Rapid and precise identification of microorganisms serves as the cornerstone for exploring and harnessing their functional potential. The genus Pantoea holds considerable promise for applications in agriculture and environmental contexts. However, due to its high interspecies genetic similarity and pronounced morphological uniformity, conventional identification methods are time-consuming and low-throughput, which severely impedes the development and utilization of this genus.

Recently, the Innovation Team of Agricultural Microbial Resources at the Institute of Agricultural Resources and Regional Planning (IARRP), Chinese Academy of Agricultural Sciences, in collaboration with multiple institutions, innovatively integrated positive dielectrophoresis-activated Raman-activated cell sorting (pDEP-RACS) technology with a deep residual network (ResNet) algorithm. They constructed a single-cell Raman spectral database and a high-precision classification model specifically for the genus Pantoea. For the first time, this study achieved rapid and accurate identification of this taxonomically challenging group at both the species and strain levels. It provides a new technical solution for the in situ rapid identification of microorganisms and the screening of functional strains, while also establishing a methodological foundation for the phenotypic analysis of complex microbial communities. The findings have been published in the internationally top-tier journal Analytical Chemistry.

(1) High-throughput single-cell phenotyping acquisition: Using a Raman flow cytometry sorting system (FlowRACS), biochemical fingerprint spectra of individual cells can be obtained directly from samples. This system collects over 7,200 high-quality single-cell Raman spectra per hour, thereby overcoming the throughput and resolution limitations of conventional Raman techniques.

(2) High-accuracy species identification model: Based on a total of 180,000 spectra derived from 12 Pantoea species (22 strains) and two closely related species, the research team developed a ResNet-18 deep learning model, achieving an average accuracy of 96.9% and a recall rate of 97.3%. Experimental investigation of the database depth threshold revealed that when the number of single-cell spectra exceeds 1,500, the classification accuracy can reach 97.6% ± 2.0%.

(3) Validation and application in complex communities: In artificially constructed microbial communities, the platform predicted species abundance with an absolute error of ≤3.21%. During in situ detection of the endophytic microbiome of rice seeds, the platform successfully identified Pantoea with a relative abundance of 34.8% and further distinguished key species such as Pantoea agglomerans and Pantoea ananatis, achieving a breakthrough from genus-level to species-level identification. The single-cell chemotaxonomic results complemented those obtained by 16S rRNA gene sequencing (relative abundance 45%) while better reflecting metabolically active cellular states, and even enabled strain-level resolution. This study provides a valuable reference pathway for the identification of other "difficult-to-culture" and "difficult-to-classify" microorganisms.

Zhang Xiaoxia, a researcher at the Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, and Wang Haisheng from the Graduate School of the Chinese Academy of Agricultural Sciences, served as co-corresponding authors. Zhang Daoshun, a PhD candidate at the same institute, is the first author. The study was jointly completed with multiple institutions, including Xingsai Biotechnology and the Qingdao Institute of Bioenergy and Bioprocess Technology, Chinese Academy of Sciences. It was supported by the National Key Laboratory for Efficient Utilization of Arable Land in Northern Arid and Semi-Arid Regions, the National Natural Science Foundation of China, and the CAAS Science and Technology Innovation Program.

Original article: https://doi.org/10.1021/acs.analchem.6c00276