(550a) Computer Recognition of Bacteria in Blood Plasma and Placement in Droplets for Antimicrobial Testing

Introduction. Blood infections by antibiotic-resistant bacteria are a tremendous burden on healthcare as they create higher patient treatment costs, and higher rates of morbidity and mortality. In particular for septicemia, doctors require an antimicrobial susceptibility profile to make correct decisions regarding antimicrobial treatment. Yet current clinical identification for suspected blood stream infection (BSI) typically requires >12 hrs from blood draw,^[1,2] since a sample must first be cultured to provide sufficient cells needed for tests to determine susceptibility to antibiotics (Abx). Furthermore, a mL of blood may only contain 10 to 100 bacteria.

Our lab has already developed a process to separate the red and white cells from bacteria in BSIs within minutes in preparation for a rapid antimicrobial susceptibility test.^[3,4] Our current research is focused on concentrating those bacteria and placing them in a microfluidic device (MFD) in which individual bacteria can be encapsulated in single droplets in an oil flow, which droplets contain various types and/or concentrations of Abx for a much faster determination of antimicrobial susceptibility. Instead of indiscriminately forming tens of thousands of droplets from all collected blood plasma, we save space and time by only forming nanoliter-sized droplets around a single bacterium as it flows through the MFD. We have combined computer vision recognition software with video capture of flow in our 3D-printed MFD. This presentation describes the results of identification of test particles, blood cells and bacteria in control suspensions and in blood plasma.

Experimental. 3D-printed microfluidic droplet generators with digital onboard pumps were built from a custom printed MFD as described previously.^[5-8] To this platform was added a visualization window with a volume of about 1 nL which was monitored using a FLIR Blackfly S monochrome camera mounted on a Nikon TE300 inverted microscope. During one digital pump cycle, an image was captured from the window and digitally processed to remove background. The resulting image was examined using a convolutional neural network (YOLOv5) to identify the presence of spherical plastic beads (10 and 2 µm diameter), red blood cells, and E. coli bacteria within the field. A positive identification for one of these objects flagged the opening of flow valved to direct the flow to our 3D droplet generator,^[5] which generated a 1 nL-sized aqueous droplet of around the single particle, suspended in a flow of oil.

Results. The identification algorithm was successfully trained on particles of 10 µM opaque beads, 2 µm fluorescent beads, red blood cells (RBC), and E. coli bacteria expressing a green fluorescent protein. The bacteria were visualized and identified using fluorescence optics. The 2 µm beads were visualized and identified with both fluorescence and bright field optics, and the RBC and 10 µm beads were visualized and identified using bright field optics. When identified, a valve was opened to send particles to the droplet generator. In some experiments the bacteria in droplets were incubated in the presence or absence of antibiotics. This presentation discusses the efficiency of correct particle identification, the efficiency of encapsulation in droplets, and the growth of bacteria in droplets containing bacteria.

Outlook. While this system is designed to rapidly capture from blood and grow bacteria in various antibiotics to examine antimicrobial resistance, it can be used to identify and shuttle any type of particle with a characteristic size greater than 1 µm. The visualization and identification algorithms are robust. This system could be applied to microfluidic identification and separation of many types of colloidal particles, both of biologic and non-biologic origin. The 3D printed devices are easily and rapidly produced for fast turn-around times during device development.

Reference

1. Burklund A, Zhang JXJ. Microfluidics-Based Organism Isolation from Whole Blood: An Emerging Tool for Bloodstream Infection Diagnosis. Ann Biomed Eng. 2019;47(7):1657-74. Epub 2019/04/14. doi: 10.1007/s10439-019-02256-7. PubMed PMID: 30980291; PMCID: PMC6687464.

2. Santella B, Folliero V, Pirofalo GM, Serretiello E, Zannella C, Moccia G, Santoro E, Sanna G, Motta O, De Caro F, Pagliano P, Capunzo M, Galdiero M, Boccia G, Franci G. Sepsis-A Retrospective Cohort Study of Bloodstream Infections. Antibiotics (Basel). 2020;9(12). Epub 2020/12/03. doi: 10.3390/antibiotics9120851. PubMed PMID: 33260698; PMCID: PMC7760988.

3. Pitt WG, Alizadeh M, Blanco R, Hunter AK, Bledsoe CG, McClellan DS, Wood ME, Wood RL, Ravsten TV, Hickey CL, Cameron Beard W, Stepan JR, Carter A, Husseini GA, Robison RA, Welling E, Torgesen RN, Anderson CM. Factors affecting sedimentational separation of bacteria from blood. Biotechnol Prog. 2019:e2892. Epub 2019/08/20. doi: 10.1002/btpr.2892. PubMed PMID: 31425635.

4. Alizadeh M, Wood RL, Buchanan CM, Bledsoe CG, Wood ME, McClellan DS, Blanco R, Ravsten TV, Husseini GA, Hickey CL, Robison RA, Pitt WG. Rapid separation of bacteria from blood - Chemical aspects. Colloids and surfaces B, Biointerfaces. 2017;154:365-72. Epub 2017/04/04. doi: 10.1016/j.colsurfb.2017.03.027. PubMed PMID: 28365426; PMCID: PMC5456291.

5. Warr CA, Hinnen HS, Avery S, Cate RJ, Nordin GP, Pitt WG. 3D-Printed Microfluidic Droplet Generator with Hydrophilic and Hydrophobic Polymers. Micromachines (Basel). 2021;12(1). Epub 2021/01/21. doi: 10.3390/mi12010091. PubMed PMID: 33467026; PMCID: PMC7830873.

6. Beauchamp MJ, Gong H, Woolley AT, Nordin GP. 3D Printed Microfluidic Features Using Dose Control in X, Y, and Z Dimensions. Micromachines (Basel). 2018;9(7). Epub 2018/11/15. doi: 10.3390/mi9070326. PubMed PMID: 30424259; PMCID: PMC6082287.

7. Gong H, Bickham BP, Woolley AT, Nordin GP. Custom 3D printer and resin for 18 mum x 20 mum microfluidic flow channels. Lab Chip. 2017;17(17):2899-909. Epub 2017/07/21. doi: 10.1039/c7lc00644f. PubMed PMID: 28726927; PMCID: PMC5571981.

8. Noriega JLS, Chartrand NA, Valdoz JC, Cribbs CG, Jacobs DA, Poulson D, Viglione MS, Woolley AT, Ry PMV, Christensen KA, Nordin GP. Highly integrated and miniaturized 3D printed serial dilution microuidic devices for dose-response assays. Nat Commun. 2021;accepted.