Emails: Mohitul Hossain - mdmohitul.hossain@postgrad.curtin.edu.au and James Hane - james.hane@curtin.edu.au 

Pathogen-secreted effectors mediate fungal crop diseases and accurate identification of effectors can help develop disease resistance cultivars. However, in fungi, accurately predicting effectors is complicated because they generally lack sequence conservation.

In the last decade, predictive tools based on the physicochemical properties of effectors have significantly improved in accuracy, but a gap remains between the number of predictions and what can be validated experimentally, so obtaining a reliably reduced candidate list is important. Effector prediction tools have not yet incorporated disease phenotype data that may identify candidates directly associated with host-specific disease outcomes.

Here we propose a method for further refinement of effector candidates - “EffectorFisher” – which mines low-cost pan-genome survey data and integrates disease phenotyping with established bioinformatic methods. EffectorFisher was benchmarked on the Parastagonospora nodorum-wheat pathosystem, for which there are 5 known effector loci (ToxA, Tox1, Tox3, Tox5, Tox267). EffectorFisher significantly improved effector prediction, demonstrated by improved ranking of known effectors by 2 to 13 times compared to previous methods and also reduced total candidate numbers by ~3.5-fold.

To further investigate potential applications of EffectorFisher, a minimum-viable experimental design was assessed with simulated datasets with reduced numbers of pathogen isolates (20-100) and/or phenotyped cultivars (2-10), which generated comparable results for known effectors for upwards of approximately ~40 isolates and ~4 cultivars.

Preliminary application of EffectorFisher to the Zymoseptoria tritici-wheat interaction (versus 6 known effectors) has yielded a similar trend of improved known effector ranking and candidate reduction. Furthermore, numerous past pan-genomic surveys of fungal pathogens typically lack quantitative phenotyping panels, yet have recorded basic cultivar metadata.

To make use of such legacy datasets, we incorporated non-quantitative cultivar data into EffectorFisher and demonstrated similar outcomes, which allows for its broader application.

EffectorFisher generates isolate-specific effector isoform profiles, which can be used to predict which effector (or candidate) isoforms confer higher or lower virulence against specific cultivars. Our new capability for effector isoform profiling has led to new development of methods to match surveyed isolate-specific isoform fingerprints to disease phenotype and yield databases, with a goal of optimal paddock-specific selection of cultivars for both yield and disease resistance.

Overall, EffectorFisher demonstrates that incorporating phenotypic data can significantly improve the effector prediction process and has led to the development of new capabilities for management of biotic stress and cultivar selection. So far EffectorFisher has been applied to two wheat pathogens - a necrotroph and hemibiotroph - but is broadly applicable to many other patho-systems.