The recent discovery that Aspergillus niger can produce fumonisins raises a new mycotoxigenic risk that requires additional studies to assess the impact on agricultural commodities. Here, we report the development and validation of a sensitive LC-fluorescence method for the identification of fumonisin B2 in black aspergilli cultures. The performances of naphthalene-2,3-dicarboxaldehyde (NDA), o-phthaldialdehyde (OPA), and 9-fluorenylmethyl chloroformate (FMOC) for the precolumn derivatization of samples were evaluated and compared. The best methodology included the following elements: NDA was used during pre-column derivatization; the derivatives were separated by a 30 min isocratic elution with acetonitrile/H2O/acetic acid (60:40:1, v/v/v), pumped at a flow rate of 1 mL · min−1 on a polar C18 reversed-phase YMC-Pack ODS-AQ column (250 × 4.6 mm, 5 µm); and fluorescence detection using excitation and emission wavelengths of 420 and 500 nm, respectively. The calibration curve was found to give a good linear response (r2 > 0.9999) from 1 to 200 µg · L−1. For FB2, which is the main fumonisin detected in black aspergilli thus far, this method measured an LOD and an LOQ of 2 and 6 µg · kg−1, respectively. Using this method, the amount of FB2 in black aspergilli cultures can be simply and rapidly determined.
Enzyme engineering is central to developing biocatalysts with improved activity and specificity, yet traditional approaches are often limited by the scale of experimental screening. Here we combine ultrahigh-throughput microfluidic droplet screening with generative machine learning to engineer specificity into an unspecific peroxygenase (UPO) from Aspergillus brasiliensis (AbrUPO). A library of 5e6 variants was expressed in Komagataella phaffii ( Pichia pastoris ) and screened for increased production of styrene oxide and reduced production of phenylacetaldehyde and phenylpropanol. From an ultrahigh throughput screening dataset of >30,000 unique sequences paired with function data. These data were used both for direct selection of enriched variants and to train a task-specific generative long short-term memory (LSTM) model using the Variational Search Distributions (VSD) framework. We compared variants selected by rank aggregation from the screening data (R series) with novel sequences generated by the refined generative model (G series). Experimental assays revealed that multiple G variants outperformed both wildtype AbrUPO and R variants, in terms of styrene oxide production and specificity. Correlation analyses further showed that the task-specific generative model outperformed protein models pretrained on large publicly available datasets in predicting experimental outcomes. Our results demonstrate that coupling ultrahigh throughput screening with generative protein models enables efficient discovery of improved enzyme variants beyond those identified by direct screening alone, providing a scalable strategy for task-specific enzyme engineering. Graphical
Ultrahigh throughput screening of an enzyme library provides a dense dataset to train a generative machine learning model. Novel enzymes designed by this model can outperform the best enzymes in the screened library.