Machine learning for classification of hypertension subtypes using multi-omics: A multi-centre, retrospective, data-driven study

Background

Arterial hypertension is a major cardiovascular risk factor. Identification of secondary hypertension in its various forms is key to preventing and targeting treatment of cardiovascular complications. Simplified diagnostic tests are urgently required to distinguish primary and secondary hypertension to address the current underdiagnosis of the latter.

Methods

This study uses Machine Learning (ML) to classify subtypes of endocrine hypertension (EHT) in a large cohort of hypertensive patients using multidimensional omics analysis of plasma and urine samples. We measured 409 multi-omics (MOmics) features including plasma miRNAs (PmiRNA: 173), plasma catechol O-methylated metabolites (PMetas: 4), plasma steroids (PSteroids: 16), urinary steroid metabolites (USteroids: 27), and plasma small metabolites (PSmallMB: 189) in primary hypertension (PHT) patients, EHT patients with either primary aldosteronism (PA), pheochromocytoma/functional paraganglioma (PPGL) or Cushing syndrome (CS) and normotensive volunteers (NV). Biomarker discovery involved selection of disease combination, outlier handling, feature reduction, 8 ML classifiers, class balancing and consideration of different age- and sex-based scenarios. Classifications were evaluated using balanced accuracy, sensitivity, specificity, AUC, F1, and Kappa score.

Findings

Complete clinical and biological datasets were generated from 307 subjects (PA=113, PPGL=88, CS=41 and PHT=112). The random forest classifier provided ∼92% balanced accuracy (∼11% improvement on the best mono-omics classifier), with 96% specificity and 0.95 AUC to distinguish one of the four conditions in multi-class ALL-ALL comparisons (PPGL vs PA vs CS vs PHT) on an unseen test set, using 57 MOmics features. For discrimination of EHT (PA + PPGL + CS) vs PHT, the simple logistic classifier achieved 0.96 AUC with 90% sensitivity, and ∼86% specificity, using 37 MOmics features. One PmiRNA (hsa-miR-15a-5p) and two PSmallMB (C9 and PC ae C38:1) features were found to be most discriminating for all disease combinations. Overall, the MOmics-based classifiers were able to provide better classification performance in comparison to mono-omics classifiers.

Interpretation

We have developed a ML pipeline to distinguish different EHT subtypes from PHT using multi-omics data. This innovative approach to stratification is an advancement towards the development of a diagnostic tool for EHT patients, significantly increasing testing throughput and accelerating administration of appropriate treatment.

Funding

European Union's Horizon 2020 Research and Innovation Programme under Grant Agreement No. 633983, Clinical Research Priority Program of the University of Zurich for the CRPP HYRENE (to Z.E. and F.B.), and Deutsche Forschungsgemeinschaft (CRC/Transregio 205/1).

Systematic review: Association between circulating microRNA expression & stroke

This systematic review aimed to establish the range and quality of clinical and preclinical evidence supporting the association of individual microRNAs, and the use of microRNA expression in the diagnosis and prognosis of ischaemic or haemorrhagic stroke. Electronic databases were searched from 1993 to October 2021, using key words relevant to concepts of stroke and microRNA. Studies that met specific inclusion and exclusion criteria were selected for data extraction. To minimise erroneous associations, findings were restricted to microRNAs reported to change in more than two independent studies. Of the papers assessed, 155 papers reported a change in microRNA expression observed in more than two independent studies. In ischaemic studies, two microRNAs were consistently differentially expressed in clinical samples (miR-29b & miR-146a) and four were altered in preclinical samples (miR-137, miR-146a, miR-181b & miR-223-3p). Across clinical and preclinical haemorrhagic studies, four microRNAs were downregulated consistently (miR-26a, miR-126, miR-146a & miR-155). Across included studies, miR-126 and miR-146a were the only two microRNAs to be differentially expressed in clinical and preclinical cohorts following ischaemic or haemorrhagic stroke. Further studies, employing larger populations with consistent methodologies, are required to validate the true clinical value of circulating microRNAs as biomarkers of ischaemic and haemorrhagic stroke.

Circulating microRNAs as Diagnostic Markers in Primary Aldosteronism

Simple Summary

Many patients remain at increased risk of primary aldosteronism (PA) and its consequences due to the difficulty of accurate diagnosis. MicroRNAs circulating in the bloodstream are emerging as biomarkers for disease, particularly specific forms of cancer. In this review article, we argue that they may also have a role in the diagnosis of PA, if observed changes in the microRNA profile of PA tissue are reflected in circulating microRNAs, which can be sampled and analysed readily in a clinical setting. However, for various practical reasons, studies of potential diagnostic circulating microRNAs have often proved difficult to reproduce consistently. We describe these problems and how they might be overcome using, as an example, our design of the circulating microRNA arm of the ongoing ENS@T-HT project, which is intended to confirm whether circulating microRNAs can serve as biomarkers for PA.

Abstract

Primary aldosteronism (PA) is a common and highly treatable condition, usually resulting from adrenocortical tumorous growth or hyperplasia. PA is currently underdiagnosed owing to its complex and protracted diagnostic procedures. A simplified biomarker-based test would be highly valuable in reducing cardiovascular morbidity and mortality. Circulating microRNAs are emerging as potential biomarkers for a number of conditions due to their stability and accessibility. PA is known to alter microRNA expression in adrenocortical tissue; if these changes or their effects are mirrored in the circulating miRNA profile, then this could be exploited by a diagnostic test. However, the reproducibility of studies to identify biomarker-circulating microRNAs has proved difficult for other conditions due to a series of technical challenges. Therefore, any studies seeking to definitively identify circulating microRNA biomarkers of PA must address this in their design. To this end, we are currently conducting the circulating microRNA arm of the ongoing ENS@T-HT study. In this review article, we present evidence to support the utility of circulating microRNAs as PA biomarkers, describe the practical challenges to this approach and, using ENS@T-HT as an example, discuss how these might be overcome.

Altered Extracellular Vesicle MicroRNA Expression in Ischemic Stroke and Small Vessel Disease

Active transport of microRNAs (miRNA) in extracellular vesicles (EV) occurs in disease. Circulating EV-packaged miRNAs in the serum of stroke patients were compared to stroke mimics with matched cardio- and cerebrovascular risk factors, with corroboration of results in a pre-clinical model. An unbiased miRNA microarray was performed in stroke vs. stroke mimic patients (n = 39). Results were validated (n = 173 patients) by real-time quantitative polymerase chain reaction. miRNA expression was quantified in total serum/EV (n = 5-7) of naïve adult spontaneously hypertensive stroke-prone rats (SHRSP), their normotensive reference strain (Wistar Kyoto, WKY) and in circulating EV (n = 3), peri-infarct brain (n = 6), or EV derived from this region (n = 3) in SHRSP following transient middle cerebral artery occlusion (tMCAO). Circulating EV concentration did not differ between stroke and stroke mimic patients. The microarray identified many altered EV-packaged miRNAs: levels of miRNA-17-5p, -20b-5p and -93-5p (miRNA-17 family members) and miRNA-27b-3p were significantly (p ≤ 0.05) increased in stroke vs. stroke mimic patients. Patients with small vessel disease (SVD) consistently had the highest miRNA levels. Circulating EV concentration was unaltered between naïve SHRSP and WKY but levels of miRNA-17-5p and -93-5p were significantly increased in SHRSP. tMCAO in SHRSP did not further alter circulating EV miRNA-17 family member expression and nor did it change total miRNA-17 family levels in peri-infarct brain tissue or in EV isolated from this region at 24 h post-tMCAO. Changes in EV packaged miRNA expression was validated in patients with stroke, particularly those with SVD and corroborated pre-clinically. Together, altered circulating EV levels of miRNA-17 family members may reflect the chronic sequelae underlying cerebrovascular SVD rather than the acute ischemic stroke itself.

Preclinical Validation of the Therapeutic Potential of Glasgow Oxygen Level Dependent (GOLD) Technology: a Theranostic for Acute Stroke

In acute stroke patients, penumbral tissue is non-functioning but potentially salvageable within a time window of variable duration and represents target tissue for rescue. Reperfusion by thrombolysis and/or thrombectomy can rescue penumbra and improve stroke outcomes, but these treatments are currently available to a minority of patients. In addition to the utility of Glasgow Oxygen Level Dependent (GOLD) as an MRI contrast capable of detecting penumbra, its constituent perfluorocarbon (PFC) oxygen carrier, combined with normobaric hyperoxia, also represents a potential acute stroke treatment through improved oxygen delivery to penumbra. Preclinical studies were designed to test the efficacy of an intravenous oxygen carrier, the perfluorocarbon emulsion Oxycyte® (O-PFC), combined with normobaric hyperoxia (50% O₂) in both in vitro (neuronal cell culture) and in vivo rat models of ischaemic stroke. Outcome was assessed through the quantification of lipid peroxidation and oxidative stress levels, mortality, infarct volume, neurological scoring and sensorimotor tests of functional outcome in two in vivo models of stroke. Additionally, we investigated evidence for any positive or negative interactions with the thrombolytic recombinant tissue plasminogen activator (rt-PA) following embolus-induced stroke in rats. Treatment with intravenous O-PFC + normobaric hyperoxia (50% O₂) provided evidence of reduced infarct size and improved functional recovery. It did not exacerbate oxidative stress and showed no adverse interactions with rt-PA. The positive results and lack of adverse effects support human trials of O-PFC + 50% O₂ normobaric hyperoxia as a potential therapeutic approach. Combined with the diagnostic data presented in the preceding paper, O-PFC and normobaric hyperoxia is a potential theranostic for acute ischaemic stroke.

Regulation of Aldosterone Secretion

Secretion of the major mineralocorticoid aldosterone from the adrenal cortex is a tightly-regulated process enabling this hormone to regulate sodium homeostasis and thereby contribute to blood pressure control. The circulating level of aldosterone is the result of various regulatory mechanisms, the most significant being those controlled by the renin-angiotensin system and plasma potassium levels. The importance of maintaining tight control over aldosterone secretion is demonstrated by cases of dysregulation, which can result in severe hypertension and significantly increased cardiovascular risk. In this article we summarize current knowledge of the major regulatory mechanisms, focusing particularly on the systems operating within the adrenocortical zona glomerulosa cells; we also describe some of the other factors that influence aldosterone production to a lesser but still significant extent. Finally, we discuss the influence of common genetic polymorphisms on aldosterone secretion in large sections of the population and also the emerging role of microRNA as significant regulators of this system.

Abstract 143: Exosomal miRNA Profiling Following Ischaemic Stroke: Clinical and Pre-clinical Data

Introduction: MicroRNA (miRNA) profiles are altered in patients with stroke. Recently, active inter-cellular transport of miRNAs in extracellular vesicles (EV), such as exosomes has been shown. We assessed whether miRNAs packaged in EVs differed in patients with stroke and patients without stroke and then performed in vitro and in vivo experiments to assess potential significance of dysregulated miRNAs.

Methods: We recruited patients with suspected stroke. Serum was obtained at 48h post-stroke. EVs were isolated from 200μL serum before RNA extraction. A miRNA microarray was performed (Openarray™ platform) (39 patients) to identify significantly dysregulated miRNAs in stroke vs non-stroke patients. Results were validated by Taqman™ real time quantitative polymerase chain reaction of specific miRNA (17 miRNAs, 169 patients). I n vitro studies of selected miRNAs were performed on a rat neuronal cell line (B50) and a rat glial cell line (B92) subjected to 9h hypoxia (1% O 2 ) and serum starvation with 24h reoxygenation in complete media (n=4-5/group). In vivo , total miRNA expression was quantified in serum (n=4) and in peri-infarct tissue of adult spontaneously hypertensive stroke-prone rats (SHRSP) following 45min transient middle cerebral artery occlusion (tMCAO) (n=6/group).

Results: The microarray identified 26 dysregulated miRNAs in stroke patients. Using Taqman™ levels of miRNAs 17, 20b and 93 were observed to be significantly (p<0.05) increased in stroke vs non-stroke patients. Patients with small vessel disease consistently had the highest levels. In vitro , a trend towards increased exosomal expression of miRNAs-17, 93 and 20b in B50 hypoxic cells was in contrast to small changes in total cell miRNA expression (vs. normoxic cells). In vivo , tMCAO did not induce significant modulation of miRNAs 17, 93 or 20b expression in SHRSP brain either 24h or 72h post-tMCAO. EVs isolated from the serum showed a significant increase in miRNA-93 expression at 72h post-tMCAO (p<0.05 vs. sham animals).

Conclusions: We have identified and validated changes in EV packaged miRNA expression in patients with stroke, particularly in small vessel disease. Our pre-clinical experiments corroborate the human data and support a functional role for these findings.

Positive impact of pre-stroke surgery on survival following transient focal ischemia in hypertensive rats

We describe a positive influence of pre-stroke surgery on recovery and survival in a commonly used experimental stroke model. Two groups of male, stroke-prone spontaneously hypertensive rats (SHRSPs) underwent transient middle cerebral artery occlusion (tMCAO). Group 1 underwent the procedure without any prior intervention whilst group 2 had an additional general anaesthetic 6 days prior to tMCAO for a cranial burrhole and durotomy. Post-stroke recovery was assessed using a 32 point neurological deficit score and tapered beam walk and infarct volume determined from haematoxylin-eosin stained sections. In group 2 survival was 92% (n=12) versus 67% in group 1 (n=18). In addition, post-tMCAO associated weight loss was significantly reduced in group 2. There was no significant difference between the two groups in experimental outcomes: infarct volume (Group 1 317±18.6 mm³ versus Group 2 332±20.4 mm³), and serial (day 0-14 post-tMCAO) neurological deficit scores and tapered-beam walk test. Drilling a cranial burrhole under general anaesthesia prior to tMCAO in SHRSP reduced mortality and gave rise to infarct volumes and neurological deficits similar to those recorded in surviving Group 1 animals. This methodological refinement has significant implications for animal welfare and group sizes required for intervention studies.

Galanin-immunoreactivity identifies a distinct population of inhibitory interneurons in laminae I-III of the rat spinal cord

Background

Inhibitory interneurons constitute 30-40% of neurons in laminae I-III and have an important anti-nociceptive role. However, because of the difficulty in classifying them we know little about their organisation. Previous studies have identified 3 non-overlapping groups of inhibitory interneuron, which contain neuropeptide Y (NPY), neuronal nitric oxide synthase (nNOS) or parvalbumin, and have shown that these differ in postsynaptic targets. Some inhibitory interneurons contain galanin and the first aim of this study was to determine whether these form a different population from those containing NPY, nNOS or parvalbumin. We also estimated the proportion of neurons and GABAergic axons that contain galanin in laminae I-III.

Results

Galanin cells were concentrated in laminae I-IIo, with few in laminae IIi-III. Galanin showed minimal co-localisation with NPY, nNOS or parvalbumin in laminae I-II, but most galanin-containing cells in lamina III were nNOS-positive. Galanin cells constituted ~7%, 3% and 2% of all neurons in laminae I, II and III, and we estimate that this corresponds to 26%, 10% and 5% of the GABAergic neurons in these laminae. However, galanin was only found in ~6% of GABAergic boutons in laminae I-IIo, and ~1% of those in laminae IIi-III.

Conclusions

These results show that galanin, NPY, nNOS and parvalbumin can be used to define four distinct neurochemical populations of inhibitory interneurons. Together with results of a recent study, they suggest that the galanin and NPY populations account for around half of the inhibitory interneurons in lamina I and a quarter of those in lamina II.