cy5), the positively charged linker, and the reactive (alkylating) moiety. The Label IT reagent covalently attaches fluorophores to any reactive heteroatom in the RNA sample without affecting downstream hybridization performance. RNA labeling involves a one hour incubation at 37°C.

Figure 4. Chemical labeling of selected miRNA sequences. Synthetic RNA oligonucleotides representing miRNAs with no Gs (miR-467a, 5′ AUAUACAUACACACACCUACAC 3′), no As (miR-328, 5′ CUGGCCCUCUCUGCCCUUCCGU 3′), no Us (miR-214, 5′ ACAGCAGGCACAGACAGGCAG 3′), and no Cs (miR-206, 5′ UGGAAUGUAAGGAAGUGUGUGG 3′) were chemically labeled in triplicate with Cy3, purified and spectophotometrically measured (molar extinction coefficient (ε) = 150 000 M^-1 cm^-1; λ_max = 550 nm) to estimate labeling density (pmol Cy3 / µg RNA). Average labeling densities are plotted. Similar results were observed with Cy5 labeling reactions (data not shown).

Since both enzymatic methods involve nucleotide addition using E.coli poly(A) polymerase I, it is possible this step is the cause of no detection of certain miRNA species. The structure of the miRNA species, particularly under the in vitro conditions of the tailing reaction, may affect poly(A) polymerase binding affinity; it has been reported that RNA molecules terminating with a stem-loop structure are poor substrates [10]. With in vitro assays, poly(A) polymerase has been demonstrated to require a short single-stranded "toe hold" of at least 2 nucleotides [10]. Because miRNAs are known to have such stem-loop structures and may not have this “toe hold”, miRNAs may not be optimal substrates for poly(A) polymerase-based labeling systems. Conversely, the bulk of the appended tails (Figure 5) may impact the hybridization performance of particular miRNAs. As another example of the importance of miRNA structure, it has recently been reported [11] that the mirVana kit cannot be used to label plant miRNA for expression profiling applications since endogenous plant miRNAs are methylated at their 3΄ end [12] (Figure 5). Further investigation is required to discern whether it is the labeling or the hybridization step that is compromised when labeling specific miRNA species using poly(A) polymerase based methods.

Figure 5. Internal chemical labeling of any miRNA species. Mammalian miRNAs can be labeled using direct chemical labeling or enzymatic methods. The poly(A) polymerase-based methods generate long 3΄ tails which dramatically extend the length of the miRNA species. The Label IT reagents covalently modify plant and animal miRNAs at internal sites. Conversely, plant miRNAs, due to their endogenous 3΄ methylation, cannot be efficiently labeled using poly(A) polymerase-based methods.

Expression profiling is a powerful technique that allows researchers to obtain global snapshots of the expression patterns of all known miRNAs in a given tissue at a particular point in time. MicroRNA expression profiling may play an important role in disease diagnostics and therapeutics in the future. To be of value however, it is imperative that consistent and representative detection of all miRNAs is achieved. The consistent inability of enzymatic labeling methods to detect certain miRNA candidates is a serious impediment for its use, especially if those miRNAs play important roles in plant or animal biology. For example, miR-219 has been associated with ovarian and breast cancer [13]. We demonstrate here that the method of labeling miRNAs, for detection by microarray analysis, can have a significant effect on the expression profile outcome. Thus, care must be taken in choosing an appropriate miRNA labeling system.

REFERENCES
1. Kim, V.N. MicroRNA biogenesis: Coordinated cropping and dicing. Nat Rev, Mol Cell Biol 6:376 (2005).

2. Esquela-Kerscher A. and Slack, F.J. Oncomirs - microRNAs with a role in cancer. Nat Rev Cancer. 6:259 (2006).

3. Griffiths-Jones, S. et al. miRBase: microRNA sequences, targets and gene nomenclature. Nucleic Acids Res. 34:D140 (2006).

4. Griffiths-Jones, S. The microRNA Registry. Nucleic Acids Res. 32:D109 (2004).

5. Ambros V. et al. A uniform system for microRNA annotation. RNA. 9:277 (2003).

6. Slattum, P.S. et al. Efficient in vitro and in vivo expression of covalently modified plasmid DNA. Mol Therapies. 8:255 (2003).

7. Shingara, J. et al. An optimized isolation and labeling platform for accurate microRNA expression profiling. RNA. 11:1461 (2005).

8. Goff, L.A. et al. Rational probe optimization and enhanced detection strategy for microRNAs using microarrays. RNA Biology 2:93 (2005).

9. Sempere, L.F. et al. Expression profiling of mammalian microRNAs uncovers a subset of brain-expressed microRNAs with possible roles in murine and human neuronal differentiation. Genome Biology 5:R13 (2004).

10. Yehudai-Resheff, S. and Schuster, G. Characterization of the E.coli poly(A) polymerase: nucleotide specificity, RNA-binding affinities and RNA structure dependence. Nucleic Acids Res. 28:1139 (2000).

11. Ambion website product page http://www.ambion.com/catalog/ProdGrp.html?fkApp=&fkSubApp=176&fkProdGrp=347 “Note: Plant miRNAs cannot be labeled with this kit due to endogenous 3' methylation.”

12. Yu, B. et al. Methylation as a crucial step in plant microRNA biogenesis. Science 307:932 (2005).

13. Zhang, L. et al. microRNAs exhibit high frequency genomic alterations in human cancer. Proc Natl Acad Sci U S A. 103:9136 (2006).

back to top