The BrAID study protocol: integration of machine learning and transcriptomics for brugada syndrome recognition

Brugada Syndrome (BrS) is a hereditary arrhythmogenic disease with an electrocardiographic (ECG) pattern characterized by ST-segment elevation in the right precordial leads and burdened by the risk of sudden cardiac death [1]. The prevalence of the syndrome is estimated to be 15 per 10,000 in South East Asia, including Japan, and 2 per 10,000 in the Western countries. BrS has been considered responsible for 4–12% of all sudden deaths and up to 20% of sudden deaths in patients with structurally normal hearts. It is 8 to 10 times more prevalent in men than in women. In a recent meta-analysis, the incidence of arrhythmic events (sustained VT or VF or appropriate implantable cardioverter defibrillator (ICD) treatment or sudden death) in patients with Brugada syndrome was 13.5% per year in patients with a history of sudden cardiac arrest, 3.2% per year in patients with syncope and 1% per year in asymptomatic patients [2]. As more data become available, the currently accepted percentages of Sudden Cardiac Death (SCD) due to BrS need to be updated to establish the real incidence of the syndrome in unexpected deaths in different populations [3].

In these last years, the research interest on BrS has significantly increased since it prevalently affects healthy young adults during their most productive years [1]. Due to its genetic component, there is a need to identify other potentially affected relatives, focusing on children [4].

There are three types of ECG changes associated with Brugada Syndrome. Type 1 shows, in more than one right precordial lead, a coved shape of the ST-segment with J wave or ST-elevation of ≥ 2 mm (mV) at its peak followed by a negative T wave with little or no isoelectric interval; in Type 2, the ST segments have a high take-off but the J amplitude of ≥ 2 mV rises to a gradually descending ST-segment remaining ≥ 1 mV above the baseline followed by a positive or biphasic T wave; these abnormalities result in a saddle-back configuration. Finally, Type 3 presents ST elevation on the right precordial leads < 1 mm (saddle-back type or coved morphology).

From a diagnostic point of view, BrS can be confirmed in non-diagnostic cases by developing the peculiar ECG patterns in V1 and V2 after administration of Na + channel blockers (i.e., ajmaline, flecainide) [5].

In addition to recognizing specific ECG patterns, genetic and environmental modulators of BrS have been identified, which play a significant role in the dynamic nature of BrS patients' ECG. In 1998 the first genetic mutation in the SCN5A (sodium voltage-gated channel alpha subunit 5) gene was associated with BrS within a family [6]. SCN5A is responsible for phase 0 of the cardiac action potential, and pathogenic variations result in the sodium channel inability to function properly. Since the identification of the first gene associated with BrS, reports of other families affected by BrS have confirmed that the disease genetic origin follows an autosomal dominant pattern of inheritance. More than 250 pathogenic variations associated with BrS have been described in 18 different genes, primarily encoded for sodium, potassium, calcium channels, or proteins associated with these channels [7]. Pathogenic variations in genes encoding desmosomal proteins have also been associated with BrS [8, 9]. Despite these ongoing developments in understanding the genetic causes of BrS, only 30–35% of clinically established cases are genetically diagnosed, and most of these (25–30%) result from pathogenic alterations in SCN5A [5]. The remaining cases may be attributable to alterations in one of the other BrS-associated genes [10]. A recent study [11] concluded that BrS ECG pattern appears not to be a pure Mendelian disorder, but rather the result of different molecular pathologies. Thus, genetic screening results do not currently influence prognosis or treatment [2], since they take into account only a part of the reported cases and variants previously classified as pathogenic may be of ambiguous significance following recent guidelines of the American College of Medical Genetics [12].

Due to disease heterogenicity in terms of genetic and phenotypic features, the use of new approaches as omics (genomics, transcriptomics, proteomics) might assume an important role in understanding the molecular mechanisms and identification of diagnostic/prognostic markers of BrS. These high-throughput technologies produce a considerable amount of data on analyzed genes, RNAs, and proteins, providing a more thorough picture of the different molecular events involved in the disease pathogenesis. In particular, omics technologies can identify new molecular mechanisms through the in-depth analysis of biological (or biochemical) pathways. This information can be extracted to differentiate BrS from similar phenocopies and develop new and reliable biomarkers for disease detection [13]. Among all techniques, the analysis of RNA profile trough transcriptomics has shown its importance in the study of several diseases [14]. This group of molecules is close to patient phenotype and can provide information on patients' condition that standard biochemical analysis or DNA variants cannot detect [15]. RNAs and their associated pathways could represent the bridge between environmental factors and genotype [16], showing the dynamic interaction during the patient's life. Among RNA molecules sub-groups, a prominent role is assumed by non-coding RNAs, particularly the micro-RNAs (miRNAs). These short coding RNA sequences can be found in the so-called microvesicles, a class of membrane vesicles ranging in size between 100 and 1000 nm in diameter, generated by outward budding or blebbing of the plasma membrane [17] during various processes and stimuli [18]. miRNAs modulate gene function by binding with specific sequences of target genes [15]. While transcription factors play a critical role in controlling ion channels/transporters at the transcriptional level, miRNAs are paramount for expression regulation at the post-transcriptional level. These small non-coding RNAs finely regulate the expression of genes. In cardiovascular disease, miRNAs analysis has assumed a leading role in several cardiovascular pathologies [19, 20].

Challenges and actual controversies do exist as far as BrS diagnosis is concerned. Although at a clinical level BrS presents specific ECG features and these criteria have a wide consensus as underlined by present guidelines [21], difficulties exist on ECG interpretation due to the extreme variability of ECG patterns that may change with time and during situations as sleep, fever and vagal stimulation [4, 5]. In contrast to diagnostic criteria for the long QT syndrome, one of the main difficulties in BrS diagnosis is the absence of exact cut-off values to base a clear-cut diagnosis.

The interpretation of these ECG patterns may vary between clinicians since different pathologies can be associated with ST-segment elevation in the right precordial leads (Table 1) [22].

Even if specific ECG patterns are described in scientific literature, they may vary in the same individual, making the diagnosis problematic [23]. The very low prevalence of the disease associated with the high observer inter-variability in ECG interpretation and the absence of univocal "genetic makers" may lead to a reduced recognition of BrS in the population, potentially increasing risks, including sudden death.

The BrAID (Brugada syndrome and Artificial Intelligence applications to Diagnosis) project aims to integrate classic clinical guidelines for BrS diagnosis evaluation with transcriptomics approach and innovative ML algorithms for ECG pattern recognition: the combined information should lead to new diagnostic strategies in cardiovascular precision medicine of this disease.

Difficulties still hamper the recognition of Brugada syndrome: first of all, the approach to asymptomatic patients since roughly 30% of the sudden death may represent the first manifestation of the disease [36]; moreover, ECG is by itself challenging due to temporal variability of the pattern and difficulties in selecting Br patients from other pathologies.

As far as disease characterization through biochemical markers, recently interesting results on autoantibodies against cardiac and skeletal alfa-actins, keratin-24, and connexin-43 in selected BrS patients have been published [37]. However, these results need to be confirmed in larger patient groups, especially in those with clear-cut Brugada pattern after administration of Na⁺ channel blockers.

Therefore, improvement in the ECG diagnosis of Brugada syndrome through artificial intelligence techniques, the definition of new biochemical markers from omics approach, and especially the combination of these markers with a well-tailored collection of clinical data may represent the key towards an unequivocal diagnosis in subjects without structural cardiac disease at risk of sudden cardiac death.

BrAID is a study intended to investigate the role of Machine Learning models to recognize specific ECG patterns associated with Brugada Syndrome, evaluate possible novel biomolecular markers of the syndrome, and integrate these data with the clinical ones in an advanced system for patient's risk stratification.

In particular, we will evaluate ML models and blood markers' power in recognition of Type 1 BrS syndrome patients with further advancement in recognition of spontaneous respect to drug-induced patients. The study will expand the current knowledge of Brugada Syndrome and will bring additional data on this disease and its clinical implications.