Individuals were selected consecutively from a prospective cohort of patients with CrS admitted to the stroke unit of Hospital del Mar (Barcelona, Spain) between October 2013 and September 2016 and meeting the following criteria6: a) ischemic stroke or transient ischemic attack; b) age between 50 and 89 years; c) undetermined origin according to the SSS-TOAST criteria.2 We excluded patients with a history of hemorrhagic stroke, prior AF or atrial flutter, permanent contraindication for oral anticoagulation, recent (< 1 month) major surgery or cardiac events, severe cardiac disease or comorbidities, short life expectancy (< 1 year), or severe stroke (modified Rankin Scale >4).6 The study followed national and international principles (Declaration of Helsinki) and was approved by the local ethics committee (2013/5055/I). All patients or next-of-kin were required to sign the specific informed consent form.

Blood was obtained from a peripheral line at the time of admission. Patients were immediately treated with antiplatelet therapy and transferred to the stroke unit under continuous electrocardiogram-monitoring. A 2-dimensional transthoracic echocardiogram was performed within the first 48hours with a General Electric Vivid E9 equipment (GE Healthcare, Chicago, IL, USA) to assess the dimensions and functionality of the left ventricle and left atrium. See further details in the supplementary data.

As per protocol, all patients with no established cause for the stroke underwent implantation of an insertable cardiac monitor when transferred to the general ward. Insertable cardiac monitor devices (Biomonitor and Biomonitor-2, Biotronik, Berlin, Germany) were programmed with the specific algorithm for AF detection, defined by an RR interval variability> 12.5% lasting> 1minute, as previously described.6 Recordings automatically detected as potential AF episodes were reviewed by a specialized cardiologist. AF burden was assessed by the percentage of monitoring time an individual was in AF, and secondarily by the number of AF episodes, as previously suggested.16 An AF time <0.05% defined participants with low AF burden, whereas AF burden was high when AF time was ≥ 0.05%.17 If AF was documented at any time, oral anticoagulation was initiated.

This was a case-control study in which patients were prospectively enrolled and allocated to cases (patients with AF, CrS-AF) or controls (patients in continuous SR, CrS-SR) after 6 months of monitoring, the final timepoint of the primary analysis. A secondary analysis was performed at 12 months of follow-up.

As commonly performed in gene expression studies,18 the study design followed a 2-step approach, including a discovery phase using high-throughput microarray data, with selection of potential candidate biomarkers, followed by a replication (and confirmatory) phase of candidate miRNAs in an independent cohort (figure 1). For the discovery phase, a total of 18 age- and sex-matched patients were selected (set I): 9 CrS-AF and 9 CrS-SR. All 9 patients in the 2 groups were followed up for 12 months to ensure that they remained equally classified either at 6 or 12 months (ie, no AF episodes occurred in the CrS-SR beyond 6 months), validating their eligibility for the study at 6 and 12 months. In parallel, an additional group of 9 patients (with no prior history of AF) presenting with AF at admission (cardioembolic stroke, CES-AF), also matched for the previous groups, was screened in this phase with the aim of further guiding the selection of miRNAs. For the replication phase, we used an independent cohort of 46 unmatched patients (set II), of which 19 had AF at 6 months. The same cohort was analyzed after 12 months of follow-up, when 23 had shown AF (figure 1).

Figure 1.

Study design and population. AF, atrial fibrillation; CES, cardioembolic stroke; CrS, cryptogenic stroke; ICM, insertable cardiac monitors; OA, Open Array; PCR, polymerase chain reaction; SR, sinus rhythm; TIA, transient ischemic attack.

(0.64MB).

MiRNAs were isolated from plasma using the miRNeasy Serum/Plasma kit (Qiagen Hilden, Germany). A TaqMan OpenArray Human Advanced MicroRNA Panel (ThermoFisher, Waltham, MA, USA) analyzing 754 human miRNAs was used in the discovery phase, and expression levels of all miRNAs were analyzed with the Cloud software (ThermoFisher). Those miRNAs found to have a significant differential expression among groups (P <.05 and fold change> 2 or <.5) were selected for replication, which was performed by quantitative polymerase chain reaction (qPCR). The results were analyzed using the Expression Suite software (Life Technologies). See further details in the supplementary data.

Clinical data are reported as mean±standard deviation or frequency (%). MiRNA expression results are expressed in fold change vs controls, where 1 equals the mean of expression of all controls and therefore represents the reference value.19 OpenArray analysis of differential expression was assessed with the specific Cloud Software (Life Technologies) using the global normalization method. Other comparisons were performed with the Student t test, 1-way ANOVA or chi-square analysis, as appropriate. Given the case-control study design, the association between variables and AF was evaluated using binary logistic regression models. Because there are no established reference values of miRNAs, the expression of the miRNAs of interest was stratified into quartiles or as above/below the median of all values obtained in our cohort. Those variables showing an association with a P <.1 in the univariable model were included in the multivariable analysis. The discriminatory ability of the model was assessed by the area under the receiver operating characteristic curve. Predicted probabilities were calculated at 6 months by generalized linear models using the variables of interest from the multivariable model and plotted according to them. Significance was set at P <.05. Statistical analyses were performed with SPSS Inc (version 26.0 for Windows, Chicago, IL, United States).

During the study period, a total of 530 patients aged 50 to 89 years were admitted with ischemic stroke. After a complete work-up, 74 patients with a CrS were eligible for the study, of which 10 were subsequently excluded. Therefore, the study population comprised 64 patients with a definite diagnosis of CrS under continuous monitoring with an insertable cardiac monitor implanted before discharge.

The clinical characteristics of the final study population are summarized in table 1. Mean age was 75±11 years, and 29 patients (45%) were women. The prevalence of cardiovascular risk factors was high. Eleven patients (17%) had had a prior stroke or transient ischemic attack. Cardiac evaluation during admission showed normal left ventricle and left atrium dimensions and function in the global population. Stroke characteristics, treatment and severity are presented in table 1.

Of the 64 patients, 28 had AF (25 paroxysmal, 3 persistent) within the first 6 months, constituting the CrS-AF group. Oral anticoagulation was initiated in all patients. Patients with AF had a median of 4.3 (0.6-2) AF episodes per month, with a median total AF time of 183.9minutes (23.7-775.6), representing 0.071% (0.009-0.30) of the monitoring time. The remaining 36 patients showed persistent SR (CrS-SR group). The characteristics of patients with AF and those in SR are summarized in table 1. Patients in the CrS-AF group tended to have higher ventricular dimensions and had higher atrial dimensions, together with lower left atrial ejection fraction compared with patients in the CrS-SR group.

The characteristics of the patients included in the discovery phase (9 patients in CrS-AF, 9 in CrS-SR, and 9 in CES-AF) are summarized in table 1 of the supplementary data. All groups were sex- and age-matched, with 4 women and a mean age of 75 years. Sixteen miRNAs were differentially expressed in CrS-AF vs CrS-SR (see complete list in table 2 of the supplementary data), of which 8 emerged as potential candidates based on their plasma stability. These are represented in figure 2A (shadowed area). All 8 candidates showed relatively consistent expression between CrS-AF and CES-AF compared with controls, and 5 of them (miR-744-5p, miR-1-3p, miR-377-3p, miR-425-5p, and miR-19b-3p) were also significantly different between CES-AF and CrS-SR patients, suggesting that these could be potential indicators of the presence of an AF-favoring environment, irrespective of whether the arrhythmia had been documented or remained subclinical. Expression values of the 8 candidate miRNAs in the Open Array assay are depicted in figure 2B.

Figure 2.

A: results of the Open Array study. Out of the 854 miRNAs screened in all 3 study groups (CrS-SR, CrS-AF, and CES-AF), 8 were differentially expressed between CrS-AF and CrS-SR (gray-shadowed area), of which 5 were also differentially expressed between CES-AF and CrS-SR (in bold). Overexpressed miRNAs are presented in blue, whereas downregulated miRNAs appear in orange. B: results of the 8 miRNAs that were selected for replication. Data are expressed in Rq (FC) vs the CrS-SR group (control), which was normalized to 1. Specific P values are provided in table 2 of the supplementary data. C: results of plasma miRNA expression in the replication cohort. miRNA-1-3p was the only candidate that was confirmed to be significantly increased in CrS-AF vs CrS-SR patients. AF, atrial fibrillation; CES, cardioembolic stroke; CrS, cryptogenic stroke; FC, fold change; SR, sinus rhythm.

(0.52MB).

All 8 candidate miRNAs were replicated in an independent cohort defined by the presence or absence of AF at 6 months (set 2, table 1 of the supplementary data). The results are shown in figure 2C. Only the plasma levels of miR-1-3p remained significantly increased in CrS-AF vs CrS-SR patients (2.81-fold increase, P=.014).

Using the global cohort (n=64), we further explored the potential association between miR-1-3p and AF. Since the “normal range” of plasma levels of miR-1-3p has not been, and fold change values do not provide real plasma levels but a value that is proportional to their concentration, the miR-1-3p results were stratified into quartiles (Q1-Q4). As shown in figure 3A, miR-1-3p levels showed a positive association with the risk of AF, where patients in Q4 and Q3 were more likely to show AF at 6 months of follow-up (69% and 56%, respectively) compared with those in the lower quartiles (31% in Q2 and 19% in Q1, P=.017).

Figure 3.

A: occurrence of AF according to quartiles of miR-1-3p in the global population. B: association between miR-1-3p values and AF burden. Percentage of time in AF (top right) and number of AF episodes per month (top left) are presented according to miR-1-3p quartiles. On the bottom panel, distribution of miR-1-3p quartiles among patients with no AF, low AF burden and high AF burden is depicted. AF, atrial fibrillation.

(0.44MB).

Moreover, miR-1-3p levels appeared moderately associated with AF burden in our cohort. figure 3B depicts the percentage of time in AF (top, right) and the number of AF episodes per month (top, left) that were recorded among patients in the 4 miR-1-3p quartiles throughout the 6 months of follow-up. AF burden measured by both parameters seemed greater among patients in higher miR-1-3p quartiles. The bottom panel of figure 3B represents the distribution of miR-1-3p quartiles among patients with no AF, low AF burden, and high AF burden. Notably, more than half of patients with high AF burden had miR-1-3p values in Q4, whereas none had miR-1-3p values in Q1.

We then evaluated whether miR-1-3p, in combination with clinical data, could provide early guidance in the identification of CrS patients with subclinical AF. For better applicability, we grouped all miR-1-3p values into above (Q3 + Q4) or below (Q1 + Q2) the median.

Table 2 summarizes the results of logistic regression analyses. Only 2 parameters remained independently associated with AF in the multivariable analysis: a) left atrial ejection fraction, associated with an 11% relative reduction in the AF risk per unit of increase; and b) high plasma levels of miR-1-3p, where values above the median were associated with an 8.5-fold increase in the risk of AF at 6 months. The model including both variables showed a discriminatory ability, evaluated by the area under the receiver operating characteristic curve, of 0.860 (0.760–0.960, P <.001). Figure 4 depicts the estimated predicted probabilities of AF documentation at 6 months after stroke according to these variables.

Figure 4.

Predicted probabilities of AF risk at 6 months in the global cohort according to LAEF and miR-1-3p levels [expressed as above (red) or below (blue) the median of all values]. Shadowed areas represent 95% confidence intervals. AF, atrial fibrillation; LAEF, left atrial ejection fraction.

(0.08MB).

To further evaluate the potential clinical impact of plasma miR-1-3p in the assessment of subclinical AF in patients with CrS, we investigated the association between miR-1-3p and the presence of AF at 12 months. Four additional patients developed AF between 6 and 12 months of follow-up (figure 1). miR-1-3p levels were significantly higher in patients showing AF at 12 months than in those with persistent SR (figure 5A). Moreover, miR-1-3p quartiles were positively associated with the presence of AF (figure 5B). Finally, by logistic regression analysis, both miR-1-3p above the median and left atrial ejection fraction remained independently associated with the presence of AF at 12 months (table 3).

Figure 5.

Association between miR-1-3p levels and AF at 12 months. A: results of plasma miRNA expression in the replication cohort, with 23 patients in SR and 23 having had at least 1 episode of AF. B: occurrence of AF at 12 months according to quartiles of miR-1-3p in the global population. AF, atrial fibrillation; CrS: cryptogenic stroke; FC, fold change; SR, sinus rhythm.

(0.12MB).

Our work has several limitations. The sample size was small, particularly in the replication cohort, but included an invaluable population of consecutive patients thoroughly evaluated under continuous monitoring. Even so, the final population was not significantly smaller than that used in previous studies on circulating miRNAs using high-throughput methodology.18 Our population included men and women aged between 50 and 89 years. Therefore, the present results should not be extrapolated to other age groups. This study was performed in a single tertiary hospital and lacks external validation in other cohorts, which would be most desirable to definitely establish the role of miR-1-3p as a biomarker of subclinical AF in patients with CrS. Finally, we cannot confirm that high circulating levels of miR-1-3p reflect high miR-1-3p expression at the atrial level, or the presence of an AF-favoring substrate. However, our results point to miR-1-3p as a potential biomarker of subclinical AF regardless its tissue expression or the underlying atrial substrate. Whether high plasma miR-1-3p levels may be directly responsible for AF promotion in the context of CrS, reach true clinical applicability, and even might provide guidance for early oral anticoagulation initiation should be evaluated after external validation in future studies.