Empagliflozin in heart failure with preserved ejection fraction: decoding its molecular mechanism of action using artificial intelligence

 

Rationale The use of sodium-glucose co-transporter 2 inhibitors (SGLT2i) to treat heart failure with preserved ejection fraction (HFpEF) is under investigation in ongoing clinical trials, but the exact mechanism of action is unclear. Here we aimed to use artificial intelligence (AI) to characterize the mechanism of action of empagliflozin in HFpEF at the molecular level.

Methods

We retrieved information regarding HFpEF pathophysiological motifs and differentially expressed genes/proteins, together with empagliflozin target information and bioflags, from specialized publicly available databases. Artificial neural networks and deep learning AI were used to model the molecular effects of empagliflozin in HFpEF.

Results

The model predicted that empagliflozin could reverse 59% of the protein alterations found in HFpEF. The effects of empagliflozin in HFpEF appeared to be predominantly mediated by inhibition of NHE1 (Na+/H+ exchanger 1), with SGLT2 playing a less prominent role. The elucidated molecular mechanism of action had an accuracy of 94%. Empagliflozin's pharmacological action mainly affected cardiomyocyte oxidative stress modulation, and greatly influenced cardiomyocyte stiffness, myocardial extracellular matrix remodelling, heart concentric hypertrophy, and systemic inflammation. Validation of these in silico data was performed in vivo in patients with HFpEF by measuring the declining plasma concentrations of NOS2, the NLPR3 inflammasome, and TGF-β1 during 12 months of empagliflozin treatment.

Conclusion

Using AI modelling, we identified that the main effect of empagliflozin in HFpEF treatment is exerted via NHE1 and is focused on cardiomyocyte oxidative stress modulation. These results support the potential use of empagliflozin in HFpEF.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Instituto de Salud Carlos IIICentro de investigaciόn biomédica en red cardiovascular (CIBERCV) Summary figureTable 1

MHC-environment interactions leading to type 1 diabetes: feasibility of an analysis of HLA DR-DQ alleles in relation to manifestation periods and dates of birth

AIM

The region on chromosome 6p21 (IDDM1) confers the largest part of genetic susceptibility to type 1 diabetes (T1D) with particular human leucocyte antigen (HLA) alleles predisposing and others protecting from it. As T1D is primarily a "sporadic" disease, the pathophysiology must involve gene-environment interactions. We searched for indirect evidence for such major histocompatibility complex (MHC)-environment interactions by asking two questions: (i) can the degree of an HLA association vary over time periods? and (ii) if a prenatal event like an intrauterine infection - that might cluster in seasons - leads to differences of HLA associations in patients with particular birth months?

METHODS

We screened the Type 1 Diabetes Genetics Consortium (T1DGC) database (in addition our own database and the original UK, US and SCAND databases) for MHC DR-DQ and CTLA4 associations. First, we separated the groups of patients with onset of disease before 1980 in comparison with onset after 1980. Second, we analysed the data according to dates of birth (grouped in months). Not all patients' dates of birth or manifestation periods were available, leading to different group sizes. There were 282 patients analysed for manifestation periods and 329 for birth month.

RESULTS

The cohorts of manifestation before 1980 demonstrated a significantly lower frequency of DQ2/X (2 vs. 14.2%; p = 0.03). There was a trend for DQ8/x to be more frequent for manifestations before 1980 (34 vs. 21.6%; p < 0.10). Other alleles did not differ significantly. The months of birth were not evenly distributed. Significant deviations from the whole group were seen in August (DQ2/8 trough and DQx/x high), whereas birth in September was more frequent in DQ8/x or DQ8/8 carriers. This pattern was significantly different from the expected distribution of months at birth (13.9 vs. 7.6%; p < 0.04).

CONCLUSIONS

We demonstrate the feasibility of an analysis that searches for indirect evidence of gene-environment interactions. These preliminary data need to be confirmed in larger data sets.