Evaluating miRNA Sequencing and Integrative Proteomics for Identifying Plasma miRNAs as Potential Pharmacodynamic Biomarkers of IFNβ‐1a Biologics

ABSTRACT

We used miRNA sequencing to identify candidate plasma pharmacodynamic (PD) biomarkers of interferon beta‐1a (IFNβ‐1a) biologics and explored miRNA‐mRNA targeted protein relationships and networks to support identified miRNA candidates. Plasma samples from 36 healthy subjects from a placebo‐controlled randomized single‐dose clinical study with IFNβ‐1a and pegIFNβ‐1a were used. Mature miRNAs were measured using an in‐house miRNA‐seq workflow at baseline, at 9 timepoints over 6 days in the IFNβ‐1a group (n = 11 [30 μg]), and at 11 timepoints over 13 days in the pegIFNβ‐1a group (n = 11 [125 μg]) and placebo‐specific groups (n = 6 each). A miRNA was only considered expressed if it had a read count ≥ 10 in more than 50% of the samples across all treatment groups. Linear mixed‐effects models (lmer) regressing miRNA changes with treatment, time, and their interaction were used to identify differentially expressed miRNAs, which were further prioritized based on the magnitude of response and biological relevance using in silico predicted miRNA‐protein targets and regulatory network analyses. Ten and 13 miRNAs were impacted by IFNβ‐1a and pegIFNβ‐1a, respectively (lmer FDR‐corrected p value < 0.1), compared to placebo, of which miR‐223‐3p and miR‐21‐5p were prioritized as candidate PD biomarkers for both products. Systems‐level analysis of integrated proteomics data highlighted miRNAs' protein targets for both miR‐223‐3p and miR‐21‐5p that were linked to previously reported top proteomic response proteins and predicted regulatory networks including the IFN beta node. Using miRNA sequencing and linking candidates to predicted target proteins and regulatory networks, results suggest miR‐223‐3p and miR‐21‐5p as potential PD biomarkers of IFNβ‐1a biologics for further investigation.

Study Highlights

• What is the current knowledge on the topic?

  • ○
    PD biomarkers are molecular indicators that can measure a drug‘s effect against a certain target/pathway and/or can inform on the functional consequences of the drug‐target interaction. Interferon beta‐1a (IFNβ‐1a) biologics used to treat multiple sclerosis have complex mechanisms of action and limited well‐characterized PD biomarkers. The availability of omics technologies, including next‐generation sequencing (NGS), has offered great potential for advances in biomarker discovery for drug development.
• What question did this study address?

  • ○
    This study evaluated miRNA‐sequencing for the identification of potential plasma IFNβ‐1a PD biomarkers and explored miRNA‐mRNA target‐encoding protein relationships in the same samples to support candidate miRNAs as new PD biomarkers of IFNβ‐1a products.
• What does this study add to our knowledge?

  • ○
    Utilizing a sensitive, accurate, reproducible automated fit‐for‐purpose miRNA sequencing workflow, we identified miR‐223‐3p and miR‐21‐5p as potential candidate PD miRNAs for IFNβ‐1a biologics. These findings were further supported by linking changes in miRNA levels with predicted protein targets from proteomics data from the same samples. The complexity and potential intrinsic platform biases of the high‐throughput miRNA sequencing from plasma may limit the generalizability of present results and the utility of small RNA transcriptomics for reproducible detection of plasma PD miRNA biomarkers.
• How might this change clinical pharmacology or translational science?

  • ○
    If well‐validated in future studies, miR‐223‐3p and miR‐21‐5p may have utility as PD biomarkers for IFNβ‐1a biologics, in early drug development studies, and in assessing the effectiveness of these drugs in MS patients. These results also provide further understanding of the mechanism of action of these drugs and the role of miRNA‐regulated protein changes.

1. Introduction

Omics technologies offer great potential for biomarker discovery in drug development, enabling the measurement of genome, transcriptome, proteome, and metabolome features [1, 2, 3]. Blood plasma contains miRNAs, which are small non‐coding RNAs of approximately 21–23 nucleotides (nts), involved in the regulation of several physiological processes [4]. In circulation, tissue‐specific and cell‐free miRNAs can act as cell‐to‐cell messengers by targeting specific cells, where they can regulate mRNA and protein expression [5, 6]. Circulating miRNA expression changes have been linked to disease development and drug response [6] with plasma miRNAs showing potential as diagnostic, prognostic, and predictive biomarkers [4]. Global miRNA profiling can measure expression levels before and after pharmaceutical interventions [7], therefore, miRNA‐sequencing of plasma may identify candidates for drug response or pharmacodynamic (PD) biomarkers.

Interferon β‐1a (IFNβ‐1a) and pegylated (peg)‐IFNβ‐1a biologics are approved for the treatment of multiple sclerosis (MS), and the mechanisms of action of these drugs in the disease process are complex [8]. Additionally, there are only a few well‐characterized biomarkers for IFNβ treatment response, including serum neopterin, blood myxovirus resistance protein 1 (Mx1), β2‐microglobulin, and 2′,5′‐oligoadenylate synthase activity [9, 10, 11, 12, 13, 14, 15], as well as some recently identified proteomic‐derived plasma PD biomarkers including I‐TAC and IP‐10 [15]. However, there is no data on plasma miRNAs as potential PD biomarkers of IFNβ‐1a products or biologics in general that may be useful for biosimilarity assessment or development.

In this study, we evaluated miRNA sequencing for the identification of plasma miRNA PD biomarkers of IFNβ‐1a biologics using samples from a clinical trial (NCT04183491) in healthy participants. Herein, we report on the development and evaluation of a miRNA sequencing workflow for plasma miRNA profiling and analysis, together with the integration of previously generated plasma proteomics data to support the identified miRNA candidate PD biomarkers.

2. Methods

2.1. Study Design

Plasma samples collected from the high‐dose groups (i.e., the approved therapeutic doses) from NCT04183491 were used for PD miRNA biomarker identification (Appendix S1: Supplementary Methods & Results). NCT04183491 was a randomized, double‐blinded, placebo‐controlled, single‐dose, parallel‐arm study. Participant recruitment, inclusion, and exclusion criteria are described in detail at https://clinicaltrials.gov/ct2/show/NCT04183491?term=IFNB+1a&draw=2&rank=7.

Thirty‐six healthy participants were randomized to IFNβ‐1a [30 μg] or pegIFNβ‐1a [125 μg] drug group or placebo (5% dextrose), administered as a single dose at time zero on Day 1 [15]. For plasma processing, blood samples were obtained at time zero (baseline), at 9 timepoints over 6 days following administration of IFNβ‐1a, and at 11 timepoints over 13 days following administration of placebo and pegIFNβ‐1a. The study was approved (Pro00040449, August 29, 2020) by the institutional review board (Advarra [https://www.advarra.com]) and was conducted in accordance with the 7th revision of the Declaration of Helsinki 2013.

2.2. RNA Extraction and miRNA Sequencing

Next‐generation sequencing (NGS) allows global, unbiased profiling of circulating miRNAs in biospecimens, but requires careful control of potential biases during sample preparation, sequencing, and analysis [16, 17]. To develop a robust workflow for quantification of plasma miRNAs, we assessed and identified an optimal: library preparation kit for miRNAs/small RNAs generation; bioinformatics pipeline; and normalization method using a pool of synthetic miRNA reference (miRXplore, Myltenyl Biotec) for miRNA sequencing (described in detail in Appendix S1: Supplementary Methods & Results, Sections 1–4). Based on these evaluations, a sensitive, accurate, and fit‐for‐purpose workflow using the miRNeasy Serum/Plasma Kit (Qiagen LLC) and the QIAcube RNA purification instrument (Qiagen LLC), automated liquid handling of the QIAseq miRNA Library Kit (Qiagen LLC), CLC Genomic Workbench pipeline (Qiagen LLC), and DESeq2 normalization method [18], together with short read sequencing on the Illumina NextSeq 2000 was developed and used for miRNA sequencing of plasma (Appendix S1: Supplementary Methods & Results, Section 1A, Tables 1–4).

Clinical trial plasma samples selected for sequencing included baseline/pre‐dose (time 0), 0.125, 0.25, 0.33, 0.67, 1.33, 2, 4, 5, 6, 9, and 13 days (with Day 9 and 13 samples only obtained for pegIFNβ‐1a and placebo groups). A total of 395 plasma RNA samples from 36 participants were sequenced together with plasma controls. Library preparations were carried out in batches of 96 samples at a time using an automated workflow on the Biomek i7 Automated Liquid Handler (Beckman). Following quality control (QC) assessment, libraries were sequenced using a read length of 75 base pairs (bp) on the NextSeq 2000 System (Illumina) according to the manufacturer's instructions. Additional details of RNA extraction, library preparation, and sequencing are described in Appendix S1: Supplementary Methods & Results, Section 1B.

2.3. Bioinformatic Analyses

Raw sequence reads were quality control checked using the FASTQC package [v0.11.2] and FASTQ files were generated using the NextSeq 2000 DRAGEN BCL convert tool v3. The QIAseq miRNA Quantification workflow in CLCBio Genomics Workbench (v21.0.1) was then used to process the FASTQ files, remove adapters, and align sequences to miRbase version 22 [19, 20]. Mature miRNA reads measured at baseline and all other timepoints were normalized using the R package DESeq2 v1.34.0 [18], and were filtered for ≥ 10 read counts in ≥ 50% of samples as previously described [21]. Normalized counts were then transformed using the DESeq2 Variance Stabilizing Transformation (VST) function [18], using a set of 8 intrinsic (in data) miRNAs with the least variance before further data analysis.

2.4. Statistical Analyses

Descriptive summaries of demographics and baseline characteristics for the three treatment groups have been previously reported [15]. Differentially expressed miRNAs among the three treatment groups at baseline were identified using the Kruskal–Wallis test. The effect of age, sex, race, and body mass index (BMI) on baseline log2 transformed normalized miRNA counts was tested using the Wilcoxon test. The effect of treatment on each miRNA at each timepoint was assessed as fold change (FC) in log2 transformed normalized counts from baseline (FC = log2 transformed normalized counts at a timepoint/log2 transformed normalized counts at baseline). Differentially expressed miRNAs by treatment were identified using linear mixed‐effects (lmer) models, regressing the log2 normalized FC values with treatment group, timepoint, and the interaction of treatment group and timepoint. In these models, timepoint was considered fixed effects, whereas study participants were considered random effects. This analysis was adjusted for categorical age, sex, BMI, race, and batch effects. For each variable in the model, an F‐statistic and a p value were generated using type‐3 ANOVA. False discovery rate (FDR) was used to account for multiple comparisons. MiRNAs with an F‐test FDR‐corrected p < 0.1 for drug group: timepoint interaction were considered differentially expressed. For each differentially expressed miRNA, the peak response time was reported as the timepoint with the highest observed median FC in log2 transformed normalized counts from baseline in the drug‐treated group. We also calculated the fold difference for each differentially expressed miRNA as the difference in log2 FC from baseline between the drug‐treated group and placebo group at the peak response time.

Differentially expressed miRNAs with at least a 20% fold difference and a t‐test p < 0.05 were tested further for their baseline‐adjusted area under the effect curve (AUEC). We calculated the AUEC for log2 normalized counts from baseline (time 0) to the last timepoint (6 days for IFNβ‐1a and 13 days for pegIFNβ‐1a), or over a shorter time interval based on the miRNA's return to baseline using the PKNCA R package. A t‐test was performed to assess if the AUEC for the miRNA was significantly different between the drug‐treated group and the placebo group. A Bonferroni corrected p value was used to account for multiple comparisons. All statistical analyses were conducted in R, version 4.1.2 (R Core Team, Vienna, Austria).

2.5. Correlation, Functional Annotation, and Upstream Regulatory Analysis of Predicted miRNA‐Protein Targets

To provide additional empirical support and biological insight into observed differentially expressed miRNAs (FDR‐corrected p < 0.1) impacted by IFNβ‐1a biologics, we used previously published proteomics data from the same samples [15] to examine in silico predicted miRNA‐protein relations and target expression changes at an individual and systems analysis level. QIAGEN Ingenuity Pathway Analysis (IPA) was used to identify (predicted and experimental) miRNA‐mRNA‐target protein relationships and network regulators [22].

First, the encoding mRNA targets (and thus proteins) of each differentially expressed miRNA were identified using the IPA miRNA Target Filter function. Correlations between each of the prioritized miRNAs and target proteins (FDR‐corrected p < 0.1) [15] were then examined to assess coherent and noncoherent response signatures for miRNA‐protein target relationships. This involved exploring specific cases where miRNAs and protein target expression changed in the expected directions (coherent) or in opposite directions (noncoherent) [23], as an indicator of potential for a drug‐mediated miRNA‐regulatory mechanism influencing protein levels. Median fold‐change (FC) response from baseline at each timepoint for each miRNA‐protein target identified (FDR‐corrected p < 0.1) from [15] were then used to identify upstream regulatory networks as previously described [15]. For activation/inhibition of regulatory networks, we used a Bonferroni‐corrected overlap p value < 0.05 and a z‐score ≥ 2 to indicate strong activation or z ≤ −2 to indicate strong inhibition.

3. Results

3.1. Participant Characteristics

As previously reported, a total of 36 subjects (12 in each group: high‐dose IFNβ‐1a, high‐dose pegIFNβ‐1a, and placebo) were enrolled in the study [15]. One participant in the IFNβ‐1a group and one in the pegIFNβ‐1a group discontinued the study. No significant differences in demographics or baseline characteristics between the two product and placebo groups were observed, and no participant tested positive for COVID‐19 at any time over the study course [15]. These published data are presented in Table S1 for reference.

3.2. miRNA‐Seq Analysis

3.2.1. QC Analysis

Variability of miRNA measurements was shown to be < 15% for 90% percentile CV of all expressed miRNAs for intra‐ and inter‐plate variance and < 23% for 90% percentile CV for expressed miRNAs in the universal plate control samples (Appendix S1: Supplementary Methods & Results, Section 1C). Seventeen test samples and six convalescent COVID‐19 control samples were shown to separate from the other test samples using PCA. Because we could not identify a reason for the observed outliers, these test samples were excluded from further analyses. In addition, PCA identified a batch effect between two runs, which was adjusted for in the statistical model (Appendix S1: Supplementary Methods & Results, Section 1C). The number, diversity, and qualitative longitudinal stability of miRNAs were also shown to be constant across all test samples over time, albeit slightly lower in number and reflecting previously observed prevalent miRNAs of plasma (Appendix S1: Supplementary Methods & Results, Section 2).

3.2.2. Analysis of miRNAs Between Treatment and Placebo Groups at Baseline

After applying a read cutoff of ≥ 10 read counts across ≥ 50% of samples [21], 108 mature miRNAs were identified for further analysis. Based on an FDR‐corrected p < 0.05, no significant differences in baseline miRNA counts were observed by treatment group or by baseline characteristics, except for sex (n = 6 miRNAs: miR‐10b‐5p, miR‐10a‐5p, miR‐146a‐5p, miR‐7‐5p, miR‐21‐5p, and miR‐23a‐3p).

3.2.3. Identification of Candidate miRNA PD Biomarkers of IFNβ‐1a and pegIFNβ‐1a

Based on a lmer FDR‐corrected p < 0.1, 11 and 13 miRNAs were impacted by IFNβ‐1a and pegIFNβ‐1a, respectively (Tables S2 and S3). Of the 11 IFNβ‐1a miRNAs, 7 were upregulated, 3 were downregulated, and 1 was biphasic (i.e., showing at least a 10% up‐ and 10% downregulation from baseline at 2 independent timepoints). Of the 13 pegIFNβ‐1a miRNAs, 6 were upregulated, 5 were downregulated, and 2 were biphasic. Of the 11 IFNβ‐1a miRNAs, 5 (miR‐223‐3p, miR‐150‐5p, miR‐125a‐5p, miR‐423‐5p, and miR‐29b‐3p) characterized a ≥ 20%‐fold difference and t‐test p < 0.05 (Table S4). miR‐223‐3p showed the strongest magnitude in fold difference (87%), whereas miR‐125a‐5p showed the smallest magnitude (20%) between the IFNβ‐1a drug group and the placebo group at maximum peak response. Of the 13 pegIFNβ‐1a miRNAs, 2 characterized a significant ≥ 20%‐fold difference: miR‐21‐5p (48%) and miR‐223‐3p (31%) (Table S6). The temporal response profiles for each of these miRNA candidates by product are presented in Figure 1 and Tables S5 and S7. Changes in miR‐223‐3p were common to both products based on the above criteria.

FIGURE 1.

Temporal response profiles for the two candidate miRNAs impacted by both IFNβ‐1a and pegIFNβ‐1a. (A) Response profile for miR‐223‐3p in response to IFNβ‐1a vs. IFNβ‐1a specific placebo groups (left side) and in response to pegIFNβ‐1a vs. pegIFNβ‐1a specific placebo groups (right side). MiR‐223‐3p was upregulated in response to IFNβ‐1a, with a maximum peak response of 0.33 days. It was biphasic in response to pegIFNβ‐1a, with a maximum peak response for downregulation at 4 days. (B) Response profile of miR‐21‐5p in response to IFNβ‐1a versus IFNβ‐1a specific placebo groups (left side) and in response to pegIFNβ‐1a versus pegIFNβ‐1a specific placebo groups (right side). MiR‐21‐5p showed an upregulation in response to IFNβ‐1a and pegIFNβ‐1a, with a maximum peak response at 4 and 2 days, respectively. The blue bars represent the placebo group, whereas the orange bars represent the IFNβ‐1a drug group. Responses are presented on the y‐axis as log2 (fold change from baseline) over time (days) on the x‐axis. FC, fold change.

3.3. Prioritization of miRNA PD Biomarker Candidates

The temporal response profile of miR‐223‐3p was classified as upregulated for IFNβ‐1a (maximum peak response at 0.33 days), and biphasic for pegIFNβ‐1a (maximum downregulation/peak response at 4 days) (Figure 1A). In response to both products, miR‐223‐3p characterized early upregulation at 0.33 days, followed by a downregulation at 1.33 days with a return to baseline or near baseline by the end of the study. MiR‐223‐3p showed a significant AUEC over 0.67 days for IFNβ‐1a when the miRNA returned to baseline (p = 0.015), and a nominally significant AUEC for pegIFNβ‐1a over 13 days (p = 0.04). While miR‐223‐3p showed an upregulated response to IFNβ‐1a over time, the AUEC was not significant over 6 days (Figure 2A and Tables S8 and S9).

FIGURE 2.

Baseline‐adjusted area under the effect curve (AUEC) results for candidate miRNAs impacted by both IFNβ‐1a and pegIFNβ‐1a. Boxplots show the baseline‐adjusted AUEC for (A) miR‐223‐3p and (B) miR‐21‐5p. The baseline‐subtracted AUEC is shown for each of the IFNβ‐1a products and their placebo‐specific groups, with response represented as log2 normalized counts. The AUEC was calculated for log2 normalized counts from time 0 (baseline) to the last timepoint for miR‐21‐5p. Since miR‐223‐3p had a biphasic response, the AUEC was calculated over a shorter time interval (over 0.67 days). AUEC, area under the effect curve.

The temporal response profile for miR‐21‐5p showed an up‐regulation for IFNβ‐1a and pegIFNβ‐1a with a maximum peak response at 4 and 2 days, respectively. In response to IFNβ‐1a, miR‐21‐5p characterized a 27% (t‐test p = 0.066) and 48% (t‐test p = 0.009) upregulated fold difference in response to IFNβ‐1a and pegIFNβ‐1a, respectively. In response to both products, miR‐21‐5p characterized an early downregulation, followed by an upregulation at 1.33 days through the end of the study (Figure 1B). MiR‐21‐5p showed a nominally significant AUEC over 13 days between the pegIFNβ‐1a group and the placebo group (p = 0.026) (Figure 2B and Table S9).

The IFNβ‐1a candidate miR‐125a‐5p showed a downregulation with a maximum peak response at 6 days, with a nominally significant AUEC (p = 0.039) over 6 days (Table S8). The IFNβ‐1a candidate miR‐29b‐3p showed an upregulation with a maximum peak response at 5 days (Tables S4 and S5). However, miR‐29b‐3p did not characterize a significant AUEC over 6 days (Table S8). A downregulation of miR‐150‐5p with a maximum peak response at 0.33 days was observed for IFNβ‐1a (Tables S4 and S5). MiR‐150‐5p was nominally differentially expressed (FDR‐corrected p = 0.19) in response to pegIFNβ‐1a and exhibited a 28%‐fold difference between pegIFNβ‐1a and placebo at 0.33 days (t‐test p = 0.02) (data not shown). However, we observed a placebo effect for miR‐150‐5p (Appendix S1: Supplementary Methods & Results, Section 3) and, as a result, this miRNA was dropped from further analyses. Lastly, the AUEC was not estimated for the IFNβ‐1a candidate miR‐423‐5p because this miRNA did not show a consistent response pattern over time (Table S5).

3.4. Correlation, Functional Annotation, and Upstream Regulatory Analysis of miRNA‐Protein Expression Levels Changes

We identified 1347 and 1245 mRNA target‐proteins, collectively regulated by the 10 (excluding miR‐150‐5p) and 13 differentially expressed IFNβ‐1a and pegIFNβ‐1a miRNAs, respectively. Of the 1347 and 1245 proteins, 96/1347 and 147/1245 were previously identified as differentially expressed (FDR‐corrected p < 0.1) in plasma from the same participants following treatment with IFNβ‐1a and pegIFNβ‐1a, respectively [15]. These proteins are described in Tables S10 and S11.

The top common PD biomarker miRNA candidate miR‐223‐3p regulated 7 proteins, including Mx1, lipase family member N (LIPN), interleukin 6 (IL6), ras homolog family member B (RHOB), prolactin (PRL), midkine (MDK), and HLF transcription factor (HLF) (Tables S10 and S11). Figures 3 and 4 show the coherent direction of effect between miR‐223‐3p and Mx1 and LIPN regulation in plasma, respectively, in IFNβ‐1a and pegIFNβ‐1a groups compared to placebo. Both Mx1 and LIPN previously survived Bonferroni‐corrected significance as proteomic PD biomarker candidates [15]. Table S12 shows the response profiles for the remaining miRNA‐mRNA target‐encoding protein relationships for miR‐223‐3p in plasma. The other top candidate PD biomarker miRNA, miR‐21‐5p, regulates 5 proteins: oxidized low‐density lipoprotein receptor 1 (OLR1), interleukin 6 receptor (IL6R), TPM metallopeptidase inhibitor 3 (TIMP3), Fas cell surface death receptor (FAS), and neurotrophin 3 (NTF3) (Tables S10 and S11). Figure 4 shows the coherent temporal response relation between miR‐21‐5p and the significant proteomic‐based PD biomarker hit OLR1 [15]. The response profiles of the remaining miRNA‐mRNA target‐encoding protein relationships for miR‐21‐5p are described in Table S13.

FIGURE 3.

Temporal response patterns of miR‐223‐3p and Mx1 in IFNβ‐1a‐ and pegIFNβ‐1a‐treated and placebo groups. Response patterns in the IFNβ‐1a‐ and pegIFNβ‐1a‐treated groups characterized a coherent change in the expected direction of expression (i.e., miRNA downregulation and protein upregulation with time) between miR‐223‐3p (IFNβ‐1a: p = 6.55E‐23, FDR‐corrected p = 7.08E‐21; pegIFNβ‐1a: p = 3.59E‐04, FDR‐corrected p = 1.70E‐02) and Mx1 (IFNβ‐1a: p = 9.35E‐15, FDR‐corrected p = 1.12E‐12; pegIFNβ‐1a: P = 7.52E‐28, FDR‐corrected p = 2.49E‐25). In contrast, noncoherent response signatures for the miRNA‐protein target relationship were observed in the placebo groups. The orange bars represent the miR‐223‐3p, whereas the blue bars represent Mx1. Responses are presented on the y‐axis as log2 (fold change from baseline) over time (days) on the x‐axis. FC, fold change; FDR, false discovery rate.

FIGURE 4.

Temporal response patterns of the prioritized PD miRNA candidates and their targets in IFNβ‐1a‐ and pegIFNβ‐1a‐treated and placebo groups. (A) Shows the temporal response patterns of miR‐223‐3p and LIPN in the IFNβ‐1a‐ and pegIFNβ‐1a‐treated and placebo groups. The patterns in the drug groups show an opposite direction between the miR‐223‐3p (IFNβ‐1a: p = 6.55E‐23, FDR‐corrected p = 7.08E‐21; pegIFNβ‐1a: p = 3.59E‐04, FDR‐corrected p = 1.70E‐02) and LIPN (IFNβ‐1a: p = 1.46E‐11, FDR‐corrected p = 1.17E‐09; pegIFNβ‐1a: p = 4.41E‐07, FDR‐corrected p = 7.68E‐06). (B) Shows the temporal response patterns of miR‐21‐5p and OLR1 in the IFNβ‐1a‐ and pegIFNβ‐1a‐treated and placebo groups. The patterns in the drug groups show an opposite direction between the miR‐21‐5p (IFNβ‐1a: p = 1.08E‐04, FDR‐corrected p = 3.91E‐03; pegIFNβ‐1a: p = 4.59E‐03, FDR‐corrected p = 4.96E‐02) and OLR1 (IFNβ‐1a: p = 1.03E‐12, FDR‐corrected p = 9.01E‐11; pegIFNβ‐1a: p = 9.56E‐15, FDR‐corrected p = 5.71E‐13). The orange bars represent the miRNA, whereas the blue bars represent the protein. Responses are presented on the y‐axis as log2 (fold change from baseline) over time (days) on the x‐axis. FC, fold change; FDR, false discovery rate; PD, pharmacodynamic.

Upstream regulatory analysis of the 96 and 147 miRNA‐regulated proteins from [15] identified 68 common upstream regulatory (UR) networks. Appendix S1: Supplementary Methods & Results Section 4 details the top 35 common UR networks. Of these, 31 were reported to be activated in response to both products using the previously published proteomics data [15]. These networks included IFN Beta node, interferon gamma (IFNG), interferon alpha, interleukin 1β (IL1B), tumor necrosis factor (TNF), and signal transducer and activator of transcription 3 (STAT3) (also listed in Tables S14 and S15). The IFN Beta node (including IFNβ and IL‐6 as the upstream regulator combined) was significantly activated in response to IFNβ‐1a at 0.33 days through day 5, and enriched with targets (IL6, galectin (LGALS9), and Mx1), predominantly regulated by miR‐223‐3p (Figure 5A,B). In response to pegIFNβ‐1a, the IFN Beta node was activated at 0.67 days through day 13 post‐dose and was enriched with proteins (CD247, interferon induced with helicase C domain 1 (IFIH1), IL6, and Mx1) also regulated by miR‐223‐3p (Figure 5C,D). Lastly, we identified a miR‐21‐5p network that was inhibited in response to IFNβ‐1a and pegIFNβ‐1a and associated with proteins predominantly regulated by miR‐21‐5p (Tables S14 and S15).

FIGURE 5.

Prediction of miRNA‐mediated regulation of the IFN Beta node and regulators over time in response to IFNβ‐1a and pegIFNβ‐1a. (A) Longitudinal activation of the IFN Beta node (i.e., role of IFNβ and IL‐6 together) or regulatory network over time using miRNA‐regulated protein targets and their response to IFNβ‐1a treatment over time. Maximum activation (yellow colored star symbol) by IFNβ‐1a was observed at 0.33, 1.33, 4, and 5 days, respectively. (B) Graphical illustration of the IFN Beta node or in this instance the IFNβ‐1a product (plus IL‐6) as the master regulator and 18 other regulators with specific activation states as predicted by the protein targets of one or more of the 10 miRNAs differentially expressed by IFNβ‐1a treatment. For the 10 IFNβ‐1a impacted miRNAs: miR‐21‐5p was predicted to regulate the expression of TNF, IL1B, STAT3, and NFKB (complex); miR‐223‐3p regulates the expression of STAT3, RELA, and IL1B; miR‐101‐3p regulates the expression of EZH2, HIF1A, and IL1B; miR.27a.3p regulates the expression of IFNG, JUN, and MYC; and miR.423.5p regulates the expression of IRF1, CEBPB, and NATC2. (C) Longitudinal activation of the IFNβ group node and regulatory network over time using miRNA‐regulated protein targets and their response to pegIFNβ‐1a treatment with significant and strong activation (z‐score > 2) occurring from 0.67 days (16 h) through 13 days. The maximum activation (yellow colored star symbol) by pegIFNβ‐1a was observed at 13 days. (D) Graphical illustration of the IFN Beta node or in this instance pegIFNβ‐1a product (plus IL‐6) as the master regulator and 22 other regulators with specific activation states as predicted by the protein targets of one or more of the 13 miRNAs differentially expressed by pegIFNβ‐1a treatment. Of the 13 pegIFNβ‐1a impacted miRNAs, miR‐223‐5p was predicted to regulate the expression of TNFSF10 and STAT1, while miR‐223‐3p regulated the expression of STAT3 and STAT1. Figures were generated using QIAGEN Ingenuity Pathway Analysis (IPA). CEBPB, CCAAT enhancer binding protein beta; CITED2, Cbp/p300 interacting transactivator with Glu/Asp rich carboxy‐terminal domain 2; CTNNB1, catenin beta 1; EZH2, enhancer of zeste 2 polycomb repressive complex 2 subunit; FOS, Fos proto‐oncogene; FOXO1, forkhead box O1; HIF1A, hypoxia inducible factor 1 subunit alpha; IFN Beta, interferon‐beta 1; IFNG, interferon gamma; IL1B, interleukin 1 beta; IRF1, interferon regulatory factor 1; IRF9, interferon regulatory factor 9; JUN, jun proto‐oncogene; MAPK1, mitogen‐activated protein kinase 1; MYC, MYC proto‐oncogene; NFATC2, nuclear factor of activated T cells 2; NFKB, nuclear factor kappa B; NFKB1, nuclear factor kappa B subunit 1; NFKBIA, NFKB inhibitor alpha; NR3C1, nuclear receptor subfamily 3 group C member 1; NR4A1, nuclear receptor subfamily 4 group A member 1; PPARG, peroxisome proliferator activated receptor gamma; RELA, RELA proto‐oncogene; STAT1, signal transducer and activator of transcription 1; STAT2, signal transducer and activator of transcription 2; STAT3, signal transducer and activator of transcription 3; TNF, tumor necrosis factor; TNFSF10, TNF superfamily member 10; TP53, tumor protein p53; VDR, vitamin D receptor.

4. Discussion

Omics technologies have the potential to facilitate and accelerate drug discovery through the identification of new biomarkers for drug response prediction [24]. PD biomarkers are molecular indicators, which can measure a drug's effect against a certain target/pathway and/or can inform on the functional consequences of the drug‐target interaction [25, 26]. Although challenges remain with respect to using PD biomarkers for biosimilarity assessment due to a lack of available and well‐characterized PD biomarkers for many biologic drugs [27], we evaluated miRNA sequencing for the identification of circulating PD biomarkers of IFNβ‐1a and pegIFNβ‐1a products. We further leveraged and linked proteomics data from the same plasma samples and a systems‐level analysis of predicted miRNA‐protein targets to support candidate miRNA PD biomarkers identified.

NGS workflows, including miRNA sequencing, can involve biases that affect miRNA detection and downstream analyses [28, 29, 30, 31]. While a detailed description of these biases [17, 30] is beyond the scope of this discussion, we optimized these parameters to develop a fit‐for‐purpose workflow. Using the QIAseq miRNA library kit, CLC Genomics Workbench pipeline, and DESeq2 normalization method [18], we achieved high accuracy (98%) and validity metrics for a synthetic human miRNA reference independent of whether liquid handling was carried out manually or robotically and/or sequenced on different Illumina systems. While the diversity of plasma miRNAs is reported to be low ~10%–35% [17, 32], compared to tissue [17], these validity metrics were reproducible for detection of miRNAs in plasma, albeit we observed a smaller‐than‐expected number of mature plasma miRNAs using our automated workflow. As discussed in detail in Appendix S1: Supplementary Methods and Results, Section 3, this may reflect one or more factors, including differences in plasma processing and RNA extraction, low template input and miRNA diversity, sequencing instrument bias, and biological differences specific to the study population. Importantly, the overall pattern of the most abundant plasma miRNAs observed were as expected indicating an unbiased detection by our method.

We identified several differentially expressed miRNAs in response to IFNβ‐1a and pegIFNβ‐1a, of which miR‐223‐3p and miR‐21‐5p were prioritized as potential candidate PD biomarkers. Both candidate miRNAs were associated with small but biologically relevant fold changes (> 20%) in response to both products at maximum peak time, but also high variability in response as seen from CV% for AUEC. MiR‐223‐3p was associated with an upregulated response to IFNβ‐1a or an overall biphasic response to pegIFNβ‐1a. The different responses of miR‐223‐3p to each of the IFNβ‐1a biologics may reflect the different dosing, shorter half‐life, and increased rate of clearance of IFNβ‐1a compared to pegIFNβ‐1a [15, 33, 34]. MiR‐21‐5p was also upregulated in response to both products but characterized an earlier and smaller magnitude of response to IFNβ‐1a compared to pegIFNβ‐1a. The duration of response of miR‐21‐5p to pegIFNβ‐1a was also much longer. However, the maximum peak time of miR‐21‐5p in response to IFNβ‐1a and pegIFNβ‐1a was at 4 and 2 days, respectively, which does not correlate well with the half‐lives and rate of clearance of both products. This may reflect variable miRNA release from the tissue(s) of origin in response to the drug.

In silico analyses of miRNA‐mRNA targeted protein relationships identified 18 miR‐223‐3p regulated proteins including Mx1, LIPN, MDK, and heat shock protein 90 beta family member 1 (HSP90B1) which were previously identified as top proteomic PD biomarkers for IFNβ‐1a biologics in the same plasma samples [15]. This concordance between miRNA and proteomic biomarkers strengthens the biological relevance of miR‐223‐3p as a potential PD biomarker. IFNβ‐1a and pegIFNβ‐1a products are approved for treating relapsing MS. However, their mechanisms of action are complex, multifactorial, and incompletely understood regarding how they elicit therapeutic responses in MS [8, 35]. Currently, only a few partially characterized PD biomarkers exist for IFNβ‐1a treatment response, including neopterin, β2‐microglobulin, and 2′,5′‐oligoadenylate synthase activity. Data on their dose–response relationships and correlations with clinical outcomes in MS also remain limited [34]. In MS, Mx1 serves as a biomarker reflecting the presence and activity of IFNβ therapy. IFNβ treatment is associated with various immunomodulatory effects, including restored regulatory T‐cell (Treg) function and reduced pro‐inflammatory cytokine production [8, 36]. Thus, Mx1, a predicted target of miR‐223‐3p as demonstrated in this study, functions as a direct downstream effector of type I interferon signaling and can represent a reliable indicator of IFNβ biological response [15, 34].

Moreover, the potential clinical relevance of miR‐223‐3p to MS has also been reported in the literature. Upregulation of miR‐223‐3p in blood cells and CD4(+)CD25(+high) T cells has been linked to relapsing–remitting MS (RRMS) pathophysiology with and without treatment [35]. Additionally, miR‐223‐3p has been found to be upregulated in active MS lesions compared to normal brain tissues [36] and serum miR‐223‐3p was reported to negatively correlate with changes in magnetic resonance imaging in RRMS patients, suggesting a potential protective and/or prognostic effect [37, 38]. These findings collectively suggest that miR‐223‐3p may serve as a potential biomarker of MS and disease progression [35, 36, 37, 38], although the mechanistic relationship between disease‐associated upregulation and therapeutic response requires further investigation.

MiR‐21‐5p regulates several pro‐inflammatory and cell signaling target proteins, including TNF [37, 38], FAS [39], IL6R [39], and transforming growth factor beta receptor 2 (TGFBR2) [40]. Additionally, predicted miR‐21‐5p regulated proteins were OLR1, C‐C motif chemokine ligand 22 (CCL22), and transforming growth factor beta induced (TGFBI). Notably, we previously demonstrated that these same proteins (OLR1, FAS, IL6R, TGFBR2, CCL22, and TGFBI) were significantly differentially expressed in response to both IFNβ‐1a biologics [15] providing mechanistic support for miR‐21‐5p as a potential PD biomarker. Beyond its pharmacodynamic relevance, miR‐21‐5p has reported associations with MS pathophysiology, though its role appears to be significantly influenced by treatment status and differs markedly between healthy individuals and MS patients. Several studies have suggested that miR‐21 is involved in IFN production [41, 42, 43] and is upregulated by IFNβ treatment in vitro [44]. Importantly, miR‐21‐5p expression patterns differ between healthy controls and MS patients. It is downregulated in CD4+ T cells in untreated RRMS patients compared to healthy controls [45], but upregulated in MS lesions and peripheral mononuclear cells [46, 47, 48]. The altered expression patterns of this miRNA in MS suggest that MS patients may exhibit different miR‐21‐5p responses to IFNβ treatment compared to the upregulation we observed in healthy subjects. Future studies in MS patients will be required to determine whether IFNβ treatment‐induced changes in miR‐223‐3p and miR‐21‐5p in plasma may indicate therapeutic response in the disease context. If confirmed, these miRNA biomarkers could enable real‐time monitoring of treatment response.

A single miRNA can regulate multiple mRNA‐protein relations, and a single mRNA can be targeted by multiple miRNAs [23]. Upstream regulatory (UR) analyses of all miRNA protein targets indicated activation and inhibition of specific pro‐inflammatory and immune signaling networks downstream of IFNβ‐1a products. While networks were predominantly driven by miRNA target proteins previously reported by us as top proteomics PD biomarkers [15], they also included other miRNA specific target proteins. Activation of IFNG, IL1B, interleukin 2 (IL2), STAT3, and interferon alpha (IFNA) networks was predominantly linked to proteins regulated by the candidate PD biomarker miRNAs, miR‐223‐3p, and miR‐21‐5p. In turn, activation of the IFN Beta node network in response to both IFNβ‐1a biologics was linked to miR‐223‐3p regulated targets including Mx1, IL6, CDC47, and IFIH1. Additionally, plasma miR‐223‐3p showed a coherent regulatory response with the plasma targets Mx1 [12, 14, 15] and LIPN, two top proteomic PD biomarkers that survived Bonferroni correction [15]. The relations between miR‐223‐3p and its remaining targets, however, were non‐coherent in plasma. Similarly, miR‐21‐5p also showed a coherent regulatory response with the plasma target OLR1, another top proteomic PD biomarker [15]. The regulatory relations between miR‐21‐5p and its other target proteins were non‐coherent in plasma. The observed noncoherence between differentially expressed miRNAs and target protein(s) in plasma might reflect differences between observed miRNA expression profiles in circulation as investigated here and the expected effects based on gene regulation in cell compartments or tissues of origin. Interestingly, even though the AUEC for miR‐21‐5p was not significant between IFNβ‐1a and placebo, inhibition of an IPA‐specific miR‐21‐5p regulatory network was identified using miRNA‐specific target proteins in response to both biologics [15]. These findings suggest that inhibition of this network in response to IFNβ‐1a and pegIFNβ‐1a treatment was mediated via upregulated miR‐21‐5p specific downregulation of protein targets.

This study has several strengths and limitations. It is the first to assess plasma miRNA sequencing for identifying longitudinal miRNA PD biomarkers of IFNβ‐1a biologics in healthy participants. Healthy subject studies provide a stable biological baseline without disease‐related confounding factors, allowing clear attribution of biomarker changes to drug activity. The miRNA sequencing method was shown to be sensitive, specific, accurate, and reproducible using a synthetic miRNA reference. Further, to reduce analytical variation, strict operating protocols were used for the sample collection, storage, processing, and sequencing. However, plasma miRNA sequencing, unlike cellular/tissue mRNA sequencing, involves low complexity samples where a few stable, prevalent miRNAs can dominate the reads. Results (see Appendix S1: Supplementary Methods & Results, Section 2) also highlight the need for sufficient or quantitated RNA input from plasma samples and adequate sequencing depth to account for the presence of short RNA fragments (e.g., 8–15 nts) when using patterned flow cells. For statistical analyses, we applied FDR correction for multiple comparisons, which is widely used for biomarker discovery studies. Although we did not technically replicate miR‐223‐3p and miR‐21‐5p as candidate PD biomarkers using a different method, we used published proteomics data generated from the same plasma samples to link miR‐223‐3p and miR‐21‐5p to protein target changes in the same samples to support our results.

In conclusion, we optimized and utilized an automated fit‐for‐purpose miRNA sequencing workflow to identify plasma miRNAs as potential IFNβ‐1a PD biomarker candidates. We identified miR‐223‐3p and miR‐21‐5p as potential candidate PD biomarkers of IFNβ‐1a biologics that were further supported by linking these miRNAs to changes in protein targets from the same interval plasma samples. However, given the overall magnitude and high variability in response of these miRNAs to these products, their use as plasma PD biomarkers of IFNβ‐1a biologics remains to be determined by future studies but provides further understanding of the mechanism of action of these drugs and the role of miRNA‐regulated protein changes. PD biomarkers are essential in drug development and are better identified in healthy subjects due to fewer confounding factors. If confirmed, these miRNA PD biomarkers may have utility beyond biosimilarity assessments in understanding the IFN‐β mechanism of action, supporting dose‐finding studies, and serving as quantitative measures of treatment response. Given the associations between these miRNAs and MS, and since biomarker dynamics may differ between healthy and diseased populations, validation in MS patients at therapeutic doses will be critical for clinical translational relevance. Finally, miRNA NGS complexity, cost, intrinsic method biases, and reduced template input from plasma, together with miRNA diversity, may limit its broad application for the identification and reproducible detection of plasma PD miRNA biomarkers.

Author Contributions

M.M., L.M.C., J.F., and P.L.H. wrote the manuscript. J.F., Y.‐M.W., S.J.S., and P.L.H. designed the research. B.A.R., D.P.S., E.G.M., E.D., C.S., and P.L.H. performed the research. M.M., L.M.C., B.A.R., I.R., and P.L.H. analyzed the data.

Funding

This study was supported by U.S. Food and Drug Administration.

Disclosure

Disclaimer: The opinions expressed in this manuscript are those of the authors and should not be interpreted as the position of the US Food and Drug Administration.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Appendix S1: cts70444‐sup‐0001‐AppendixS1.docx.

Tables S1–S15: cts70444‐sup‐0002‐TablesS1‐S15.xlsx.

Mehanna M., Chekka L. M., Rosenzweig B. A., et al., “Evaluating miRNA Sequencing and Integrative Proteomics for Identifying Plasma miRNAs as Potential Pharmacodynamic Biomarkers of IFNβ‐1a Biologics,” Clinical and Translational Science 18, no. 12 (2025): e70444, 10.1111/cts.70444.