A Quasi-direct LC-MS/MS-based Targeted Proteomics Approach for miRNA Quantification via a Covalently Immobilized DNA-peptide Probe

MicroRNAs (miRNAs) play a vital role in regulating gene expression and are associated with a variety of cancers, including breast cancer. Their distorted and unique expression is a potential marker in clinical diagnoses and prognoses. Thus, accurate determination of miRNA expression levels is a prerequisite for their applications. However, the assays currently available for miRNA detection typically require pre-enrichment, amplification and labeling steps, and most of the assays are only semi-quantitative. Therefore, we developed a quasi-direct liquid chromatography-tandem mass spectrometry (LC-MS/MS)-based targeted proteomics approach to quantify target miRNA by innovatively converting the miRNA signal into the mass response of a reporter peptide via a covalently immobilized DNA-peptide probe. Specifically, the probe containing the targeted proteomics-selected substrate/reporter peptide, GDRAVQLGVDPFR/AVQLGVDPFR, and the DNA sequence complementary to the target miRNA (i.e., miR-21) was first immobilized on APMTS modified silica nanoparticles using PDITC. After the immobilized probe was recognized and hybridized with the target miRNA, the excess probe was degraded using MBN and followed by a trypsin digestion of the hybrids. The reporter peptide was released and quantified using LC-MS/MS. The obtained LOQ was 5 pM. Finally, the developed assay was used for the quantitative analysis of miR-21 in breast cells and tissue samples.

MicroRNAs (miRNAs) are single-stranded, noncoding RNAs with a typical length of 18-25 nucleotides. They generally take part in a number of biological processes as regulatory factors via repression of messenger RNA (mRNA) translation [1][2][3] . Recently, accumulating evidence has demonstrated that miRNA plays a very important role in the development of various cancers, including breast cancer [4][5][6][7] . For example, the differential levels of miRNA in healthy volunteers and breast cancer patients indicated the potential use of miRNA as a diagnostic/ prognostic marker and therapeutic target 5,8,9 . Therefore, accurate determination of miRNA expression levels is a prerequisite for their application.
The relatively short length and low abundance of miRNAs are disadvantages that must be overcome in a detection platform 10,11 . To date, many detection techniques have been reported for the study of the miRNA expression profile, including indirect methods (e.g., microarrays 12 , polymerase chain reaction (PCR) 13,14 , next-generation sequencing 15 , electrocatalysis 16 and fluorescent resonance energy transfer (FRET)) 17 and direct methods (e.g., differential interference contrast (DIC) imaging 18 , spectral detection assisted by duplex-specific nuclease 10 , electrochemical-based methods and capillary electrophoresis (CE)-based methods) 19,20 . Indirect detection methods normally require pre-amplification and chemical or enzymatic modification of the target miRNA. These additional steps often result in information loss, increased assay time and sequence-related biases during quantification 20 . Notably, most of these methods made use of fluorescence detection. In a previous study, the beads array for the detection of gene expression (BADGE) assay immobilized oligonucleotide capture probes on fluorescence microspheres, and the beads were then hybridized to labeled cRNA and passed through a counting device, which records the identity of each microsphere and its probe intensity 21 . The subsequent studies mostly combined from baker's yeast (Sigma, St. Louis, MO, USA) 45 . The concentrations of the calibration standards were 5 pM, 10 pM, 25 pM, 100 pM, 1 nM, and 10 nM. The QC standards for the lower limit of quantification (LLOQ), low QC, mid QC and high QC were prepared at 5 pM, 15 pM, 500 pM and 8 nM and frozen prior to use. For the reporter peptide, the corresponding isotope-labeled synthetic peptide was used as the internal standard. A 100 μM internal standard stock solution was prepared, and a 1 nM working solution was prepared by diluting the stock solution with an ACN:water mixture (50:50, v/v) containing 0.1% FA.
Cell Culture and Tissue Collection. MCF-10A cells (ATCC, Manassas, VA) and MCF-7/WT (ATCC, Manassas, VA) cells were cultured routinely in DMEM media supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin in a humidified atmosphere with 5% CO 2 at 37 °C. MCF-7/ADR (Keygen Biotech, Nanjing, China) cells were grown in an RPMI 1640 media (with L-glutamine and sodium pyruvate) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin at 37 °C under a 5% CO 2 atmosphere. Cells were split every 5-7 days by lifting the cells with 0.25% trypsin, and the cells were maintained by the addition of fresh medium. The MCF-7/ADR cells were periodically reselected by growing them with 1000 ng/mL DOX to maintain a highly drug-resistant cell population 46 . Experiments were conducted using the cells incubated without DOX for 48 h. The cells were counted using a hemocytometer (Qiujing, Shanghai, China). The cell viability was assessed using trypan blue (0.4%) exclusion. The cell suspensions, PBS and trypan blue were mixed in a 2:3:5 ratio, and the viable cells were counted after incubation for 5 min at 37 °C.
The breast tissue collection in the present study was approved by the ethical review committee of Nanjing Medical University, and informed consent was obtained from all patients. All experiments were performed in accordance with the relevant guidelines and regulations. Thirty-six pairs of tumors and adjacent normal tissue samples were collected from breast cancer patients between January 2012 and December 2012 at the First Affiliated Hospital of Nanjing Medical University and Nanjing Drum Tower Hospital, Nanjing, China (mean patient age, 52.9 ± 8.6 years; age range, 38-65 years). The pathologic examinations of the tissue sections were confirmed as normal and/or cancerous by the hospital pathologists. The histological evaluations of the adjacent tissue samples indicated that there was no contamination from the tumor or other abnormal cells. All the participants were biologically unrelated and belonged to the Han Chinese ethnic group from the Jiangsu province in China. Substrate peptide containing the sequence of reporter peptide and tryptic cleavage site was tagged to DNA at 3′ end, which was complementary to the target miRNA. With PDITC reaction, the DNA-peptide probe was immobilized on the amino-modified silica nanoparticles. This immobilized probe was then hybridized with the target miRNA, followed by sequential digestion by mung bean nuclease and trypsin. The reporter peptide was ultimately released and quantified using LC-MS/MS. In this way, the quantity of target miRNA can be inferred from the measurement of reporter peptide.
Scientific RepoRts | 7: 5669 | DOI:10.1038/s41598-017-05495-7 After collection, the samples were stripped of adhering fat, cut into small pieces, and stored at −80 °C. Before RNA extraction, they were thawed to room temperature and rinsed thoroughly with DEPC water. The total RNA was isolated from the cells, and using TRIzol Reagent, the concentrations of approximately 50 mg tissue homogenates were estimated using a nanodrop spectrophotometer (Thermo Fisher Scientific Inc., MA, USA). The RNA extracts were stored at −80 °C until further processing.
Preparation of the DNA-peptide Probe. The experimental procedure to prepare the DNA-peptide probe has been described in our previous work 38 . Briefly, 1 OD DNA (4.55 nmol) was first reduced using 300 μL of TCEP reducing beads at 37 °C for 2 h with vigorous shaking. After centrifugation at 1000 g for 5 min, the supernatant was collected, and an equal volume of 50 nmol maleimidohexanoic acid-modified substrate peptide was added to the solution. Prior to purification, the conjugation reaction was performed at 37 °C for 4 h with vigorous shaking. The DNA-peptide compound was separated from the excess non-conjugated DNA and peptide using high-performance liquid chromatography (HPLC). Finally, the collected fraction was ultra-filtered using an Amicon Ultra 3 K device (Merck Millipore, Darmstadt, Germany). Quantification was performed using an external calibration peak area measurement. The HPLC conditions are provided in the Supporting Information.

Amino Modification of the Silica Beads and Amino Loading Estimation. Silica beads (~50 mg)
were washed with ethanol and resuspended in 1 mL of ethanol, followed by an addition of 10 μL (1%) APTMS and shaking for 1 h at room temperature. After amino modification, the beads were washed 5 times with ethanol and resuspended in 1 mL of methylene chloride, which was followed by an addition of 20 μL (0.12 mmol) of DIEA and sonication for 5 min. To estimate the available amino groups, 25.8 mg (0.1 mmol) of Fmoc-Cl were added into the suspension, and the mixture was shaken for 1 h at room temperature. Following centrifugation at 5000 rpm for 5 min, the beads were washed with methylene chloride and ACN 5 times, sequentially. In this study, the unreacted amino groups were capped using a mixture of 0.2 M acetic anhydride and 0.2 M DIEA in methylene chloride (1 mL for 50 mg of beads) with shaking overnight at room temperature. The beads were then washed with methylene chloride and ACN 5 times again. Finally, the Fmoc protecting group was removed using 1 mL of 20% piperidine in DMF. After shaking for 30 min, the supernatant was collected and analyzed using HPLC.
Immobilization of the DNA-peptide Probe. The amino-modified beads (50 mg) were suspended in 1 mL of DMF containing 10% pyridine and 0.2% PDITC and shaken for 2 h at room temperature. The beads were then washed with DMF 5 times, with ethanol 3 times and with methylene chloride 3 times. Afterward, 0.75 nmol of the DNA-peptide probe and 10 mg of the beads were mixed in 400 μL of the reaction buffer, which included 2 M sodium chloride and 0.05 M sodium borate buffer at pH 8.5. After shaking overnight at 37 °C, the beads were washed with PBS 3 times.
miRNA Hybridization with the DNA-peptide Probe. After immobilization, 1 nM miR-21 in the hybridization buffer (200 μL) was mixed with 20 μL of the probe immobilized beads (0.25 mg), and the reaction was conducted in an oven. Similar to previous work 38 , the hybridization buffer (Table 1S), time and temperature were optimized. After hybridization, the beads were thoroughly washed to remove any unbound miRNAs. To evaluate the hybridization specificity of the probe, the miR-21 molecules with a single-base mismatch (5′-UAGCUUAUCAGUCUGAUGUUGA-3′) and two-base mismatch (5′-UAGCUUAUCAGUGUGAUGUUGA-3′) sequences were also examined. The underlined bases in the sequences represent the mismatch sites.

In-solution MBN Digestion and Tryptic Digestion.
After hybridization, 0.25 mg of the beads were treated with 40 units of MBN for 20 min at room temperature in 200 μL of the reaction buffer containing 30 mM sodium acetate (pH 5.0), 50 mM sodium chloride and 1 mM zinc chloride 47,48 . The beads were then washed with 10 mM PBS 3 times. After centrifugation, 200 μL of 50 mM NH 4 HCO 3 were mixed with the beads and followed by the addition of 1 μg of sequencing grade trypsin and incubation at 37 °C for 24 h. Then, 10 μL of 0.1% TFA were added to stop the reaction. After that, the tryptic mixture, with an addition of 100 μL of the internal standard solution, was transferred into a microspin C18 column (The Nest Group, Inc., MA, USA). The column was preconditioned using 100 μL of ACN and 100 μL of water in advance. After loading, the column was washed with 50 μL of 5% ACN containing 0.1% TFA and eluted with 50 μL of 80% ACN containing 0.1% FA. The elution was repeated for another 3 times. For additional detailed information, please see the manufacturer's operating instruction.

LC-MS/MS Method Development and Validation. The LC-MS/MS analysis was performed using an
Agilent Series 1290 UPLC system (Agilent Technologies, Waldbronn, Germany) coupled to a 6460 Triple Quad LC-MS mass spectrometer (Agilent Technologies, Santa Clara, USA). Chromatographic separation was achieved using an Agilent SB C18 (2.7 µm, 30 mm × 2.1 mm) at room temperature. The analysis was performed in the positive ion mode. The mobile phase A was 0.02% FA in water, and the mobile phase B was 0.02% FA in methanol with a flow rate of 0.3 mL/min. The elution gradient started with 10% of eluent B run at isocratic conditions for 1 min, and then, it was increased to 90% in 4 min and held for another 4 min. Finally, the initial condition was reached again in 1 min. The sample injection volume was set at 5 μL.
The mass spectrometer was equipped with an electrospray ion source. The following parameters were used: Q1 and Q3: set at unit resolution; the drying gas: 10 L/min, 350 °C; needle voltage: 4000 V; nebulizer pressure: 35 psi. The data were obtained and processed using the Agilent MassHunter Workstation Software (version B.06.00) The precision and accuracy of the assay were determined using QC samples. The detailed validation procedures and acceptance criteria for the assay have been discussed in a number of studies 31

Results and Discussion
A Quasi-direct Targeted Proteomics Approach. The goal of this approach was to convert the miRNA signal into a reporter peptide for mass spectrometric analysis. For the scheme to work, an appropriate reporter/ substrate peptide must be found. Then, the N-terminal maleimidohexanoic acid-modified substrate peptide was conjugated with the 3′-end thiol-modified complementary DNA to form the DNA-peptide probe (Fig. 2, reaction  3). Instead of the previous modification of the target miRNA, the DNA-peptide probe was covalently immobilized on the amino-modified silica nanoparticles via a PDITC reaction in this study (Fig. 2, reactions 1, 2 and 4). After reacting with the target miRNA (i.e., miR-21) under the optimized conditions, the miRNA-DNA-peptide hybrids were formed. The excess single-stranded DNA-peptide probe was degraded via a single-stranded specific nuclease MBN. Finally, the beads bearing the hybrids were subject to trypsin digestion. Thus, the reporter peptide could be released from the beads and detected using LC-MS/MS.

Selection of the Reporter and Substrate
Peptides. In our previous work, the criteria for the selection of suitable reporter and substrate peptides were extensively described 38 . While evidence indicated that our previously designed surrogate/reporter peptide was applicable to targeted proteomics analysis, these peptides may not be suitable for the present work because of the probe immobilization process.
To date, oligonucleotide immobilization has been widely used in a number of important molecular applications 50 . There are several strategies available for immobilization 51 . Among them, the most accessible approach is covalent attachment of oligonucleotides to the solid support 52 . Normally, oligonucleotides are modified at their end-group and then react with the functional groups on the solid support 53,54 . The combinations of oligonucleotide/surface functional groups that have been reported so far include thiol/acrylamide 55 , amine/isothiocyanate 52 , activated carboxylic acid/amine 56,57 , amine/aldehyde 58,59 and epoxide/amine 60 . Among them, the amine/isothiocyanate reaction under mildly alkaline conditions can provide a high immobilization efficiency for oligonucleotides, which has been proven by many studies 52,61,62 . Thus, PDITC, a member of the isothiocyanate family, was selected to crosslink the DNA-peptide probe to the beads. In principle, this cross-linker is capable of reacting with the amino-modified beads at one end and with the DNA-peptide probe at the other end (Fig. 2, reactions 3 and 4) 52 . Its preferred reactivity with a primary amine is essential for the product yield 63 . However, additional consideration is required in this study because of the potential interference from the peptide component of the probe.
As is well known, the primary amino groups of the peptides exist both at the N-terminus (α-amino group) and in the side chain of the lysine residues (ε-amino group) 64 . The N-terminal amino group was formerly modified and conjugated with the DNA in the probe formation, whereas the ε-amino group was free and could form a stable phenylthiourea (PTU) derivative during immobilization, if it exists 65 . Thus, the previously selected substrate peptide, GDKAVLGVDPFR, was no longer suitable because of the presence of the lysine residue at position 3. Given the biochemical similarity of lysine and arginine 66,67 , the replacement of lysine with arginine was easily accomplished while preserving the original tryptic site. The new substrate peptide was GDRAVQLGVDPFR, and the corresponding reporter peptide was AVQLGVDPFR. Fortunately, the sequence of the reporter peptide did not match any protein based on a BLAST search, and the peptide had a high predicted mass response (ESP Predictor (0.90, full score is 1). Experimentally, the maximum response was afforded via its doubly charged ion at m/z 551.3. Its product ion spectrum is shown in Fig. 3A. The sequence-specific b ions and y ions were observed and were indicative of the precursor peptide. The corresponding stable, isotope-labeled peptide was also synthesized to serve as an internal standard. For this peptide, the stable, isotope-labeled [D 8 ] Val (V*) was coupled to AVQLGVDPFR at positions 2 and 6 to form AV*QLGV*DPFR, which yielded a molecular mass shift of 16 Da from the non-labeled peptide and a monoisotopic molecular mass of 1117.6 Da. The retention times for AVQLGVDPFR and its isotope-labeled peptide were identical (∼3.3 min, Fig. 1S-A). Finally, the MRM transitions afforded by the product ions, b2 m/z 171.1, y3 m/z 419.2 and y6 m/z 690.3, gave the best signal-to-noise (S/N) and limit of quantification (LOQ) (Fig. 3B). This characteristic pattern was also observed in AV*QLGV*DPFR (Fig. 1S-B). The quantification limit of these MRM transitions was 5 pM (Fig. 2S). Therefore, the peak areas from the three transitions, m/z 551.3 → m/z 171.1, m/z 551.3 → m/z 419.2 and m/z 551.3 → m/z 690.3, were summed and used in the following quantitative analysis 68 .
The digestion efficiency of GDRAVQLGVDPFR was calculated by comparing the response ratios of the equimolar peptide before and after digestion. The estimated value was 99.91% (Fig. 3S). To note, the subsequent conjugated DNA, hybridized miR-21 and the probe-coated silica beads did not have a significant impact on the digestion (please see the following section). The reason may be that the tryptic site was intentionally designed using the three amino acid residues (GDR) and the C 6 spacer (on DNA) away from the DNA sequence to avoid any potential steric hindrance in the trypsin cleavage. Meanwhile, this design also reduced the potential spatial interference of the peptide tag on the reactivity of the DNA.

Preparation and Characterization of the DNA-peptide Probe.
To prepare the DNA-peptide probe, the disulfide at the 3′ end of the DNA was first reduced to a thiol, and, then, a Michael addition reaction was performed between the reactive thiol and the maleimide at the substrate peptide N-terminus to form the thiol-maleimide linkage (Fig. 2, reaction 3) 69 . In addition to the factors previously discussed 65 , there is another factor that deserves consideration. The newly designed amino-modified 5′ end of the DNA as well as its thiol group at the 3′ end can react with the double bond of the maleimide to form stable carbon-nitrogen and carbon-sulfur bonds, respectively 70 . Fortunately, there is evidence that the maleimide reaction is specific for thiols at pH 6.5 ~ 7.5 and for the amino group when the pH is greater than 8 71 . After careful evaluation of the pH effect, 10 mM PBS (pH = 7.35) was selected as the conjugation buffer in this study. To further rule out the possibility of cross-reactivity (Fig. 4S), we used disulfide-protected DNA to react with the peptide. As shown in Fig. 5S, no new compound was observed in the HPLC chromatogram.
Under the above optimized conditions, almost all the DNA molecules were conjugated with the substrate peptide in the presence of excess peptide. The product was immediately purified using preparative HPLC. As shown in Fig. 4A, the DNA before/after reduction and the DNA-peptide were detected in the HPLC chromatograms using a detection wavelength of 260 nm; their retention times were 7.8 min, 16.8 min and 21.4 min, respectively. The peptide was undetectable at this wavelength. To ensure that the DNA-peptide conjugate we collected did not co-elute with the substrate peptide, a 220 nm detection wavelength was chosen for peptide detection under the same conditions. Figure 4B indicates that the substrate peptide was not eluted out of the column. For another probe characterization, trypsin was added to the collected fraction. The expected products, DNA-GDR and AVQLGVDPFR (i.e., reporter peptide), were detected using HPLC and LC-MS/MS, respectively. Consistent with the previous observation 38 , the retention time of the DNA product (9.25 min in HPLC) was not the same as that of the premier DNA because of the addition of three amino acid residues from the substrate peptide (Fig. 5A). However, the retention of the peptide product agreed with that of the reporter peptide (Fig. 5B). Furthermore, the estimated amount of the reporter peptide was not significantly different from the amount of the DNA-peptide reactant, which confirmed that the conjugation of the DNA to the substrate peptide did not have a significant impact on digestion.
Optimization of the DNA-peptide Probe Immobilization. As mentioned earlier, a variety of immobilization strategies, including entrapment, adsorption and chemical binding, have been used to attach oligonucleotide to solid supports, such as ceramics, silicon, glass, magnetic beads, nylon and polymers 57, 72-74 . Among them, covalent immobilization, owning to the involvement of chemical derivatization, is one of the most common approaches. In the present study, the silica beads were first silanized via APTMS to generate an active amino group prior to the PDITC reaction.

Optimization of the APTMS Concentration.
To evaluate the influence of the reactant APTMS concentration on the availability of amino groups on beads, the reaction was performed with an increase in the APTMS concentration from 0.01% to 2.5%. Because it is normally difficult to directly detect immobilized amino groups, a derivatizing agent, Fmoc-Cl, that can react with free amino groups was introduced. After attachment, the Fmoc group was removed, and its absorbance was monitored using a UV detector to estimate the available amino groups 69,75 . The results indicated that the amount of the amino groups increased as the concentration of APTMS increased from 0.1% to 1% and reached a plateau afterward (Fig. 6A). Therefore, 1% APTMS was used in the subsequent experiments.
Optimization of the Probe Concentration. After determining the availability of the amino group, the PDITC reaction was performed with various concentrations of the DNA-peptide probe, ranging from 0 μM to 3.75 μM. The immobilization efficiency was calculated as the input of the DNA-peptide probe divided by the amount of the probe immobilized, which can be estimated from the probe amount left in the solution. As shown in Fig. 6B, there was a direct and positive relationship between the input and the immobilized probe. However, the level of immobilization efficiency decreased throughout the range. This phenomenon may result from the steric hindrance of the immobilized probes that block the access of the free probes to the adjacent amino sites 52 . Taking the immobilization amount and efficiency into account, we chose 1.875 μM as the probe reaction concentration with 87.3% immobilization efficiency.
Hybridization Efficiency. In this study, the hybridization efficiency was assessed using the molar ratio of the hybridized miRNA and reactant miRNA 76,77 . Its value was influenced by many parameters, including hybridization buffer, temperature and time. Six commonly used buffers were examined, and a buffer of 10 mM Tris, 100 mM KCl, and 1 mM MgCl 2 at pH 7.4 (buffer 3) provided the highest efficiency (Table 1S, Fig. 6S). In addition, the optimum temperature and time were 62 °C and 12 h, respectively. Under these conditions, the hybridization efficiency reached 90.2%.
The distinct nature of miRNA concerning detection specificity is the presence of highly homologous family members. Their sequences differ by only 1-3 nucleotides 10,11 , which may create extra difficulties when it is necessary to distinguish specific miRNAs from their family members. To evaluate the specificity of this assay, we also applied the assay to single-base mismatched miR-21 (SM) and double-base-mismatched miR-21 (DM). The hybridization efficiency decreased in the order of matched (100%) > SM (9.6%) > DM (3.5%). As predicted, the non-complementary sample did not display a peak.
Mung Bean Nuclease Treatment. MBN is a single-stranded specific nuclease purified from the sprouts of the mung bean, Vigna radiate 72 . This enzyme prefers to degrade single-stranded DNA (ssDNA) over double-stranded DNA (dsDNA) or DNA-RNA complexed by 30,000-fold 78,79 . In addition, the enzyme can cleave at a single-nucleotide mismatch when two nucleotide strands are not a perfect match. However, the enzyme can still potentially degrade double-stranded DNA at very high concentrations 78,80 . Thus, the specificity of MBN was evaluated. The results indicated that up to 200 U/mL of MBN can digest ssDNA completely without affecting dsDNA (Fig. 7S).

Validation of the Quasi-direct Targeted Proteomics Assay.
After determining the conditions for the immobilization, hybridization and digestion, the quasi-direct targeted proteomics assay was validated for further application to biological samples. Calibration curves were generated using a weighted linear regression model with a weighting factor of 1/x 2 . The relative peak area ratio between the reporter peptide and the stable isotope-labeled internal standard was plotted against the concentration. For the assay, the LOQ was 5 pM, and the established concentration range was from 5 pM to 10 nM (Fig. 7). Notably, the LOQ was chosen as the concentration corresponding to the lowest standard of the calibration curve that gave good accuracy and precision as suggested by FDA 81 . However, we have to admit that the LOQ is at least 5 pM in neat buffer conditions, but may be higher when realistic sample matrix is present in the sample. The other results are provided in the Supporting Information (Table 2S).

Quantification of miR-21 in Breast Cells and Tissue Samples.
Using the assay, the levels of miR-21 were accurately quantified as (7.94 ± 1.62) × 10 3 copies/cell in MCF-10A cells, (3.06 ± 0.45) × 10 4 copies/cell in MCF-7/WT and (4.90 ± 0.66) × 10 4 copies/cell in MCF-7/ADR cells. Their difference was statistically significant (p < 0.05). This result agrees well with the results reported in our previous work 38 , but all the values are higher. One of the possible reasons could be the indirect nature of the previous method, i.e., miR-21 was biotinylated and bound to streptavidin agarose. The biotinylation process may cause information loss of the target miRNA because of the complexity of the biological samples, e.g., the ubiquitous presence of biotin 44 . Finally, miR-21 was measured using quantitative qRT-PCR for a comparison (Fig. 8S). The level of miR-21 determined using our assay was slightly higher than that obtained using qRT-PCR, but the difference was not significant. Such uncertainty could be attributed to the larger variations in the qRT-PCR experiment. qRT-PCR is the most widely reported method and is regarded as a gold standard for quantifying miRNAs because of its high sensitivity 82 . However, this method suffers from practical issues such as low efficiency, high false positive rates for amplification, and sophisticated and expensive analysis 83 . Currently, most qRT-PCR analyses determine the relative miRNA abundance (often with respect to a non-validated reference miRNA) 84 . Absolute quantification via qRT-PCR can provide a quantity of unknowns, but it is labor-intensive 85 . Minor variations in the reaction components, thermal cycling conditions, and mispriming events during the early stages of the reaction can lead to large changes in the overall amount of the amplified product 86 . Furthermore, the subsequent analysis is also mathematically complex 83 . Thus, our assay is able to provide the miR-21 level in copies per cell while avoiding complicated sample manipulation and data analysis.

Figure 7.
A representative calibration curve (5 pM to 10 nM) for the miR-21 standards. Relative peak area ratio of the reporter peptide and the stable isotope-labeled internal standard with the same sequence was plotted against concentration.

Conclusions
In this study, we immobilized the DNA-peptide probe on amino-modified silica beads and developed a quasi-direct targeted proteomics assay for miRNA quantification. Using this assay, the target miR-21 was quantified in 3 breast cell lines and 36 pairs of breast tissue samples. The advantages of the quasi-direct targeted proteomics approach were demonstrated by converting the miRNA signal to the mass response of the reporter peptide. More importantly, immobilization of the DNA-peptide probe circumvents the miRNA biotinylation and subsequent dependence on the biotin-streptavidin interaction and allows the use of RNA samples without any further manipulation. Technically, this quasi-direct targeted proteomics method can be easily applied to other miRNAs by replacing the DNA sequence with one that is complementary to the target miRNA while keeping the reporter peptide the same. However, the parameters, including conjugation, immobilization, hybridization and digestion, deserve careful optimization to achieve the highest sensitivity and specificity for each miRNA. Furthermore, this assay has more potential for simultaneous detection of multiple miRNAs. Indeed, a key advantage of the LC-MS/MS-based quasi-targeted proteomics assay is its multiplexing ability, which is valid as long as a mass spectrometer can manage the concomitant analysis of multiple reporter peptides while retaining a degree of selectivity. However, the challenges of optimizing the assay format for each peptide, selecting a common dilution factor, addressing the variability and cross-interference, and establishing a robust quality control algorithm are substantial and require further analytical and statistical development. We anticipate that the quasi-targeted proteomics approach described here can ultimately be applied to the profiling of miRNAs in biological samples. This type of quasi-direct analysis may increase the quantitative accuracy and precision, but more evidence is required to confirm this feature.