Abstract

Background

   Atrial fibrillation (AF) is one of the most prevalent causes of
   cryptogenic stroke. Also, apart from AF itself, structural and
   remodelling changes in the atria might be an underlying cause of
   cryptogenic stroke. We aimed to discover circulating proteins and
   reveal pathways altered in AF and atrial cardiomyopathy, measured by
   left atrial volume index (LAVI) and peak atrial longitudinal strain
   (PALS), in patients with cryptogenic stroke.

Methods

   An aptamer array (including 1310 proteins) was measured in the blood of
   20 cryptogenic stroke patients monitored during 28 days with a Holter
   device as a case-control study of the Crypto-AF cohort. Protein levels
   were compared between patients with (n = 10) and without AF (n = 10)
   after stroke, and the best candidates were tested in 111 patients from
   the same cohort (44 patients with AF and 67 without AF). In addition,
   in the first 20 patients, proteins were explored according to PALS and
   LAVI values.

Results

   Forty-six proteins were differentially expressed in AF cases. Of those,
   four proteins were tested in a larger sample size. Only DPP7,
   presenting lower levels in AF patients, was further validated.
   Fifty-seven proteins correlated with LAVI, and 270 correlated with
   PALS. NT-proBNP was common in all the discovery analyses performed.
   Interestingly, many proteins and pathways were altered in patients with
   low PALS.

Conclusions

   Multiple proteins and pathways related to AF and atrial cardiomyopathy
   have been revealed. The role of DPP7 as a biomarker for stroke
   aetiology should be further explored. Moreover, the present study may
   be considered hypothesis-generating.

   Keywords: Atrial fibrillation, Atrial cardiomyopathy, Biomarkers,
   Cryptogenic stroke, Atrial function

1. Introduction

   Atrial fibrillation (AF) is a prevalent cardiac rhythm disorder
   underlying up to one-third of all ischemic strokes [53][1]. Paroxysmal
   AF remains undetected in a high proportion of patients after stroke and
   is one of the most prevalent causes of cryptogenic stroke [54][2].

   Also, there is increasing recognition that atrial dysfunction itself is
   associated with an increased risk of thromboembolism, even in patients
   without AF [55][3]. Therefore, atrial substrate or atrial
   cardiomyopathy has been proposed as an important cause of cryptogenic
   strokes [56][4], [57][5].

   Left atria (LA) enlargement and atrial fibrosis are two structural
   hallmarks of the atrial substrate [58][3]. LA size has been associated
   with cardioembolic stroke and AF detection in patients with embolic
   stroke of undetermined source (ESUS) [59][6]. Similarly, peak atrial
   longitudinal strain (PALS), which measures the LA wall deformability
   and is a surrogate of LA fibrosis, has been associated with AF in
   cryptogenic stroke patients [60][7].

   Some blood biomarkers (e.g. natriuretic peptides) have been proposed as
   useful tools to detect paroxysmal AF [61][8]. In addition, circulating
   markers might allow noninvasive assessment of atrial cardiomyopathy
   before AF appears and might guide the selection of patients for more
   intensive post-stroke monitoring to personalize the secondary
   prevention treatments [62][2], [63][5].

   The present study aims to discover circulating proteins and reveal
   pathways altered in AF and atrial cardiomyopathy, measured by left
   atrial volume index (LAVI) and PALS, in patients with cryptogenic
   stroke.

2. Methods

2.1. Study population

   The study population represented a subpopulation of the Crypto-AF study
   [64][9], [65][10]. Non-lacunar acute ischemic stroke patients over
   55 years of age with cryptogenic stroke after standard evaluation were
   included in the study by four Spanish public Stroke Centers from
   January 2015 to July 2017. All patients included had no prior history
   of AF. Patients were monitored for 28 days with a wearable Holter
   device (Nuubo^TM) following a published protocol [66][9]. The software
   algorithm classified every episode of irregular ECG rhythm
   lasting > 120 s as possible AF. An expert cardiologist blinded to
   clinical data verified the episodes. From this cohort, a total of ten
   consecutive patients with AF detected during the monitoring period, and
   ten matched controls (by sex and age) without AF were selected for the
   present discovery study. In addition, 111 patients with available blood
   samples were used for the validation study (44 patients with AF and 67
   matched controls).

   Blood samples were collected into EDTA and serum separator tubes within
   72 h after symptoms onset. After centrifugation at 1500 g and 4 °C for
   15 min, plasma and serum aliquots were frozen at −80 °C until further
   analysis.

   Left atrial size was measured by biplane transthoracic echocardiography
   to obtain the left atrial volume adjusted to body surface index (LAVI,
   ml/m^2) following the latest guidelines [67][11]. Peak atrial
   longitudinal strain (PALS) was evaluated by speckle tracking software
   (GE EchoPAC®) following expert recommendations [68][12].

   Written informed consent was obtained from all participants, and the
   study was approved (PR (AG)49/2014) by the Ethical Committee of Vall
   d’Hebrón Hospital, Valladolid Hospital, Virgen Macarena Hospital and
   Virgen del Rocio Hospital, in line with Helsinki guidelines.

2.2. Aptamer array

   Protein levels in plasma were assessed using the SOMAscan® platform
   (SomaLogic Inc., Boulder, CO, USA), which is an aptamer-based proteomic
   assay that allowed the simultaneous measurement and quantification of
   1310 proteins [69][13]. This approach uses SOMAmers® reagents, which
   are short single-stranded DNA sequences with protein affinity. The
   platform transforms the proteins present in the biological sample into
   a corresponding SOMAmer signal, which then is quantified using the
   microarrays technology. Three different dilutions (depending on each
   protein abundance) were used. Normalization and calibration procedures
   were performed by SomaLogic according to their protocol [70][14]. All
   samples passed SomaLogic quality controls. A set of control calibrator
   samples were used to detect and remove systematic variability between
   independent assay runs. Seventy-nine proteins were marked as “flags”
   due to high inter-plate variability and eliminated from the analysis.
   Data were reported in relative fluorescent units (RFU) after
   normalization and calibration.

2.3. Elisa

   Serum coiled-coil domain-containing protein 80 (CCDC80)(BosterBio), and
   plasma dipeptidyl peptidase 7 (DPP7)(R&D Systems), bone morphogenetic
   protein 1 (BMP-1)(Elabscience), and cystatin-D (BosterBio) were
   determined by ELISA. All assays were performed blinded to clinical
   information and according to the manufacturer’s instructions. All
   samples were tested in duplicate, and inter-assay variation was
   determined by a commercial control (Human Serum, male AB, USA origin
   from clotted, SIGMA, ref number [71]H16914; Human plasma K2 EDTA,
   Innovative Research, ref number IPLA-N) tested in duplicate in each
   plate. When inter-assay variation was > 20%, biomarker levels were
   standardized by the common control sample. Samples with a CV
   (coefficient of variation) > 20% between duplicates were eliminated
   from the analysis.

2.4. Statistics

   R software version 3.6.1 and SPSS version 20 were used to conduct
   statistical analysis. Categorical variables were expressed as numbers
   and percentages and continuous variables as mean ± SD, or median
   (interquartile range) for continuous variables, depending on their
   distribution. Student’s t-test, Mann–Whitney or x^2 were used to
   compare variables between AF cases and controls depending on the type
   and distribution of each variable.

   SOMAscan data were log-transformed as presented a skewed distribution.
   Differential expression analyses were performed using the “limma”
   package (Bioconductor) version 3.42.2, optimized for omics studies with
   large amounts of data and few samples [72][15].

   Spearman correlations were calculated between LAVI or PALS and all the
   analyzed proteins. The R package “Venndiagram” version 1.6.20 was used
   to visualize the common proteins between the different analyses. All
   p-values were adjusted using Benjamini and Hochberg (BH) false
   discovery rate (FDR). Group matching by sex and age was used to select
   control samples in the discovery experiment. The validation sample size
   was estimated based on Somascan results (power of 80%, α = 0.05) (Ene
   3.0, GlaxoSmithKline, UK).

   The addition of DPP7 to a logistic regression model fitted by age, sex,
   echocardiographic markers (LAVI and PALS), and NT-proBNP was tested
   using the Likelihood Ratio Test. Odds ratios (OR) for an increment of
   one unit of concentration were shown. The classification performance of
   the models was compared using Reciever Operating Curves. The R package
   “ggeffects” version 1.1.1 was used to plot the average predicted
   probability of the model when varying the variable of interest.

2.5. Pathway analysis

   Pathway enrichment analysis was conducted following a published
   protocol [73][16]. Gene Set Enrichment Analysis (GSEA) software was
   applied to all the SOMAscan proteins ordered by T-statistic or
   correlation coefficient against Reactome Pathways and Gene Ontology
   (biological processes) databases. Gene sets with < 15 genes or > 200
   genes were excluded. GSEA calculates a normalized enrichment score
   (NES) for each gene set. Positive and negative NES values represent
   enrichment of the corresponding gene set at the top (i.e., upregulated)
   or bottom (i.e., downregulated) of the ranked list. P-values were
   computed by gene set permutation for 1000. Then, multiple testing using
   a false-discovery rate (FDR) was applied to obtain the Q-values.
   Results were visualized via Cytoscape Enrichment Map with a Jaccard
   Overlap Combined Coefficient > 0.375. Significant pathways were
   considered at Q-value < 0.25.

3. Results

   The descriptive characteristics of the 20 patients included in the
   discovery study are provided in [74]Table 1. The median age was 71.5,
   and 55% were women. Clinical variables were similar between the two
   groups.

Table 1.

   Clinical characteristics of the patients included in the discovery
   experiment and comparison according to atrial fibrillation detection.
   All (n = 20) AF (n = 10) No AF (n = 10) p-value
   Sex (%female) 11 (55%) 6 (60%) 5 (50%) 0.65[75]^&
   Age (years) 71.5 (67–80) 73.5 (69.75–80) 67.5 (62.25–81.5) 0.247[76]^$
   Hypertension 14 (70%) 7 (70%) 7 (70%) 1.00[77]^&
   Diabetes 5 (25%) 3 (30%) 2 (20%) 1.00[78]^&
   Vasculopathy 1 (5%) 0 (0%) 1 (10%) 1.00[79]^&
   Renal failure 1 (5.3%) 1 (11.1%) 0 (0%) 0.47[80]^&
   COPD 1 (5.3%) 1 (11.1%) 0 (0%) 0.47[81]^&
   Obesity 8 (40%) 5 (50%) 3 (30%) 0.65[82]^&
   Heart disease 2 (10%) 2 (20%) 0 (0%) 0.47[83]^&
   Basal NIHSS 4 (2–7) 3 (1–7) 5 (3–7) 0.29[84]^$
   LVEF (%) 64.74 ± 34.19 63 ± 7.84 66.30 ± 7.51 0.362[85]^#
   PALS (%) 25.76 ± 12.98 29.89 ± 14.06 20.59 ± 10.00 0.134[86]^#
   LAVI (ml/m^2) 31 (27–37) 30 (27–34) 34 (22.5–37.75) 0.815[87]^$
   Number of AF episodes 28 (7–42.5)
   Longest AF episode (min) 780.48 (121.25–1510.71)
   [88]Open in a new tab

   COPD, chronic obstructive pulmonary disease; NIHSS, National Institutes
   of Health Stroke Scale; LVEF, left ventricular ejection fraction; PALS,
   peak atrial longitudinal strain; LAVI, left atrial volume index.

   The “heart disease” terminology included any cardiopathy that the
   investigator considered of interest, including into this category
   ischemic cardiopathy, and hypertensive cardiopathy, between others. The
   two patients with heart disease in this cohort had a mild mitral and
   aortic valvulopathy, and a hypertensive cardiomyopathy respectively.
   ^#

   Student’s t-test.
   ^$

   Mann–Whitney test.
   ^&

   x^2 test.

3.1. Differential protein expression and altered pathways in AF

   Among the tested proteins, 46 were differentially expressed in AF cases
   at a nominal p-value of 0.05 (22 down-regulated and 24 up-regulated).
   Although no protein remained significant after multiple comparison
   correction, NT-proBNP showed the strongest association
   (p-value = 0.001, logFC = 1.86) and BNP was ranked sixth
   (p-value = 0.008, logFC = 0.39). Both natriuretic peptides were
   well-known biomarkers of AF, already validated in this cohort in a
   published study [89][8]. Proteins with p-values between NT-proBNP and
   BNP were selected to evaluate their usefulness in a larger group of
   patients: CCDC80 (p-value = 0.0013), DPP7 (p-value = 0.0039), BMP-1
   (p-value = 0.0051), and Cystatin-D (p-value = 0.0080) ([90]Fig. 1 and
   Supplemental Table 1).

Fig. 1.

   [91]Fig. 1
   [92]Open in a new tab

   Volcano plot of differentially expressed proteins between AF cases and
   no AF. Black dots above the red line indicate significant proteins,
   while grey dots below the red line indicate non-significant proteins
   according to nominal p-value < 0.05. Labeled proteins are those with a
   nominal p-value < 0.01 or nominal p-value < 0.05 and |logFC|>1.Proteins
   with positive logFC had higher levels in the AF group and vice-versa.
   (For interpretation of the references to colour in this figure legend,