Combined RT-qPCR of mRNA and microRNA Targets within One Fluidigm Integrated Fluidic Circuit

Abstract

The ability to profile expression levels of a large number of mRNAs and microRNAs (miRNAs) within the same sample, using a single assay method, would facilitate investigations of miRNA effects on mRNA abundance and streamline biomarker screening across multiple RNA classes. A protocol is described for reverse transcription of long RNA and miRNA targets, followed by preassay amplification of the pooled cDNAs and quantitative PCR (qPCR) detection for a mixed panel of candidate RNA biomarkers. The method provides flexibility for designing custom target panels, is robust over a range of input RNA amounts, and demonstrated a high assay success rate.

INTRODUCTION

Multiplexed or highly parallel formats for reverse transcription qPCR (RT-qPCR) are useful tools for sensitive RNA quantification¹ of sets of gene targets (typically, 48–384). Assay panels may include an entire class of targets, such as all known miRNAs² or all genes involved in a biologic process,³ or may represent a set of candidates identified by larger, whole-genome assays and chosen for independent confirmation. In either application, the ability to profile short miRNA and long mRNA targets in the same panel would provide an efficient means to relate miRNA regulatory effects to mRNA expression,⁴ or to profile both kinds of RNA biomarkers in the same biologic sample.⁵

Fluidigm (South San Francisco, CA, USA) dynamic arrays use microfluidics [an “integrated fluidic circuit” (IFC) connected to reagent input wells] and high-resolution imaging to perform qPCR with fluorescence detection in nanoliter reaction volumes.^{6, 7} IFC formats include 12 × 12 (12 cDNA samples tested for each of 12 targets in individual parallel reactions), 48 × 48, 96 × 96, and 192 × 24. Compared with 96- or 384-well plate formats, this microfluidics system allows more rapid screening of large sample sets and consumption of substantially lower amounts of PCR reagents, while avoiding the reaction-compatibility requirements of multiplex systems that combine several target detection assays in 1 reaction. Fluidigm provides separate protocols for panels of long or short RNA targets, with the long RNA version designed for custom panels of mRNAs and the short RNA protocol for the Applied Biosystems (Foster City, CA, USA) predesigned MegaPlex Pools of miRNA assays. The latter is an Advanced Development Protocol available upon request from Fluidigm. As a result of differences in reverse transcription, preassay amplification, target abundances and dynamic ranges, and qPCR conditions, no protocol for a customized combination of mRNA and miRNA targets has been previously reported to our knowledge.

By building on standard Applied Biosystems TaqMan^{8, 9} and Fluidigm^10–12 RT-qPCR methods, we describe the development of an mRNA + miRNA custom panel and a protocol for expression profiling within one IFC. For each RNA sample, separate reverse-transcription reactions produce cDNA and microcDNA products, which are then combined and used for a limited-cycle preassay amplification. This approach thus preserves the specificity of miRNA reverse transcription yet avoids the expense of 2 preassay amplifications. As qPCR input, the amplified cDNA is shown to provide robust target detection in 48 × 48 and 96 × 96 IFC assays for both classes of targets. Quantitative RNA profiling is demonstrated for candidate long and short RNA biomarkers in blood samples from orthopedic trauma patients over a range of input RNA amounts.

MATERIALS AND METHODS

Subject Enrollment and Sample Collection

Patients in the Penn Orthopaedic Trauma and Fracture Service and healthy volunteers were evaluated for inclusion/exclusion criteria and enrolled in an ongoing study. Adult, skeletally mature human subjects (age 18 yr or older) from 3 distinct groups, including 1 control group and 2 groups related to their sustained injuries, were recruited for this investigation. Healthy subjects were eligible if they had not sustained a bodily injury within 12 mo of enrollment. Acutely injured subjects who were hospitalized provided blood samples daily up to 10 d. Subjects with delayed healing were enrolled in the outpatient setting and provided samples at time of presentation and any subsequent interventions when necessary. Blood samples from human subjects were collected with their informed consent for research use, including genetic analyses. This study was approved by the institutional review board of the University of Pennsylvania (IRB Protocol #809094).

RNA Extraction and Quality Control

Whole blood collected in PAXgene Blood RNA tubes was processed, according to the manufacturer’s protocol (PreAnalytiX/Qiagen, Valencia, CA, USA), for purification of total RNA, including miRNA. RNA quality and quantity were assessed by spectrophotometry, fluorometry (Qubit RNA HS; Thermo Fisher Scientific, Waltham, MA, USA), and Bioanalyzer RNA LabChips (Agilent Technologies, Santa Clara, CA, USA).

Biomarker Panel Development

Microarray data (not shown) were analyzed for differential RNA expression between pairs of patient outcome categories. Biomarker candidates were ranked by false discovery rate (FDR)-adjusted P values after multiclass ANOVA and included 30 long mRNAs; 12 long-noncoding RNAs; 22 miRNAs; and 5 short-noncoding RNAs. The “short” class includes small nuclear and nucleolar RNAs (snRNA and snoRNA) that are typically longer than miRNAs and can be reverse transcribed by random priming. Two endogenous reference targets were added to the panel—human TATA box–binding protein (TBP) for mRNA and U6 small nuclear RNA 6, pseudogene (RNU6-6P) for short RNA—as well as 2 synthetic spike-in references—lacZ in Escherichia coli plasmid pUC19 (New England BioLabs, Ipswich, MA, USA) and an oligonucleotide for human miRNA hsa-miR-450a-5p (IDT Integrated DNA Technologies, Coralville, IA, USA). miR-450a-5p was chosen as a synthetic reference because it had not been detected in any of the blood RNA samples screened by microarrays. For RT-qPCR detection of the 73 targets, 60 TaqMan assays were ordered (Thermo Fisher Scientific), and 13 custom primer pairs with hydrolysis probes incorporating 5′-6-FAM/ZEN and 3′-Iowa Black fluorescence quencher were synthesized (IDT Integrated DNA Technologies). All targets, assays, and custom oligonucleotides are listed in Supplemental Table 1.

Reverse Transcription

These and subsequent reactions were assembled in 8-tube strips or 96-well plates by multichannel pipetting and performed in a PTC 225 Tetrad Thermal Cycler (MJ Research, now Bio-Rad Laboratories, Hercules, CA, USA). Long RNA targets were reverse transcribed by random priming, using up to 750 ng total RNA and the High-Capacity cDNA Reverse Transcription Kit (4374966; Thermo Fisher Scientific). A reaction master mix was assembled and added to each RNA sample, along with sufficient water, for a final volume of 10 μl, followed by incubation in a thermal cycler at 25°C for 10 min, 37°C for 120 min, 85°C for 5 min, and hold at 4°C (Fig. 1). A no-template control for long cDNA (long NTC) that contained no RNA was included among these reverse-transcription reactions.

Figure 1.

Sample preparation, part 1. rxn, reaction; NTC, no-template control.

miRNA targets were reverse transcribed with a pool of the RT primers taken from TaqMan Small RNA Assays (Thermo Fisher Scientific) and up to 350 ng total RNA using the TaqMan MicroRNA Reverse Transcription Kit (4366596; Thermo Fisher Scientific). A reaction master mix was assembled, spiked with 0.05 fmol/reaction synthetic miR-450a-5p, and added to each RNA sample, along with sufficient water, for a final volume of 12 μl. Reactions were held on ice for at least 5 min, followed by incubation in a thermal cycler at 16°C for 30 min, 42°C for 30 min, 85°C for 5 min, and hold at 4°C (Fig. 2). A micro-NTC that contained no sample RNA was included among these reverse-transcription reactions. cDNA products from long and short RNA reverse transcription were stored at −20°C.

Figure 2.

Sample preparation, part 2. TE, Tris-EDTA.

Preassay cDNA Amplification

Custom-synthesized primer and probe oligonucleotides (Supplemental Table 1) were combined in TE buffer to create assay mixes for each target, containing 18 μM of each of the 2 primers and 5 μM of the hydrolysis probe (equivalent to a 20× TaqMan assay mix). A diluted primer pool was created (also containing probes, irrelevant to this reaction) by combining 10 μl of each long RNA assay and 6 μl of each miRNA assay for all targets in the panel. The resulting concentration of each assay was determined and used to calculate the dilution factor necessary for a final concentration of 0.05× for long RNA primers and 0.03× for miRNA primers (Fig. 3). Spreadsheets for making these calculations, based on the target panel composition and desired reaction volume, are provided in Supplemental Tables 2 and 3. cDNA amplifications were performed in 7 μl reactions using TaqMan PreAmp Master Mix (4488593; Thermo Fisher Scientific), 1.1 μl of the primer pool, 1.2 μl long cDNA or reverse-transcription long NTC, and 1.2 μl microcDNA or reverse-transcription micro-NTC (Fig. 3). The thermal cycler program was 95°C for 10 min, 55°C for 2 min, and 72°C for 2 min, followed by 15 cycles of 95°C for 15 s and 60°C for 4 min, and completed with 99.9°C for 10 min and hold at 4°C. A 5 μl aliquot of the amplification product was archived at −20°C. The remaining 2 μl was mixed with 29 pg pUC19 (2 μl), and 16 μl TE buffer was added to dilute the amplified cDNA 1:10 before storage at −20°C.

Figure 3.

Sample preparation, part 3.

qPCR

Real-time quantitative fluorescence detection of PCR products was performed in a Fluidigm system consisting of a BioMark HD instrument, IFC MX and HX Controllers, and 48 × 48 or 96 × 96 Dynamic Arrays using the manufacturer’s protocols for standard TaqMan assays (PN 68000089 H1, 68000130 D1, and 68000088 J1; Fluidigm). A pilot 48 × 48 array tested a subset of the target panel (IFC 1; Supplemental Table 1), 47 amplified cDNA samples, and one NTC from preassay amplification. A set of 285 amplified cDNA samples (including the pilot 47; IFC 2–4) and a subsequent set of 631 amplified cDNA samples (including all previous samples; IFC 5–11), were assayed on 96 × 96 arrays with an NTC on each array. Assays on these arrays included the pilot 48 target panel, supplemented with 25 additional candidate RNA biomarkers (Supplemental Table 1) and a second well for 23 targets to evaluate intra-array variation.

Data Analysis

BioMark HD data processing parameters were Linear (derivative) for the baseline correction method and Auto (global) for the cycle threshold (Ct) method, software version 4.1.3. Technical performance was assessed using the Ct Value column from BioMark HD output tables (rawCt). rawCt values of 38 or higher were designated as no detection. Targets with rawCt < 38, in at least 90% of cDNA samples, were designated as consistently detected. Biomarker expression was normalized to reference genes by the comparative Ct (ΔCt) method¹³ (normCt) and transformed to linear scale before statistical testing. ANOVA t tests in Partek Genomics Suite (Partek, St. Louis, MO, USA) were used to test for likelihood of differences between technical classes, and the significance threshold was set at 0.05 for unadjusted P values and the step-up FDR (an adjusted P value incorporating multiple test correction).

RESULTS

Detection of Reference Targets

Reverse transcription and preassay amplification (Figs. 1–3) were performed once for each RNA sample, and the resulting cDNA was used for 3 qPCR batches. IFC 1 tested 48 targets in 47 cDNA samples, IFC 2–4 tested 73 targets in 285 samples, and IFC 5–11 tested 73 targets in 631 samples. Each batch included all targets and samples tested in the previous batch, and IFC 2–4 and IFC 5–11 included replicate reactions for 23 targets. In IFC 5–11, the endogenous reference targets TBP and RNU6-6P were detected in 99.7% and 96.7% of blood RNA samples, with 1 case of weaker detection in an NTC (Table 1). The synthetic spike-in reference targets, lacZ and hsa-miR-450a-5p, were intentionally included in NTC samples and were detected in 99.2% and 99.7% of samples (Table 1).

TABLE 1.

Reference Target Detection and Average rawCt

RT-qPCR arrays, IFC	No. cDNA samples	TBP			lacZ			RNU6-6P			hsa-miR-450a-5p
		Det	Avg Ct	sd	Det	Avg Ct	sd	Det	Avg Ct	sd	Det	Avg Ct	sd
1	47	47	11.15	0.92	47	18.37	0.66	47	19.59	1.88	47	15.05	1.20
1	1 NTC	0			1	19.21		0			1	14.01
2–4	285	284	11.65	1.30	281	17.93	1.07	263	19.85	1.35	284	14.87	1.11
2–4	3 NTC	1	21.78		3	19.10	1.42	0			3	13.91	1.13
5–11	631	629	11.79	1.25	626	17.90	0.96	610	18.66	1.47	629	14.01	1.44
5–11	7 NTC	1	23.87		7	18.34	0.39	0			7	13.24	0.97

Det, Number of samples with detected target (rawCt < 38).

Detection of Experimental Targets

Approximately 85% of targets were consistently detected (rawCt < 38 for at least 90% of samples) in 48 × 48 (IFC 1) or 96 × 96 format (Table 2 and Supplemental Table 1). Long-noncoding RNA targets and custom-synthesized primer/probe assays had lower detection rates (70–78%) than the other classes of targets and catalog TaqMan assays. The maximum RNA expression level was a rawCt of 2.5, and intra-array CVs were lower than interarray CVs (Table 2). Long-noncoding RNA and custom assays again were outliers with higher interarray CVs. Both intra-array (Fig. 4A) and interarray variance (Fig. 4B) tended to be greater for high-abundance targets (rawCt 5–10) compared with moderate-abundance RNAs (rawCt 10–15).

TABLE 2.

Assay Performance by Target Class and Assay Source

Performance metric	All	Long mRNA	Long noncoding	Short noncoding	miRNA	TaqMan	Custom
IFC1 target	48	18	9	6	15	35	13
IFC1 targets detected	42	17	7	5	13	32	10
IFC2–11 targets	73	32	11	7	23	60	13
IFC2–11 targets detected	61	30	8	6	17	52	9
IFC1–11 rawCt minimum	2.50	2.50	2.50	2.50	2.50	2.50	2.50
IFC1–11 rawCt maximuma	37.91	29.47	37.91	36.63	37.56	37.56	37.91
IFC2–11 intra-array median CV	1.65%	1.65%			1.53%	1.76%
IFC1–11 interarray median CV	6.21%	6.51%	11.99%	6.12%	6.24%	6.13%	10.70%

CV, coefficient of variation.

^aNot including rawCt > 38.

Figure 4.

Assay variance for the combined mRNA + miRNA protocol. A) CVs were calculated for 10 long RNA targets (circles) and 9 miRNAs (squares) using rawCt values from IFC 2–4 and IFC 5–11 (2 qPCR batches, plotted as separate points). The CV was derived from 3 measurements for each target performed within every IFC and thus. reflects intra-array variation for consistently detected targets. Averages of CV and rawCt are across all samples in a qPCR batch. B) CVs were calculated for 27 long RNA targets (circles) and 14 miRNAs (squares) shared by both IFC formats (IFC 1 and IFC 2–11) and consistently detected. The CV was derived from 4 assays of the same sample, 1/qPCR batch (IFC 1, IFC 2–4, IFC 5–11), to measure interarray variation. Averages of CV and rawCt are across 47 samples. The 3 average CVs at or above 15% were from custom-synthesized assays for short-noncoding RNA targets.

The total RNA samples extracted from blood were generally purified to 10–1000 ng/μl, and the original concentration of the RNA sample had no apparent influence on assay success (Fig. 5A). Of the 631 samples assayed on IFC 5–11, 516 had sufficient RNA concentrations to use the maximum input mass allowed by the protocol’s reverse-transcription reactions. Rather than attempting to concentrate the remaining samples further, the maximum input volume was used, and thus, a range of input RNA mass was tested. As shown in Fig. 5B, the measured abundances of TBP and RNU6-6P were consistent to a lower boundary of ∼200 ng input RNA. The protocol’s tolerance for reduced input mass was similarly apparent when assessed by counting the total number of assays that successfully detected a reference or experimental target (Fig. 5C).

Figure 5.

Assay performance over a range of input RNA concentrations and masses. A) Abundance measurements for TBP (circles) and RNU6-6P (squares) are plotted for the 631 samples assayed in the third qPCR batch (IFC 5–11). Lower Ct values reflect higher RNA abundance on the inverted y-axis. Detection failures (rawCt 38) were rare and occurred in samples throughout the range of RNA concentrations achieved after extraction from blood. B) TBP and RNU6-6P assay results are plotted as in A but over the range of input RNA mass used for reverse transcription. Maximum inputs for the protocol are 750 ng for long RNA and 350 ng for miRNA, and the maximum was used if allowed by the original RNA concentration and standard reaction volume. Samples with insufficiently high RNA concentration were used at the largest volume possible for the standard reaction, resulting in the range of input masses shown. C) The number of targets detected in each sample is plotted versus input mass, as described for B. In the IFC 5–11 panel, detection is possible for a total of 49 cDNA assays (circles) and 24 microcDNA assays (squares).

Normalization across Experimental Conditions

Endogenous reference targets are used to control for variations in sample quality and mass by providing a common denominator to normalize measurements of all other targets tested. An important assumption for this normalization is that the experimental conditions under study do not affect the natural abundance levels of the reference target. Synthetic reference targets are spiked into biologic samples before testing and can measure technical sources of variability but are less informative about biologic sources (other than the potential presence of PCR inhibitors). The endogenous and synthetic references in our biomarker panel were analyzed relative to subjects’ clinical categories, outcomes, and time points (Fig. 6). Whereas fluctuations in rawCt were observed over the time courses (particularly for RNU6-6P), this variation was generally within 1 sd, and no trends appeared to correlate with time. Likewise, no confounding differences were apparent among the 3 healing outcomes of acute injury patients (Fig. 6A) or among acute injury, healthy (Fig. 6B), and nonunion (Fig. 6C) subjects. Multiclass ANOVA for IFC 5–11 data indicated no significant differences (P < 0.05) between healing times, within or among subject categories, for the normalization references (Supplemental Table 4).

Figure 6.

Reference target measurements within and among experimental conditions. A) The average expression levels (average rawCt, left axis) of TBP (red), lacZ (orange), RNU6-6P (blue), and hsa-miR-450a-5p (purple) are plotted from IFC 5–11 for samples from each blood-collection time period: 1a, wk 0–2; 1b, wk 3–4; mo 2–6; and mo 7 and later. Time 0 is the date of fracture for acute-injury patients who healed normally (AInorm), healed slowly (AIslow), or were later diagnosed with nonunion (AInu). The number of samples used to calculate average rawCt is indicated by circles (right axis), and error bars are ±1 sd. B) Samples from healthy volunteers (HV) are plotted as described in A. Time 0 is the date of first blood collection. C) Samples from patients with a confirmed fracture nonunion (NU) are plotted as described in A. Time 0 is the date that therapy for nonunion (usually revision surgery) was initiated.

rawCt values for short- and long-noncoding RNA targets and mRNA targets were then normalized to TBP or lacZ, and miRNA target data were normalized to RNU6-6P or hsa-miR-450a-5p. The likelihood of difference between qPCR batches was tested by multiclass and pairwise ANOVA for the 41 consistently detected targets and 47 samples in common across IFC 1, IFC 2–4, and IFC 5–11 (Supplemental Table 5). Numerous significant differences (P value or FDR < 0.05) were detected using rawCt values, and pairwise tests showed that the 96 × 96 IFC 2–4 and IFC 5–11 batches were more similar than when either was compared with 48 × 48 IFC 1. Normalization to endogenous reference targets was somewhat more effective than normalization to synthetic references for reducing the batch effect. The Partek Batch Removal tool, an algorithm that transforms datasets by normalizing on a technical variable, while preserving differences between nonconfounded experimental conditions, eliminated the qPCR batch effect (Supplemental Table 5).

DISCUSSION

Despite a challenging mix of mRNA, noncoding RNA, snRNA, snoRNA, and miRNA targets and the inclusion of custom assays not previously validated, the combination RT-qPCR protocol provided a high rate of assay success across all RNA classes (84%; Table 2 and Supplemental Table 1). Four miRNAs and 3 mRNAs did not meet the consistent detection threshold (90% of samples) but had robust expression levels in at least 30% of blood samples and may be under on/off regulation in this patient cohort. The inclusion of these 7 targets increases the overall assay success rate to 93%. As with any multiplexed screening assay, results that are key to a biologic conclusion or identify important candidates should be confirmed for accuracy and reproducibility using an independent test.

Detection of reference targets TBP and hsa-miR-450a-5p (Table 1) was consistent across all samples but 2, both of which showed low or no signal for all attempted assays and likely contained a PCR-inhibiting contaminant. An additional 3 samples were negative for lacZ, a reference target that was added individually to each preassay amplification product and may have suffered drop-out during multichannel pipetting. Besides the 2 problematic samples, endogenous RNU6-6P was not detected in 19 samples, suggesting that a small proportion of subjects may not express detectable levels of this small RNA in blood. For such samples, alternative normalization references include synthetic hsa-miR-450a-5p or total signal from all targets.

A technical batch effect was observed using the same amplified cDNA samples and a set of assays shared by the 3 qPCR runs. This effect was apparent in the higher interarray CVs (Fig. 4B vs. A) and in batch ANOVA results (Supplemental Table 5). We did not test sample replicates on different IFCs within 1 batch, but we suspect the resulting interarray CVs would be closer to intra-array CVs if unaffected by batch artifacts. The first and second batches, IFC 1 and IFC 2–4, were performed within days of each other using PCR reagents from the same manufacturing lot and differed mainly in IFC format. The third batch, IFC 5–11, was performed several months later using Fluidigm and PCR reagents from different manufacturing lots. IFC 2–4 and IFC 5–11 data were more similar to each other than to IFC 1, so array format (48 × 48 vs. 96 × 96) may be a primary cause of this technical effect. Surrogate variable analysis,¹⁴ Partek Batch Removal, and similar algorithms can enable fair comparisons between data sets from different assay formats, times, or laboratories, but this should only be considered for retrospective studies and not as part of a preplanned analysis strategy. A project design that uses multiple IFC formats or substantially different qPCR batches is not recommended.

We believe the protocol will be applicable to many target panel designs and sample types. The panel used to develop this protocol contains candidate biomarker RNAs chosen after microarray screening, so this may have biased the observed assay success rate by selecting for targets that were detectable by another assay method. Sensitivity for detection and differential expression may differ from our results when using other tissue sources and RNA extraction methods. A pilot panel can be useful for validating assays and determining the optimal amount of input RNA for a particular experimental design. The inclusion of synthetic reference targets is optional, and the use of such back-up normalization references must be weighed against the reduced capacity for experimental targets in the panel. Whether endogenous or synthetic, we believe it is important to include references that reflect all assay types within the panel, which in this study, included random-primed cDNAs and specifically primed microcDNAs.

The most common method for normalization to a reference target (ΔΔCt) requires similar PCR amplification efficiencies for the test and reference genes before accurate biologic interpretations can be made.¹ We did not use such interpretations in this study, but we calculated amplification efficiencies for a separate panel of 90 mRNAs and 64 miRNAs assayed using the combined RT-qPCR protocol, and also measured assay responses over a dynamic range of input RNA. Starting with a standard RNA sample, a series of 10-fold dilutions showed the expected linear signal response for most assays (Supplemental Fig. 1). A cDNA pool derived from the standard RNA sample was used to create a 2-fold dilution series, and qPCR efficiencies (E) were determined⁷ (Supplemental Fig. 2). The average E for mRNA targets was 0.8048 compared with 0.8099 for TBP, and average E for miRNA targets was 0.9264 compared with 1.0282 for synthetic hsa-miR-450a-5p. E for RNU6-6P was 0.4503, which may contribute to the lower performance observed for this assay, as discussed above. These results illustrate that a multiplexed panel of analytes will have a range of responses for any assay characteristic, and assessments of dynamic range, limit of detection, PCR efficiency, and other parameters should be made for each iteration of a panel design. Data analysis can then be customized to adjust for a subset of outlier assays; for example, alternative normalization algorithms (see http://www.gene-quantification.com/efficiency.html) will be evaluated for targets in the distribution tails of Supplemental Fig. 2B.

The protocol’s random-primed reverse-transcription step is a standard reaction that accommodates a wide range of input RNA volumes. The miRNA reverse-transcription reaction (Figs. 2 and 3) uses a master mix that must contain a pool of all of the miRNA-specific primers and represents a significant proportion of the reaction volume. The limited remaining volume available for RNA samples may create difficulties when using samples with low RNA concentrations. This can be addressed by using an input mass <350 ng (apparently well tolerated; see Fig. 5), scaling up the component volumes, and/or further concentrating the RNA (the least preferred option). The creation of the miRNA primer pool by 1:100 dilution of 5× stocks also limits the panel design to 100 miRNA targets, unless primer stocks with higher concentrations can be obtained, or more than 1 reverse-transcription reaction per sample is performed using different primer subpools. Alternatively, larger dilutions to an even lower concentration of primers may still support successful reverse transcription, but this was not tested. Assembly of the preassay amplification primer pool, and the consequent volumes of reaction components (Supplemental Tables 2 and 3) are highly dependent on the mix of assay types in the custom panel. Preassay amplification is typically performed in 5–20 μl reactions, and therefore, this step would have a maximum capacity of ∼175 mRNAs + 1 miRNA, 300 miRNAs + 1 mRNA, 100 mRNAs + 120 miRNAs, and various intermediate compositions within these ranges. Again, >1 reaction could be performed for each sample using separate subpanels if a larger overall panel is desired and the resulting products pooled during the dilution step that precedes loading onto IFCs.