MicroRNA-126 in dogs with immune complex-mediated glomerulonephritis

2.1 Animal specimens

In this retrospective study, samples submitted to the International Veterinary Renal Pathology Service at the time of renal biopsy between 2008 and 2016 were used, along with samples from clinically healthy dogs. Archived serum and urine samples were from 18 dogs with glomerular disease (6 from each diagnostic category) and 6 clinically healthy dogs for RNA-seq (Table S1). For qRT-PCR, additional archived urine was selected from 49 dogs [AMYL (n = 14), GS (n = 14), and ICGN (n = 21)] and 13 clinically healthy dogs (Table S2). For RNA-seq, all dogs in the ICGN category had membranous glomerulonephritis (MGN). For qRT-PCR, the ICGN category consisted of 12 dogs with MGN, 7 dogs with membranoproliferative glomerulonephritis (MPGN), and 2 dogs with mesangioproliferative glomerulonephritis.

For clinically healthy dogs, a physical examination, complete blood count, chemistry panel, urinalysis, and urine protein : creatinine ratio (UPC) were performed to confirm their health status. For dogs with glomerular diseases, the diagnostic category of glomerular disease was confirmed by comprehensive renal biopsy.¹⁵ Samples were selected based on age (≥1 year), urine sediment within 2 weeks of biopsy without discoloration or cloudiness on gross examination and with <5 WBCs/high-power field, <100 RBCs/high-power field, and no bacteriuria, adequate sample volume [serum: ≥1 mL; urine: ≥3 mL (RNA-seq) or 1 mL (qRT-PCR)], and nonhemolyzed serum base on a hemolysis score (see Section 2.3).

Within each diagnostic category, CKD dogs were further divided into Stage A (serum creatinine [sCr] <1.4 mg/dL and with biopsy findings limited to minimal to mild tubulointerstitial fibrosis) and Stage B (sCr ≥1.4 mg/dL or sCr <1.4 mg/dL but inappropriately high for the breed and biopsy results demonstrating moderate tubulointerstitial fibrosis). Tubulointerstitial fibrosis was evaluated by a board-certified anatomic pathologist (REC), scoring 0 to 5, as described.¹⁶ Each diagnostic category had approximately equal numbers of dogs in each stage.

2.2 Sample processing and storage

Urine and serum samples corresponding with the renal biopsy were centrifuged to separate supernatant per standard protocol, shipped overnight on wet ice by the submitting clinician, and stored at −80°C until analysis (1-6 years). For clinically healthy dogs, uncoagulated blood was allowed to sit at room temperature for 30 to 60 minutes after collection, then centrifuged at 1500g for 10 minutes at room temperature to separate serum. Urine was collected via cystocentesis, and urine remaining after the urinalysis was centrifuged at 1000g for 10 minutes at 4°C. Serum and urine were aliquoted and stored at −80°C for approximately 3 years until RNA isolation. For clinically healthy dogs, the Texas A&M University Institutional Animal Care and Use Committee (IACUC) approved the protocol (AUP 2012-021), and client consent was obtained.

2.3 RNA isolation

All serum samples were screened for hemolysis by measuring the A385 and A414, using the NanoDrop 2000 (Thermo Fisher Scientific, Wilmington, Delaware) and the following formula: hemolysis score = A414 − A385 + lipemia correlation factor * A385.¹⁷ Incorporating A385 in the calculation was done to minimize the lipemic interference when measuring A414.¹⁷ Scores were compared with an in-house hemolysis score cutoff value generated from a set of 28 grossly nonhemolyzed, leftover clinical samples submitted to the Texas A&M University Veterinary Medical Teaching Hospital (data not shown).

Circulating RNA was isolated from serum by a modified protocol using the Direct-zol RNA Miniprep Kit (Zymo Research, Irvine, California). For each dog, 1 mL serum was first homogenized with 5 mL QIAzol Lysis Reagent (Qiagen, Germany). The mixture was vortexed, then incubated at room temperature for 5 minutes. Next, 1.2 mL chloroform was added, and lysates were vortexed and incubated at room temperature for 5 minutes, followed by 4°C centrifugation at 13 400g for 15 minutes. After centrifugation, the upper aqueous phase was mixed with 4.8 mL 100% ethanol, added to a Zymo-Spin Column (Zymo Research, Irvine, California), and centrifuged at 12 000g for 30 seconds at room temperature. The spin column was then washed twice with Zymo RNA pre-wash buffer (Zymo Research, Irvine, California), once with Zymo RNA wash buffer (Zymo Research, Irvine, California), and once with 500 μL 80% ethanol, for a total of 4 washes. RNA was eluted with 25 μL 50°C RNase-free water.

Urinary RNA was isolated from 3 mL (for RNA-seq) or 1 mL (for qRT-PCR) urine supernatant from each dog using the Qiagen exoRNeasy Serum/Plasma Maxi Kit (Qiagen, Germany). The manufacturer's protocol was followed up to the point of adding QIAzol Lysis Reagent (Qiagen, Germany). The subsequent steps were identical to the serum isolation protocol, except that the RNeasy MinElute Spin Columns (Qiagen, Germany) were washed 3 times: once with Buffer RWT (Qiagen, Germany) and twice with Buffer RPE (Qiagen, Germany). RNA was eluted with 25 μL 50°C RNase-free water.

2.4 Small RNA sequencing and data analysis

RNA samples were measured using the Fragment Analyzer High Sensitivity RNA Analysis Kit (Advanced Analytical Technologies, Inc., Ankeny, Iowa). The Texas A&M University Genomics Core Lab generated a cDNA library using the NEXTflex Small RNA Library Prep Kit (Bio Scientific Corp, Austin, Texas). Under a 50 base-pair, single-end setting, all 48 cDNA libraries were multiplexed and sequenced in parallel on 3 lanes of a flow cell in an Illumina Genome Analyzer (HiSeq 2500v4) to minimize technical variation and ensure sufficient data output.

Preprocessing of raw reads (fastq files) included removal of the 3′ adapter sequence (TGGAATTCTCGGGTGCCAAGG), trimming of the first and last 4 bases from the adapter-clipped reads (as recommended by the manufacturer),¹⁸ filtering out reads fewer than 16 base-pairs to prevent false degraded RNA or adapter dimers,¹⁹ and removal of low-quality reads (quality score <30). Untrimmed raw reads were discarded as they were unlikely to be miRNAs based on read lengths. FASTX-Toolkit (version 0.0.14) was used to transform the fastq format into collapsed fasta files as proper inputs for CPSS 2.0 (https://mcg.ustc.edu.cn/bsc/cpss/).²⁰ Default settings along with the canine genome (Canis familiaris, CanFam 3.1) and microRNA annotation in miRBase (release 21) were used for analysis. The DESeq2 package in R was used to identify DE miRNAs.²¹ For multiple testing, Wald test P-values were corrected to the false discovery rate (adjusted P-values) by the Benjamini-Hochberg procedure. An adjusted P-value <.05 was set to select DE miRNAs robustly. NormFinder (updated January 2015) was applied to the read count table to identify candidate miRNAs for internal controls for qRT-PCR based on the RNA-seq data,²² and we identified urinary miR-151 and miR-28 as the most appropriate internal controls (Table S3). These miRNAs were not identified as DE miRNAs in any given pair of comparisons in this study.

2.5 qRT-PCR and data analysis

Based on sequencing data and preliminary qRT-PCR results (Figures S1 and S3), urinary miR-126, miR-21, miR-182, and miR-486 were selected as target miRNAs, along with miR-151 and miR-28 as reference miRNAs. The miRNA primers were acquired through GeneGlobe (Qiagen, Germany). RNA was reverse transcribed into cDNA using the T100 Thermocycler (Bio-Rad Laboratories, California) and MiRCURY LNA RT Kit (Qiagen, Germany). The cDNA was diluted using RNAse-free water at a 1:15 ratio and incorporated in the 10 μL PCR reaction using the miRCURY LNA SYBR Green PCR kit (Qiagen, Germany). The epMotion 6000 (Eppendorf, Massachusetts) was used for plate set-up to decrease pipetting errors, replicate, and plate differences. The PCR was run on 384-well plates using the CFX384 Touch Real-Time PCR Detection System (Bio-Rad Laboratories, California). Each sample was run in triplicate to account for technical variability and interplate calibrators. A standard curve for each miRNA and controls (NRT, NTC, SYBR neg, and water) were also included. The mean expression among the triplicates was used for analysis. If there were inconsistencies in the amplification curves, primer dimers were ruled out by running the PCR products on 1% agarose gels.

The mean, SD, coefficient of variation, and amplification efficiency for each miRNA were evaluated, and qbase+ was used to normalize the expression of target miRNAs to the reference miRNAs, calibrate expression differences between plates, and remove samples that did not meet QC requirements. Samples with a Cq difference >1.0 between the technical replicates and samples from any plates with miRNA amplification efficiencies <80% were excluded from the analysis. Mann-Whitney tests and 1-way ANOVA were used to determine significant differences in log₁₀ normalized mean expression between diagnostic categories (miR-126) and clinical stage (miR-21, miR-182, miR-486). Mann-Whitney pair-wise nonparametric tests were used to evaluate miRNAs with non-normal distributions, while 1-way ANOVA was used to evaluate miRNAs with normal distribution. The diagnostic performance of miRNAs as biomarkers using qRT-PCR was evaluated with the Wilson/Brown receiver operating characteristic (ROC) curve analysis. Area under the curve (AUC), 95% confidence intervals (95% CI), and cut points were calculated using non-parametric ROC methods. AUCs were considered statistically significant when P < .05.

The Shapiro-Wilk goodness of fit test assessed the normality of clinical variables (age, sCr, UPC, urine specific gravity [USG], and fibrosis). Differences in clinical variables among diagnostic categories and stages were performed using Mann-Whitney pair-wise nonparametric tests. Correlations between clinical variables and target miRNAs were evaluated using Spearman correlations, with P < .05 considered significant. All analyses were performed in GraphPad Prism9 (GraphPad Software, San Diego, California) except for the AUC analysis, which used STATA17 (StataCorp LLC, College Station, Texas).

3.1 Clinical variables for each diagnostic category and clinical stage

The median age was significantly lower for the ICGN category compared with AMYL (P < .05) and GS (P < .05) for dogs in the RNA-seq cohort and significantly lower compared with GS (P < .05) in the qRT-PCR cohorts. The median age of dogs in the GS category was significantly higher than controls in the qRT-PCR cohort (Figure 1A,B, P < .05). No significant difference in median sCr was observed among diagnostic categories for the RNA-seq cohort; however, in the qRT-PCR cohort, sCr was significantly higher in dogs with ICGN compared with dogs with AMYL and controls (Figure 1C,D, P < .05). UPC was significantly higher in dogs with ICGN compared with dogs with GS, and UPC was significantly higher in all diagnostic categories compared with controls (Figure 1E,F; P < .05). As expected, based on our study design, sCr was significantly higher in stage B dogs than in stage A dogs and controls (Figure S2; P < .05). UPC was also significantly higher in stage A and B dogs compared with controls (Figure S2; P < .05). There was no difference in age between stage A and B dogs (Figure S2). For complete clinical information, refer to Tables S1 and S2.

3.2 Small RNA-seq: Differentially expressed (DE) serum and urinary miRNAs

Forty-eight RNA samples were used for RNA-seq, 24 each from serum and urine. One serum sample representing stage B ICGN was excluded for low reads (<5 million; Table S3). Sequenced serum samples averaged 6.5 million reads and a 97.6% genome mapping rate, with 1.6 million reads mapped to miRNAs. Urine samples had an average of 6.9 million reads and a 90.1% genome mapping rate, with 79 019 reads mapped to miRNAs. On average, 167 and 88 miRNAs with at least 10 mapped reads were detected in serum and urine samples, respectively.

Overall, 38 serum and 16 urinary miRNAs were DE in CKD dogs vs controls (Figure 2). In each diagnostic category, CKD dogs were compared with controls regardless of disease stage (Figure 2A,D). CKD dogs were further divided into stages A (Figure 2B,E) and B (Figure 2C,F) and compared with controls. Comparing stage A dogs with controls, 8 serum and 7 urinary DE miRNAs were detected, while 39 serum and 22 urinary DE miRNAs were discovered comparing stage B dogs with controls. Regardless of diagnostic category or stage, 5 serum miRNAs (miR-107, miR-129, miR-186, miR-365, and miR-371) and 5 urinary miRNAs (miR-7, miR-9, miR-22, miR-203, and miR-423a) were DE in dogs with glomerular disease when compared with controls (Figure 2A,D). Downregulated serum miR-186 was also among the 6 common DE miRNAs among all 3 diagnostic categories when comparing stage B CKD dogs to controls (Figure 2C). Downregulated urinary miR-7 and miR-22 were the only 2 common DE miRNAs among all 3 diagnostic categories identified in stage A dogs compared with controls. They also comprised 2 of the 5 DE urinary miRNAs in stage B dogs compared with controls (Figure 2E,F). Lists of all serum and urinary DE miRNAs are provided in Tables S5 and S6.

3.3 Small RNA-seq: DE miRNAs associated with disease progression

When all CKD dogs were combined, regardless of the glomerular diagnostic category, no serum miRNAs were DE between stages A and B. However, 3 urinary DE miRNAs were identified between stage A and B dogs, including upregulated miR-182 and miR-21 and downregulated miR-486 (Figure S3). The expression of urinary miR-486 was significantly decreased in stage B dogs compared with both stage A dogs and controls (Figure S3).

3.4 Small RNA-seq: Glomerular disease-specific urinary miRNAs

We aimed to identify biofluid-derived miRNAs exclusively differentially expressed in specific diagnostic categories of glomerular diseases in dogs. No serum miRNAs were differentially expressed among the different diagnostic categories when including all dogs in each category, regardless of disease stage.

When comparing disease stages within each diagnostic category, there was upregulation of serum miR-1836 for AMYL in several comparisons (Table 1). The annotation of cfa-miR-1836 overlaps with snoRNA 20 (SNORA20)²³; therefore, miR-1836 was excluded from further analysis. For stage A, 3 serum miRNAs (miR-335, miR-101, and miR-32) and 5 serum miRNAs (miR-320, miR-99b, miR-218, miR-335, and miR-485) were differentially expressed in ICGN compared with AMYL and GS, respectively. No urinary DE miRNAs were identified among stage A samples. In stage B dogs, serum miR-350 and miR-374a were differentially expressed in GS compared with AMYL and ICGN, respectively, while 5 urinary DE miRNAs were detected in at least 1 pair of comparisons among the 3 glomerular diseases. Notably, the distinctive expression of urinary miR-126, miR-335, and miR-128 could correctly group stage B dogs into ICGN, GS, or AMYL (Figure 3). Based on preliminary qRT-PCR results, miR-126 appeared most promising and was therefore selected for further evaluation as a marker of ICGN.

3.5 qRT-PCR of urinary miR-126, miR-21, miR-182, and miR-486

The qRT-PCR confirmed that the normalized mean expression of urinary miR-126 was significantly higher in dogs with ICGN compared with AMYL, GS, and controls (Figure 4), being 10.5 times higher than in dogs with GS (95% CI: 4.1-26.7; P < .001) and 28.9 times higher compared with dogs with AMYL (95% CI: 10.5-79.8; P < .001). Urinary miR-126 expression was also 2.8 times higher in dogs with GS compared with dogs with AMYL (95% CI: 1.0-7.6; P < .05). No significant difference was observed within ICGN dogs between stage A and B or the ICGN subcategories (not shown). Urinary miR-126 expression was weakly to moderately correlated with sCr (r = .330, P = .008), UPC (r = .493, P < .001), and fibrosis (r = .39, P = .006; Table S7).

The normalized mean expression of urinary miR-21 and miR-182 differed between stages and were 1.63 (95% CI: .86-3.1) and 1.45 (95% CI: .82-2.6) times higher, respectively, in stage B vs stage A dogs (Figure 5). Despite being differentially expressed based on RNA-seq, there was no difference in miR-486 between stage A and B dogs based on qRT-PCR (Figure 5). MiR-21 was moderately correlated with UPC (r = .42, P < .001) and weakly correlated with fibrosis (r = .32, P = .03; Table S7). Likewise, miR-182 was weakly correlated with UPC (r = −.25, P = .05) and fibrosis (r = .28, P = .05; Table S7). No significant correlations with clinical variables were observed for miR-486. Fibrosis was weakly correlated with clinical stage (r = .39, P = .006; Table S7), but no significant difference in fibrosis score between Stage A and B dogs was seen (Figure S4). There was also no difference in the normalized mean expression of miR-21 or miR-182 between dogs with minimal, mild, and moderate fibrosis (Figure S4).

3.6 Diagnostic evaluation of urinary miR-126

ROC curve analysis was performed to assess the ability of urinary miR-126 to distinguish between ICGN and non-ICGN (AMYL and GS; Figure 6). With a normalized mean expression and Log₁₀ cut-point of 1.48, the AUC for miR-126 to differentiate dogs with ICGN vs non-ICGN was 0.95 (95% CI: 0.88-1.0, P < .001), with a sensitivity and specificity of 90% (95% CI: 71%-98%) and 92% (95% CI: 77%-99%), respectively.