Extending digital biology: bacterial survival and morphological heterogeneity under antibiotic stress

The morphology of bacteria is intimately linked to their biological state: It is modified by antibiotic stress while also serving to survive the antibiotic. As such knowing which morphologies are selected by the surviving cells would help explain how individual cells are able to escape the antibiotic pressure. However associating morphological descriptors with quantitative measurements of cell survival remain elusive. Here we present a workflow to generate morphological signatures for the progeny of individual cells in the presence or absence of antibiotics. The workflow uses stationary microfluidic droplets, to encapsulate and grow bacteria, and confocal microscopy to image the contents of each droplet. Experiments are performed for 168 antibiotic conditions, corresponding to 82,000 droplet images. This massive data set is treated using a custom image analysis pipeline that enables rapid labeling of the morphologies within a subset of the images. The curated images are then used to train a neural network that assigns a positive or negative label, for each of these non-exclusive features, yielding a multidimensional signature of labels within each droplet. The results show the co-existence of different morphologies, even for the progeny of individual cells. The mix of morphologies changes as a function of antibiotic type and concentration, thus providing a way to distinguish antibiotics by their mode of action. By combining these morphological signatures with the digital detection of survival, this workflow can serve to understand the emergence of antibiotic resistance or to identify antimicrobial activity of unknown substances. This dataset contains the SQL database, network weights, and imaging data needed to recreate the interactive web viewer. More details are available at the associated GitHub repository.细菌的形态学特征与其生理状态紧密相关：其既会因抗生素胁迫发生改变，同时也可助力细菌在抗生素环境中存活。正因如此，明确存活细菌所选择的形态类型，有助于阐释单个细胞如何摆脱抗生素施加的选择压力。然而，将形态学描述与细胞存活的定量检测结果相关联，仍是一项尚未攻克的难题。本研究提出一套标准化工作流程，用于在有无抗生素的培养环境下，为单个细菌细胞的子代群体生成形态学特征标签。该流程采用静态微流控液滴（stationary microfluidic droplets）封装并培养细菌，通过共聚焦显微镜（confocal microscopy）对每个液滴内的菌群内容物进行成像。实验共覆盖168种抗生素处理条件，总计获取82000张液滴图像。针对这一大型数据集，研究团队开发了定制化图像分析管线，可快速对部分图像中的细菌形态完成标注。随后，经人工审核标注的图像被用于训练神经网络：该网络可为每个非排他性特征赋予阳性或阴性标签，最终为每个液滴生成多维度的标签特征集。研究结果显示，即使是单个细胞的子代群体，也会同时存在多种不同的细菌形态。菌群的形态组成会随抗生素种类与浓度的变化发生显著改变，这为通过作用模式区分不同抗生素提供了可行的分析途径。将上述形态学特征与细菌存活的数字化检测相结合，该工作流程可用于解析抗生素耐药性的产生机制，或是鉴定未知物质的抗菌活性。本数据集包含重建交互式网页查看器所需的SQL数据库、神经网络权重以及成像数据。更多详细信息可查阅关联的GitHub代码仓库。