Speaker: Saiful Islam, graduate student in the lab of Kyaw (Joe) Aung, genetics, development and cell biology associate professor

Title: Using artificial intelligence to study plasmodesmata and plant immune responses

Abstract: Plasmodesmata (PD) are essential cytoplasmic channels that facilitate direct cell-to-cell communication in plants, playing critical roles in plant growth, development, and immunity. 

Despite their fundamental role, the molecular components of PD and their regulatory mechanisms remain understudied. My dissertation explores the use of AI to uncover novel functional aspects of PD and their role in plant immune responses. 

Identification of PD proteins is a major first step toward revealing the underlying mechanisms of those components. It has been reported that in Arabidopsis, plasmodesmata-located protein 5 (PDLP5) regulates callose accumulation at PD and is involved in plant immunity through an unknown mechanism. I developed PLASID (Proximity Labeling and Artificial Intelligence-Guided Structural Interactome Discovery), a novel experimental and AI-guided framework to identify proteins that interact with PDLP5. 

I developed a novel membrane topology-aware Python package, Membrane Protein Interactors (MNPInteract), to narrow down 27 proteins to the top-ranked PDLP5 interactors. We provided empirical evidence showing that PDLP5 mediates homodimerization via its ectodomains. 

I also characterized transmembrane kinase 4 (TMK4) as a negative regulator of PD callose deposition and innate plant immunity. Next, my dissertation presents our novel findings that PDLP5 physically interacts with plasma membrane intrinsic proteins (PIPs, aquaporins) and negatively regulates PIP-mediated hydrogen peroxide transport in Arabidopsis. AlphaFold3 predicted that PDLP5 ectodomain engages the extracellular face of PIP channels, and genetic evidence identified PIP2;5 as a key element in this regulation. 

These findings reveal that PDLP5 associates with PIPs to maintain redox homeostasis between the apoplast and cytosol. To evaluate bacterial colonization and plant disease severity in infected Arabidopsis thaliana leaves, I developed Ebio-Net, a deep learning framework for ordinal classification of bacterial colonization levels in Arabidopsis using bioluminescence imaging. Trained on six disease severity classes, Ebio-Net achieves 90% accuracy, providing finer-resolution of genotype-specific disease trends. 

Collectively, this work demonstrated the use of artificial intelligence to advance our understanding of plasmodesmal proteins, their role in redox signaling, and plant immunity.