Optimization of a high-throughput whole blood expression profiling methodology and its application to assess the pharmacodynamics of interferon (IFN) beta-1a or polyethylene glycol-conjugated IFN beta-1a in healthy clinical trial subjects

The study of the blood transcriptome in the context of clinical pharmacogenomics has generated much interest in recent years [1, 2]. The cellular and molecular components of peripheral blood exhibit dynamic responsiveness to physiological, environmental, or pathological stimuli and are in contact with nearly every tissue in the body, allowing for assessment of systemic responses to disease or treatment. As such, peripheral blood is a source of clinically accessible diagnostic, prognostic and pharmacodynamic (PD) markers [3, 4]. This idea is supported by a growing body of research that describes the identification of expressed transcripts from human and animal peripheral blood that can function as indicators of disease, as prognostic markers of clinical outcome, of risk of toxicity, and as evidence of a therapy’s pharmacodynamic effects [5–8].

The successful use of gene expression microarrays in basic research studies has spawned great interest in the application of this technology to large clinical pharmacogenomics and population-based studies [9–11]. However, microarray cost, the complexity of sample processing and tracking, and practical limitations in sample throughput have restricted its utilization in clinical investigations [12, 13]. Microarray manufacturers have responded to these needs with the recent development of higher-throughput solutions such as the high-throughput (HT) plate array or “array of arrays” [14]. This platform was made possible through reduction and optimization of probe content and advances in photonics, collectively enabling the miniaturization and assembly of 96 arrays into the spatial arrangement of a conventional microtiter plate. Our laboratory’s internal validation studies have confirmed that data from the HT plate array platform is highly concordant to that of industry standard cartridge arrays [15].

RNA is often amplified using T7 RNA polymerase-driven in vitro transcription (IVT) [16] to produce complementary RNA (cRNA) targets for hybridization to microarrays. However, the high concentration of hemoglobin transcripts in peripheral blood can induce a globin interference effect, effectively reducing a microarray’s detection sensitivity and increasing its signal variability [17]. Although effective methods have been developed to reduce globin interference [18–20], current methods of mitigation also induce variance in microarray results [21].

The challenges associated with utilizing cRNA targets from peripheral blood as probes for microarray investigations have led to the development of alternative methods of amplification and the use of single-stranded complementary DNA (sscDNA) targets from peripheral blood for microarray hybridization [22, 23], effectively improving the sensitivity of microarray hybridizations for detecting peripheral blood transcripts. Results from our laboratory’s internal benchmarking experiments analyzing peripheral blood samples have verified that sscDNA targets improve microarray sensitivity and decrease signal variance as compared with cRNA targets analyzed using globin blocking, degradation, and depletion methods (data not shown).

In the current study, we have systematically optimized sscDNA/HT plate array target mass, hybridization parameters and washing parameters using 2 highly characterized test RNAs with the goal of developing a HT methodology for whole blood transcriptional profiling. Comparative analysis of optimization data against peer-reviewed expression array [15] and quantitative polymerase chain reaction (qPCR) [24] datasets were used to select conditions that improved assay reproducibility and sensitivity. The utility of this new array method (BIIB_HT) was also confirmed through analyses of whole blood samples from a clinical trial comparing pharmacodynamic changes following dosing with either interferon (IFN) beta-1a or polyethylene glycol-conjugated IFN (PEG-IFN) beta-1a.

Peripheral blood transcriptional expression profiling is an attractive technology for large pharmacogenomics studies. However, there have been technical limitations to generating robust transcriptional profiles from this important tissue. Although microarray technologies have been standardized and miniaturized to allow much larger numbers of samples to be processed in parallel than was previously possible from tissues and cell lines, there are few robust methods to utilize these highly parallel profiling technologies for the analysis of large numbers of peripheral blood samples. Therefore, development of new methodologies that enable the reproducible generation of expression profiles from thousands of patient blood samples are of paramount importance to translational research.

We report the development and validation of a highly sensitive and reproducible HT whole blood expression profiling methodology, designated BIIB_HT. This methodology can be utilized in conjunction with clinical studies to identify expressed transcripts that may be useful for elucidating drug or disease mechanisms of action, or that can function as prognostic, diagnostic, or toxicity markers. This method was applied to the analysis of whole blood samples collected from healthy clinical trial subjects before and after treatment with a single dose of either IM IFN beta-1a or SC or IM PEG-IFN beta-1a. Study results demonstrate that PEG-IFN beta-1a induces a more sustained and prolonged pharmacodynamic response than unmodified IFN beta-1a. These results provide validation of the utility of this new methodology and support potential therapeutic benefits of PEG-IFN beta-1a.

The BIIB_HT method is unique in that it has been optimized specifically to provide the most robust detection of transcripts from 96 peripheral blood samples in parallel. It can be used to analyze peripheral blood samples from large clinical studies in order to identify expressed transcripts that may be useful for elucidating disease and therapeutic mechanisms of action, as well as for the identification and validation of prognostic, diagnostic, or toxicity markers. BIIB_HT generates sscDNA targets using Ovation® (NuGEN Technologies, San Carlos, CA) amplification technology and has been optimized to provide the maximum sensitivity and specificity when used in combination with the HTA plate array platform from Affymetrix (Santa Clara, CA). sscDNA targets were selected as the amplification moiety for the development of blood profiling methodologies based on internal benchmarking experiments (data not shown) and published reports [21, 22]. Because there are currently no reports describing the validation of sscDNA targets for use with the HTA array platform, a systematic optimization was required. Specifically, hybridization and washing conditions and mass/array parameters were optimized using 2 RNA samples from the MicroArray Quality Control (MAQC) project to identify conditions yielding maximum detection and lowest variance sscDNA targets [24]. The current results were referenced against native HTA plate array conditions as well as independent qPCR published results.

TMAC is the native HTA plate array hybridization buffer used with cRNA probes. It has been shown to stabilize adenine-thymine (AT) base pairs (bp) and minimize the effect of base composition on oligonucleotide hybridizations of up to 200 bp. The TMAC hybridization buffer effectively equalizes the melting points of different probes therefore allowing probes with different nucleotide compositions to be hybridized under identical conditions [30, 31]. On the other hand, sscDNA hybridizations on glass slide or cartridge arrays typically utilize a 10% DMSO-based buffer. In the presence of DMSO, denatured DNA has been shown to renature with homologous DNA and is not retained by the substrate, thereby reducing the occurrence of background signals [32]. In the current study, the assay results for TMAC-based versus DMSO-based hybridizations were markedly different. Hybridization chemistry (DMSO or TMAC buffer) influenced hybridization quality, as indicated by its ranking as the third most important factor impacting assay results after target type and sample type (Figure 1). In general, use of TMAC-based hybridizations with sscDNA probes produced lower numbers of detectable transcripts and higher background signals across the various masses tested (Figure 2). Given these data, the DMSO-based hybridization buffer was superior to the TMAC-based buffer when sscDNA targets were used in conjunction with the HTA platform.

When assessing possible errors induced as a function of experimental variables, NUSE plots are useful for graphical assessment of array quality. With this method, standard error estimates for each probe set are normalized to a median value of 1 across all arrays. Box plot representation of NUSE values are then drawn for each array and comparative analysis can be conducted for the entire dataset. Arrays or sets of arrays with a larger spread are determined to be of higher variance and are therefore of lower quality. Based on the NUSE plots, use of DMSO buffer with HSH and LSW conditions generated the most reproducible data (Figure 3). Interestingly, mass type and sample type did not strongly influence assay results. These observations may stem from a masking effect due to averaging across sample types. Nevertheless, DMSO was clearly superior to TMAC for use with sscDNA targets on an HT array.

Finally, a PCA in fold change space was used to assess correlations between the variable conditions tested and a “gold standard” qPCR reference dataset. In an effort to normalize all comparisons, qualifiers were selected that were present in the qPCR reference set, changed at least 1.5-fold between HBRR and UHRR datasets, and had a p- value of ≤ 0.0001 in any of the conditions tested. This strategy would allow for penalization of any assay conditions generating a false call as well as for benefiting conditions resulting in a correct call. Results showed a clear difference between DMSO and TMAC, with each buffer condition isolated to distinct clusters. Interestingly, the use of DMSO with LSH, HSW, and 2 μg of sscDNA produced results that clustered most closely with the qPCR reference set (Figure 4). Additionally, these optimized conditions markedly outperformed the standard cRNA/HT array hybridization conditions (black data point), suggesting that the methods reported here represent a significant improvement over the current technology. As with the other analyses that were performed, of all the variables tested, the mass of sscDNA for each array had the smallest effect.

Following optimization of the BIIB_HT technical parameters, we sought to apply this new methodology to the analysis of peripheral blood that was collected as part of a clinical trial evaluating administration of IFN beta-1a or PEG-IFN beta-1a to healthy subjects. Human IFN beta-1a is a first-line therapy for patients with relapsing forms of multiple sclerosis (MS). In multiple clinical trials and long-term observational studies, IFN beta-1a has been shown to reduce the development of MS-associated brain magnetic resonance imaging (MRI) lesions, reduce clinical relapse rates, and slow the advancement of physical disability [33–35]. PEG-IFN beta-1a is being developed with the aim of providing a treatment option that is at least as safe and effective as current first-line therapies, but with the added benefits of less frequent dosing and improved convenience. IFN beta-1a was PEGylated by the attachment of a 20 kDa methoxy-PEG-O-2-methylpropionaldehyde to the α-amino group of the N-terminus of IFN beta-1a, a site that is not critical for binding to the type 1 IFN receptor [36, 37].

Previous pharmacokinetic studies have shown that after a single parenteral injection, PEG-IFN beta1-a was detectable in peripheral blood after 7 days as compared with 2 days with unmodified IFN beta-1a [29]. This increased drug exposure was accompanied by enhanced and sustained expression of the pharmacodynamic IFN biomarkers 2^′,5^′-oligoadenylate synthetase and neopterin. We sought to verify if the transcriptional response would reflect previous observations of a sustained and prolonged pharmacodynamic response to PEG-IFN beta-1a as compared with IFN beta-1a using BIIB_HT assay.

We observed a peak median induction score of 11 at 6 hours post IFN beta1-a dosing. A p-value of 0.007 was calculated by comparison of the IFN induction scores of the 2 groups at 6 hours post dose (Student’s t test). As expected, at 48 hours post-dose the induction score of subjects treated with IFN beta1-1a had declined to 10 while the subjects that received PEG-IFN beta-1a reached a peak Interferon induction score of 11. Again, the p-value that was calculated between the induction scores of the 2 groups was statistically significant (p = 0.00001). These transcriptional data reflect the previously reported translational sustained pharmacodynamic response observed with PEG-IFN beta-1a.

The application of HT microarray technologies to large clinical pharmacogenomics studies represents a unique opportunity to discover prognostic and predictive markers of efficacy and safety on a genome scale. These studies allow a greater understanding of the variable expression of the human transcriptome in response to therapy in a highly controlled setting. A barrier to the execution of these studies is the ability to produce mRNA expression profiles from peripheral blood in a reproducible and robust manner. We believe that the methods presented in this report support the use of HT genome scale expression analysis for biomarker discovery from whole blood samples derived from large clinical trials.