Digital-PCR for gene expression: impact from inherent tissue RNA degradation

Subtle molecular differences indicate the heterogeneity present in a number of disease settings. Digital-PCR (dPCR) platforms achieve the necessary levels of sensitivity and accuracy over standard quantitative RT-PCR (qPCR) that promote their use for such situations, detecting low abundance transcript and subtle changes from gene expression. An underlying requisite is good quality RNA, principally dictated by appropriate tissue handling and RNA extraction. Here we consider the application of dPCR to measures of gene expression in pathological tissues with inherent necrosis, focusing on rheumatoid subcutaneous nodules. Variable RNA fragmentation is a feature of RNA from such tissues. Increased presence of transcript fragmentation is reflected in a proportionate decrease in Agilent DV200 metric and downstream, a reduction in endogenous control genes’ expression, measured by RT-dPCR. We show that normalisation of target gene expression to that for endogenous control genes sufficiently corrects for the variable level of fragmented RNA. Recovery of target gene values was achieved in samples comprising as much as 50 percent fragmented RNA, indicating the suitability and appropriate limitation of such data treatment when applied to samples obtained from inherently necrotic tissues.

RNA concentration measurement prior to RT-dPCR 16 , although because many transcripts may not be sufficiently intact for complete reverse transcription, correction is required to ensure that the final data reflects gene expression in functional, non-necrotic areas of the tissue.
Here we consider the application of nanofluidic partitioning chip-based dPCR to quantitative gene expression in tissues complicated by necrosis. We sought to utilise this technique to exploit the increased detection sensitivity and precision capabilities attributed to dPCR, when compared to qPCR, while maintaining an existing, well-validated, sample processing workflow. Results show that measures of the expression of appropriate endogenous control genes, like GAPDH and HPRT1, are proportionate to the level of mRNA of quality sufficient to allow reliable reverse-transcription. We also sought to establish suitable inclusion criteria for samples according to the accuracy of normalisation. This approach limits the undesirable impact on target gene expression values that can arise when working with tissues containing inherent necrosis, while still maintaining the high level of detection sensitivity of the digital technology.

RNA fragmentation.
Retrospective histological analysis showed that all nodule tissues used for RNA extraction contained palisading monocyte/macrophages, focused on a central necrotic area (data not shown). This is a characteristic feature of granulomatous rheumatoid nodules, that we have previously described 5,17 .
RNA quality analysis revealed that RNA purified from nodule tissues was variably fragmented between samples, despite consistent and best practice for sample handling and RNA preservation (Fig. 1a). The extent of fragmentation in each sample was quantitatively expressed by using the DV 200 metric, a measure of the percentage of RNA transcripts >200 nucleotides in length. We compared the DV 200 metrics measured in nodule RNA samples with those in synovial membrane RNA, finding greater variability in RNA fragmentation in nodules ( Fig. 1b). Consequently, synovial RNA samples contain a significantly higher proportion of longer RNA transcripts (p = 0.0005). Both tissue types undergo similar handling suggesting that the quality of nodule RNA reflects in vivo cellular processes characteristic of these pathological, necrotic lesions.
Finally, we assessed GAPDH and HPRT1 expression in nodule samples and found a significant positive correlation (r = 0.9087, p < 0.0001) between combined expression levels for these two endogenous control genes and the DV 200 metric (Fig. 1c).

Validation study.
To investigate the application of normalisation to dPCR data, we generated samples of intact THP-1 cell-derived RNA spiked with known proportions of naturally, highly fragmented nodule RNA. As expected, the DV 200 metrics closely paralleled the proportion of spiked, fragmented RNA present in each of these samples (Fig. 2a). Expression of the endogenous control genes, GAPDH and HPRT1, significantly correlated with the DV 200 metrics (r = 0.9729; p = 0.0053) and with the proportion of intact starting RNA (r = 0.9891, p = 0.0014) (Fig. 2a). Interestingly, only when the proportion of intact RNA within spiked samples fell below 50 percent was there a noticeable drop in RNA Integrity Number (RIN) values (detailed in Fig. 2a legend).
RNA-seq data indicated low or undetectable expression of proteinase 3 (PRTN3) in nodules per se. Therefore, irrespective of quality, nodule RNA is not expected to contain transcript that might significantly influence measures of PRTN3 expression. We measured PRTN3 expression in the spiked samples using RT-dPCR. Predictably, when assessed per unit of starting RNA, the absolute copies of PRTN3 expression reduced as the proportion of fragmented RNA increased (Fig. 2b). Normalisation of PRTN3 expression values effectively corrected for the variable proportion of fragmented RNA, even in samples comprising 50 percent fragmented RNA (Fig. 2b). However, in those samples where the amount of fragmented RNA increased beyond 50 percent, there were indications that normalisation could result in some over-estimation of PRTN3 expression.
Application of data normalisation to pathological samples. We investigated IL27A gene expression in rheumatoid nodule tissues using dPCR, applying normalisation and appropriate corrections for RNA fragmentation. Within our nodule sample set (n = 27), absolute measures of IL27A expression ranged from 0.25 to 14.16 copies/ng RNA. As a point of reference, a sample returning absolute measures of 0.63 copies of IL27A/ng RNA using dPCR corresponded to an average cycle threshold (C T ) value of ~37 by qPCR.
Based on DV 200 metrics, nodule samples returning values below 50 percent were then excluded from analysis and we compared absolute measures with normalised values generated from the remaining 19 nodule samples (Fig. 3). Normalisation, resulted in significantly higher median values for IL27A expression (p < 0.0001).

Discussion
Key factors in conducting research in human disease are both the availability and quality of appropriate tissue samples. This combination is often a fait accompli, with potential confounders associated with available human tissue, including characteristics of any inherent pathological processes, likely to influence the quality of the tissue and yield of accurate molecular results. This study examined the characteristics of human rheumatoid subcutaneous nodules, an inflammatory lesion with inherent necrosis, associated with granulomatous inflammation. The goal was to identify those features that determine the suitability of variably necrotic tissue for downstream molecular analysis, particularly the impact on the detection of low abundance gene expression using digital PCR.
The critical influence of RNA integrity on downstream molecular analysis is well recognised. Development of measures of RNA integrity from the long-time standard based on 28 S to 18 S ratio, to a predictive model of RNA integrity, the RIN, now provides a much more reliable and robust measure that subsequently correlates with downstream analysis of gene expression 18 . Effectively, measures of RNA integrity generate exclusion criteria that dictate the suitability of samples. The dilemma is the application of these criteria to tissues where there is inherent necrosis as part of the underlying pathological process, which is likely to impact on RNA integrity. In rheumatoid subcutaneous nodules, necrosis is a focus of the inflammatory granuloma. However, tissue necrosis is common during the development of a wide range of diseases, varying in aetiology that includes a "programmed" or "regulated" process, activated in response to death receptor ligands or cellular stress [19][20][21] , or through a contribution to the cytotoxicity of many chemotherapeutic agents in cancer 10 . In such settings, restricting molecular analysis to high quality tissues limits the available tissue and is unlikely to reflect the true pathological process in operation.
Our results show that despite nodule tissue samples being harvested and treated using optimal RNA-preservation procedures, the presence of inherent, necrotic tissue contributes considerable amounts of highly fragmented RNA. In order to gauge the extent of RNA fragmentation, we considered the DV 200 metric, indicative of the proportion of extracted RNA transcripts greater than 200 nucleotides in each sample. This metric correlated significantly with expression levels of the endogenous control genes GAPDH and HPRT1, measured by dPCR. This indicates that amongst tissue samples where RNA fragmentation has been naturally generated in-vivo, the downstream quantitation of target gene expression predominantly reflects the levels present in that portion of RNA containing the most intact transcripts.
Reference gene normalisation is recommended for gene expression analysis using alternative dPCR approaches, thereby circumventing any impact from RNA sample contaminants 1,22 . Using experimentally generated mixtures of RNA and dPCR assays for proteinase 3 (PRTN3) expression, our results show that normalisation of gene expression data to endogenous control genes' expression, also accounts for the proportion of intact transcript present amongst highly fragmented RNA. Further, the normalisation of target PRTN3 expression in THP-1 cell RNA, was effective in samples containing as much as 50% fragmented nodule RNA. However, when the proportion of fragmented RNA increased beyond 50%, and up to 75%, normalisation tended to slightly over-estimate target gene expression. Thus, when considerable RNA fragmentation is present, the expression of endogenous control genes is affected and becomes a less accurate reflection of the intact, measurable portion of RNA. Consequently, normalisation becomes less reliable. Our data suggest that implementing a 50% fragmentation threshold, circumvents any problems associated with more fragmented RNA in pathological samples.
We evaluated nodule RNA samples for IL27A gene expression, comparing absolute (non-normalised) and normalised outcomes of dPCR. In these samples, expression of this cytokine gene was generally low by conventional qPCR. Applying the criterion of a DV 200 metric >50% to ensure reliable normalisation, we excluded 9 of 27 samples from the final data set but subsequently revealed meaningful values for IL27A expression in those remaining. We conclude that normalisation of target gene levels to suitable endogenous controls in RT-dPCR gene expression studies is appropriate when applied to pathological samples where inherent RNA fragmentation is present. The generation of reliable, normalised data also depends on the level of fragmentation within samples, and may require the implementation of appropriate exclusion criteria. In that regard, we do anticipate similar normalisation strategies could be applied when assessing gene expression in formalin-fixed, and paraffin-embedded (FFPE) tissues, where RNA fragmentation and low RIN scores 23 are an inevitable consequence of the treatment. A more likely factor limiting the assessment of gene expression within FFPE tissues is chemical modification of the RNA, fragmented or not. Consequently, adherence to the recommendations for processing and storage of FFPE tissue that is destined for gene expression analysis 23 and careful selection and validation of appropriate endogenous control genes 24 are additional factors for consideration.

Material and Methods
Patient demographics and tissue collection. Subcutaneous nodule tissues (n = 27) or joint synovial tissues (n = 27) were obtained from patients (n = 42) with rheumatoid arthritis (RA), as defined by 1987 American Rheumatism Association Classification criteria 25 , undergoing elective surgery for nodule removal. The patient cohort comprised those of age 64 ± 9 years (mean ± SE), disease duration 21 ± 3 years, with the majority positive for both rheumatoid facter and anti-citrullinated peptide antibodies. Five patients provided multiple nodules, and 3 further patients multiple synovia, either at the same or different times. The study was approved by the New Zealand Health and Disability Multi-Regional Ethics Committee (MEC06/02/003). All patients provided informed written consent prior to tissue donation and all experiments were performed in accordance with relevant guidelines and regulations.
Nodule and synovial membrane tissues were processed immediately following surgical excision in an RNase-free environment or collected into RNAlater (Invitrogen) at 4 °C. Apportioned tissues destined for gene expression studies were snap-frozen and stored in liquid nitrogen.
RNA extraction, analysis and reverse transcription. The RNeasy Mini Kit (Qiagen) was used according to manufacturers' recommendations, with DNase digestion either on-column using the RNase-free DNase Set (Qiagen) or post-extraction with PerfeCTa DNase I, (Quantabio). Eluted RNA concentration was determined using a Qubit Fluorometer (Invitrogen) with low or high sensitivity Qubit assay kit (Molecular Probes). Up to 1000ng of total RNA was added per 20ul reaction with Superscript III Reverse Transcriptase (Invitrogen) and primed with Oligo (dT) 20  Prior to running duplex reactions, assays were validated to ensure the duplex readings were consistent with respective single-plex assay readings over a range of input cDNA concentrations (data not shown). Quantstudio 3D AnalysisSuite Cloud Software (Applied Biosystems, version 3.1.2-PRC-build-03) with Poisson Plus quantification algorithm (version 4.4.10) and 95% CI was used. Imported eds files were individually assessed and fluorescence thresholds set manually to generate transcript copies/ul, then converted to copies/unit RNA.
To achieve data normalisation, copies/unit RNA of the target gene were multiplied by the corresponding normalisation factor (geometric mean of GAPDH and HPRT1 normalisation factors) and final data displayed as normalised copies/unit of RNA. The stable expression of the two endogenous control genes was previously assessed among equivalent nodule samples using qPCR with qbasePLUS v2.1 (Biogazelle) 26 . An average coefficient of variation (CV) of 23.3%, and M value of 0.658 for the two genes was generated.
In each sample, RNA quality parameters were assessed by NZ Genomics using an Agilent BioAnalyzer. Two hundred and fifty ng of total RNA from each prepared sample, verified by Qubit fluorometry (Invitrogen), was reverse transcribed and expression of GAPDH and HPRT1 endogenous control genes and the PRTN3 target gene measured with RT-dPCR.
Quantitative real-time PCR. Samples identified as having high, medium or low IL27A expression levels by dPCR were compared for IL27A expression in routine qPCR. Ten uL reactions containing Quanta ToughMix (Quanta Biosciences), Taqman Assay (Hs00377366_m1 (VIC)), and cDNA representing 31.25ng starting RNA, were assayed in triplicate using a Roche LC480 (Roche Molecular Systems) and standard cycling and detection protocols.
Statistics. Graphpad Prism V7.0b (Graphpad Software) was used for statistical analyses. An unparied t test with Welch's correction for unequal variance was used to compare the DV 200 metrics between 27 nodules and 27 synovia, following a D' Agostino & Pearson test establishing normal distribution (p > 0.05). The geometric mean of the GAPDH and HPRT1 expression was correlated with the associated DV 200 metric values using a two-tailed pearson correlation test with 95% CI for log-transformed nodule data (n = 27) and spiked samples (n = 5) (Shapiro-Wilk normality test p > 0.05). Absolute and normalised IL27A gene expression values were compared using a two-tailed non-parametric Mann-Whitney test, following assessment with D' Agostino & Pearson normalisty test (p < 0.05, n = 20). Data availability. Nodule RNAseq datasets referred to during the current study are not yet publicly available, as complete data analysis is part of on-going research. Information therein is available from the corrsponding author on reasonable request.