Diagnostic potential of circulating cell‐free microRNAs for community‐acquired pneumonia and pneumonia‐related sepsis

Abstract Cell‐free microRNAs (miRNAs) are transferred in disease state including inflammatory lung diseases and are often packed into extracellular vesicles (EVs). To assess their suitability as biomarkers for community‐acquired pneumonia (CAP) and severe secondary complications such as sepsis, we studied patients with CAP (n = 30), sepsis (n = 65) and healthy volunteers (n = 47) subdivided into a training (n = 67) and a validation (n = 75) cohort. After precipitating crude EVs from sera, associated small RNA was profiled by next‐generation sequencing (NGS) and evaluated in multivariate analyses. A subset of the thereby identified biomarker candidates was validated both technically and additionally by reverse transcription quantitative real‐time PCR (RT‐qPCR). Differential gene expression (DGE) analysis revealed 29 differentially expressed miRNAs in CAP patients when compared to volunteers, and 25 miRNAs in patients with CAP, compared to those with sepsis. Sparse partial‐least discriminant analysis separated groups based on 12 miRNAs. Three miRNAs proved as a significant biomarker signature. While expression levels of miR‐1246 showed significant changes with an increase in overall disease severity from volunteers to CAP and to sepsis, miR‐193a‐5p and miR‐542‐3p differentiated patients with an infectious disease (CAP or sepsis) from volunteers. Cell‐free miRNAs are potentially novel biomarkers for CAP and may help to identify patients at risk for progress to sepsis, facilitating early intervention and treatment.


| INTRODUC TI ON
According to the World Health Organization's global health estimates, lower respiratory infections are the fourth leading global cause of deaths and the most deadly communicable disease, causative for three million deaths worldwide in 2016. 1 With the current SARS-CoV-2 pandemic, this fact has very recently been brought to worldwide attention. Data from a prospectively followed multicentre trial revealed an overall mortality of 17.3% for patients with CAP within an 18 months follow-up. 2 Despite the introduction of antibiotic therapies in the 1950s, pneumonia mortality has not decreased substantially, 3 and sepsis, septic shock or acute pulmonary failure (eg acute respiratory distress syndrome, ARDS) are frequent secondary complications. 4,5 At present, initial pneumonia diagnosis is based on suggestive clinical features such as fever, shortness of breath, sputum production, cough and leukocytosis supplemented by evidence of pulmonary consolidation found in chest X-rays or computed tomography (CT) if required in order to arrive at a final diagnosis. To improve management and treatment of pneumonia, supporting microbiological and virological tests from throat swabs, sputum or blood cultures might be indicated to identify the responsible pathogen(s) and to allow targeted antimicrobial or antiviral therapy. This can be supported by urine antigen tests, molecular assays, serology or bronchoscopy in selected cases. Blood biomarkers such as procalcitonin (PCT), C-reactive protein (CRP), Interleukin-6 (IL-6), white blood cell count and lactate 6,7 are commonly used to differentiate between patients with pneumonia and individuals with pneumonia at risk for sepsis. Scoring systems including the Confusion, Blood Urea, Respiratory Rate, Blood Pressure, Age ≥ 65 (CURB-65) score 8 are applied in patients with pneumonia to simplify site-of-care decisions such as outpatient treatment vs. hospital and intensive care unit (ICU) admission and to facilitate the decision whether to prescribe antibiotics or not. 6 However, the sensitivity and specificity of these scoring systems are limited. 9 The reliable diagnosis of pneumonia can be a time-consuming and complex process. It is particularly challenging in high-risk groups 10 such as the elderly or infants, which often present with atypical symptoms and are at an increased risk for sepsis or acute pulmonary failure as secondary complications.
At present, there are no valid and reliable biomarkers allowing an on-site diagnosis and the identification of high-risk patients.
Circulating EVs are a heterogeneous group of small-sized membranous vesicles that are loaded with biomolecules, particularly proteins, lipids and diverse types of nucleic acids and are exchanged in cell-cell signalling during various physiological and pathological processes. 11 EV-associated miRNAs are key regulators in the pathogenesis of infectious and non-infectious pulmonary disorders. [12][13][14][15] Therefore, differential miRNA expression in EV samples from liquid biopsies may indicate the presence of an inflammatory lung disease and discriminate between different disease stages or even predict the course of the disease. 16,17 miR-NAs might reflect progression from physical health, to mild and more severe forms of pneumonia. 18 Additionally, EVs from sepsis patients contain miRNAs and messenger RNAs (mRNAs) related to disease-associated pathways, such as inflammatory response, oxidative stress and cell cycle regulation. 19 Therefore, extracellular miRNAs might be attractive diagnostic biomarkers for pulmonary inflammation and prognostic indicators for disease progression.
In the present study, we identified cell-free miRNA biomarker candidates by high-throughput small RNA sequencing (small RNAseq) to differentiate between patients with CAP and healthy volunteers, and to distinguish CAP patients from those with sepsis. The candidate biomarker signature was first technically validated by RT-qPCR in the same training cohort of individuals and subsequently confirmed in a second, independent validation cohort.

| Blood sampling
Blood from ICU patients was drawn from 20 gauge catheters within the radial artery (8 cm polyethylene catheter, Vygon, Aachen, Germany) on the day of admission to the ICU, while patients with CAP and healthy volunteers were sampled by venipuncture using 21-gauge needles (Safety-Multifly, Sarstedt AG & Co, Nümbrecht, Germany). We recently showed, that arterial vs. venous blood sampling has insignificant effects on EV miRNA expression. 22 Blood was drawn into 9 ml serum collection tubes (S-Monovette, Sarstedt AG&Co, Nümbrecht, Germany) each, allowed to clot for 30 minutes and centrifuged at 3400 g for 10 minutes at room temperature (RT).
Within 10 minutes of separation, serum was aliquoted and immediately stored at −80°C.

| Sample preparation
Samples were processed according to the manufacturer's protocols. Crude EVs were precipitated (miRCURY Exosome Isolation Kit-Serum and Plasma, Qiagen, Venlo, the Netherlands) from either 1 ml (small RNA-seq in the training cohort and additional RT-qPCR confirmation in the validation cohort) or 0.75 ml (technical RT-qPCR validation in the training cohort) of serum, respectively.
As previously shown by our group, in contrast to other EV isolation methods, precipitation allows for reliable separation of sepsis patients and healthy volunteers in small RNA-seq analyses. 23 As precipitation co-isolates cell-free non-EV miRNA carriers, such as high-and low-density lipoproteins, argonaute-2 protein complexes and others, 24,25 it should be noted that miRNAs from this study are not exclusively EV-derived. Samples were therefore designated as crude EVs. After extracting cell-free total RNA, size distribution and yield were assessed by capillary electrophoresis

| Small RNA sequencing
As described previously, 23  Single-end sequencing ran in 50 cycles on the HiSeq2500 (Illumina Inc, San Diego, USA).
Quality control of small RNA-seq data, trimming of adaptor sequences and alignment of reads was performed as described before. 22 Only samples with a minimum of 1 million reads altogether and 15% of miRNA reads in relation to total library size were included for analyses. DGE analysis was conducted by DESeq2 (version 1.22.1) 26 for R (version 3.5.1) with the implemented normalisation strategy based on library size correction and the Benjamini-Hochberg method to correct for the false discovery rate (FDR). miRNAs were filtered by setting a mean expression across all samples of ≥50 reads (baseMean), a minimum twofold up-or down-regulation (log2 fold change, log2FC ≥1 or log2FC ≤ −1) and adjusted P-value padj ≤ 0.05. miRNAs that failed to be detected in more than one sample per group were removed prior to selecting the most drastically dysregulated miRNAs based on log2FC. Unsupervised clustering was performed by principal component analysis (PCA). Additionally, after filtering small RNA-seq data for baseMean ≥500, supervised clustering was performed by sparse partial-least-squares discriminant analysis (sPLS-DA) with four components and maximum five features each using mixOmics 27 to assess the minimal number of miRNAs required to separate groups. Combining data from both, DGE analysis and sPLS-DA, a subset of candidate miRNAs serving as a whole biomarker signature was selected. Statistical significance of RNA mapping was tested with the non-parametric Kruskal-Wallis test followed by Dunn's multiple comparison test using Graphpad Prism (version 8.3.0). Relative mapping frequencies were reported as mean values ± standard deviation.

| RT-qPCR validation
The most stably expressed miRNAs among all groups were evaluated from the NGS data set as potential reference miRNAs by NormFinder. 28 Validations of the biomarker signature were performed by RT-qPCR using the LNA-optimized miRNA PCR system (miRCURY LNA RT kit, miRCURY LNA SYBR Green PCR kit, Qiagen, Venlo, the Netherlands).
For reverse transcription, 6.5 µl of cell-free total RNA was used as template for cDNA synthesis. qPCR reactions were prepared according to the manufacturer's recommendation with the appropriate miR-CURY LNA miRNA PCR Assays (Qiagen, Venlo, the Netherlands) for the biomarker candidates and the reference miR-30d-5p. The UniSp6 assay (Qiagen, Venlo, the Netherlands) was used as control for cDNA synthesis and PCR amplification. qPCR reactions were run in trip- Spearman's rank-order correlation. Spearman's r and 95% confidence intervals were reported. Additionally, group classification of RT-qPCR data from the independent validation cohort was performed based on partial-least-squares discriminant analysis (PLS-DA) of RT-qPCR data from the training cohort. Expression values of validated miRNAs were depicted as median and IQR.

| Statistical analysis of demographics and clinical data
Demographic characteristics and clinical data were compared using the non-parametric Mann-Whitney U test or in case of more than two groups by ANOVA on Ranks followed by Dunn's post hoc test. Data in the text and in tables are reported as median and IQR. All statistical tests were two-tailed, and a P-value <0.05 was considered statistically significant.

| Identification of pathways relevant to community-acquired pneumonia
Ingenuity Pathway Analysis (IPA, Spring version 2020, Qiagen Bioinformatics, Redwood, USA) was used for the in silico identification of gene targets and causal networks from the high-throughput miRNA expression data of the training set (n = 67 patients). Only the 29 miRNAs meeting the predefined cut-off values (baseMean ≥50, log2FC ≥1 or log2FC ≤ −1 and padj ≤0.05) were entered into IPA, and only experimentally confirmed relationships were considered for the identification of miRNA targets and the characterisation of regulatory effects. Possible gene targets were identified using the IPA 'microRNA Target Filter' to identify target genes and to construct networks of relevance to CAP. Disease filtering was set to 'infectious disease', and the network for 'cellular and humoral immune response' was selected.

| Ethics approval and consent to participate
Approval of the study was granted by the Ethics Committee of the Medical Faculty of the Ludwig-Maximilians-University of Munich under Protocol #551-14. All samples were anonymised during analyses. The study was conducted in accordance with approved guidelines, and written informed consent to participate was obtained from each participant or the patient's legal representative.  Table S4 of the online supplement for details).

| RNA yield, sequencing quality and mapping distribution
As different RNA isolation kits might influence miRNA recovery and composition of RNA, 30  Library sizes ( Figure 1A), as well as numbers of mapped miRNAs ( Figure 1B) tended to be higher in volunteers when compared to patients, while no apparent difference was present between both patient groups. Relative miRNA frequencies were 40.9 ± 11.3% for volunteers, 28.0 ± 9.3% for CAP patients and 31.5 ± 14.4% for patients with sepsis. Additionally, patient groups had higher frequencies of short sequences < 16 nucleotides in size (CAP: 20.3 ± 15.6%, sepsis: 15.0 ± 12.3%) when compared to volunteers 8.5 ± 8.4%, which probably represent degradation products from longer coding and non-coding RNA species (CAP vs. volunteers: P = 0.012, sepsis vs. volunteers: P = 0.061, CAP vs. sepsis: P = 0.831). The mapping distribution is visualised by relative mean frequencies ( Figure 1C).

| Small RNA sequencing data analyses
When performing unsupervised clustering based on the 500 miR-NAs with highest variance, different patient groups overlapped, but could be distinctly separated from volunteers (Figure 2A). Patients with CAP and lower overall disease severity tended to reflect the miRNA profiles of volunteers more closely than sepsis patients, who were more distinctly separated.
Analysing DGE data, the comparison of CAP patients with volunteers revealed 29 significantly up-or down-regulated miRNAs (see Table S5 of the supporting information).
Among these cell-free miRNAs, expression levels of several transcripts correlated with indicators of disease severity in CAP patients including the CURB-65 score as assessment score for pneumonia and the associated risk of mortality, 8 total duration of hospital stay, and plasma levels of NGAL (see Table 1 for details). Only miR-127-3p was related to demographic variables of CAP patients (age: r = 0.627, P = 0.029) but none of the other miRNAs appeared to be influenced by demographics.
Comparing patients with CAP to those with sepsis, we detected 25 miRNAs with altered expression (see Table S5 of the supporting information). A number of these miRNAs correlated significantly with treatment variables, indicating face validity of these transcripts.
In particular, expression values of miR-1-3p were related to plasma concentrations of the inflammatory markers IL-6, NGAL and vasopressor requirements in sepsis (see Table 2 for details).
Based on sPLS-DA, group separation was achieved on the basis of twelve miRNAs as discriminators ( Figure 2B). Out of these, four miRNAs (miR-182-5p, miR-193a-5p, miR-215-5p, miR-93-5p) were also detected using DESeq2 with log2FC ≥1 or log2FC ≤ −1 (see Table S5 of the supporting information for comparison), while all other miRNAs showed no regulation according to our threshold settings for DGE analysis.

| RT-qPCR validation
The eighteen candidate miRNAs were quantified by RT-qPCR, as a supplementary method, in the same cohort that was used for highthroughput sequencing and additionally confirmed in an independent second cohort of individuals. After quality control of RT-qPCR data, twelve miRNAs remained for analyses.
By reading in RT-qPCR data of all twelve miRNAs measured in the training cohort to PLS-DA, groups were separated and equally distributed compared to the small RNA-seq data ( Figure 3A). According to findings from PLS-DA, samples in the validation cohort could be correctly classified using principal components 1-10 and displayed F I G U R E 1 Library sizes (A) and mapped miRNA reads (B) are depicted for volunteers, patients with community-acquired pneumonia (CAP) and sepsis as median values and 95% confidence intervals. Library sizes, as well as numbers of mapped miRNA reads tended to be higher in volunteers compared to patients, with no apparent difference between patient groups. Read counts in the mapping distribution (C) are visualised as mean relative frequencies. Volunteers have higher relative miRNA frequencies, as well as fewer incidences of short sequences (<16 nucleotides) when compared to patient groups. No adaptor: sequence lacking adaptors; short: sequence < 16 nucleotides; unmapped: sequence did not align to any of the mapped RNA classes; rRNA: ribosomal RNA; snoRNA: small nucleolar RNA; snRNA: small nuclear RNA; tRNA: transfer RNA; miRNA: microRNA; * P < 0.05; ** P < 0.005; ns: not significant high discriminatory power with 73.3% of correctly assigned samples for reduced models from principal components 1-6 or 1-8, respectively ( Figure 3B).
Out of these miRNAs, some transcripts showed the same expression pattern for both group comparisons independent of cohort and approach used for analysis, albeit with reversed trends ( Figure 4A and C). With one exception, Spearman's rank-order correlation of miRNA expression from the NGS and RT-qPCR data sets showed an overall significant positive relationship ( Figure 4B and D). The three most drastically dysregulated miRNAs (CAP vs. volunteers: miR-193a-5p, miR-542-3p and miR-1246; CAP vs. sepsis: miR-1246) with equal expression pattern for both approaches and cohorts of individuals were selected as candidate biomarkers for further analysis.

| 'The cellular and humoral immune response' in community-acquired pneumonia
When IPA target filtering was set to 'cellular and humoral immune response', the resulting canonical network in CAP patients was primarily affected by 6 of the 29 significantly regulated miRNAs with 42 possible target mRNA transcripts. Figure  One of the identified molecular targets of miR-542-3p was inflammatory mediator PTGS2 (prostaglandin G/H synthase 2).

| D ISCUSS I ON
Our study outlines the possibility of using cell-free miRNA biomarkers to discriminate patients with CAP from healthy volunteers and from those with sepsis as a severe secondary complication. We characterised a subset of twelve miRNAs as a potential biomarker for this purpose, technically validated miR-193a-5p, miR-542-3p and miR-1246 and confirmed our findings in an independent cohort.
When analysing relative miRNA expression, miR-1246 showed significant increases with more severe overall disease from volunteers to patients with CAP and to those with sepsis, whereas miR-193a-5p and miR-542-3p differentiated patients with an infectious disease (CAP or sepsis) from healthy individuals.
The miR-193a/b-5p family was previously shown to be a possible indicator for CAP, as expression of miR-193b-5p was related to the CURB-65 score, a validated clinical prediction score for pneumonia and the associated risk of mortality 8 and also to NGAL plasma levels. NGAL has recently been described as a useful biomarker for lower respiratory tract infections 33 and the associated mortality 34 and may also serve as an indicator of an ongoing risk for renal injury, 35 which is common in patients with pulmonary disorders. 36 Moreover, expression levels of miR-193a-5p were related to the duration of the required hospital stay in our study. Identified molecular targets of miR-193a-5p included IL-10, which is known to correlate with the CURB-65 score and is associated with increased mortality in patients with CAP 37 and mTOR, which has been shown to be down-regulated in sepsis due to CAP. 38 Our group previously showed the positive correlation of cell-free miR-193a-5p expression with disease severity in sepsis patients. 17 Expression levels of miR-193 in serum were also associated with death from sepsis in recent studies. 39 miR-542-3p was shown to be a causal mediator of mitochondrial dysfunction in muscle tissue of patients with sepsis. 40 In our study, one of the molecular targets of miR-542-3p was PTGS2, a pro-inflammatory mediator, which is known to be responsible for the production of prostaglandins that are involved in the inflammatory response. 41 Moreover, elevated miR-1246 levels have been shown to mediate lipopolysaccharide-induced apoptosis of pulmonary endothelial cells and acute lung injury. 42 In conclusion, our findings indicate that these three cell-free miRNAs may be valuable for the diagnosis of CAP and sepsis as a severe secondary complication.
The fact that these miRNAs were also associated to EVs isolated from the peripheral circulation suggests a possible role of these vesicles in mediating inflammatory signals from the lung to the periph- However, the fact that our final biomarker signature has been confirmed for two different RNA extraction methods (1), miRNA analysing platforms (2) and individual cohorts (3) might represent additional proof of its stability.
In conclusion, our findings indicate that in patients with suspected CAP, cell-free miR-193a-5p, miR-miR-542-3p and miR-1246 may serve as indicators for CAP, whereas a further increase in miR-1246 may suggest an increased risk to develop sepsis. For the successful use of these miRNAs as biomarkers, studies in larger patient cohorts with both conditions will be required to confirm our data of this novel diagnostic approach.
F I G U R E 4 Log2 fold changes (log2FC) for CAP vs. volunteers (a), and CAP vs. sepsis (c) calculated from next-generation sequencing (NGS, black) and reverse transcription quantitative real-time PCR (RT-qPCR) data in the training (TC, dark blue) and validation cohort (VC, light blue). Significant changes are marked by asterisks. Spearman's correlation matrix of expression levels (log2FC) from NGS and RT-qPCR data sets shown as heatmaps for CAP vs. volunteers (b) and CAP vs. sepsis