Testing and validation of a new automatized analysis method for cell volume changes

GALIZIA, LUCIANO; CAVIGLIA, AGUSTÍN F.; GIBSON, OLIVER M.; AMODEO, GABRIELA; OZU, MARCELO

Rosario

Congreso; L Reunión Anual de la Sociedad Argentina de Biofísica; 2022

Sociedad Argentina de Biofísica

Xenopus laevis oocytes is a widely used cellular model for aquaporin expression.Biophysical characterization of aquaporins expressed in oocytes is based onmonitoring cell volume changes. This model allows recording sequential imagesor videos with low-cost videomicroscopy equipment. Traditional image processingis performed manually. The first step is oocyte segmentation, which includesdetermination of an initial frame (ZF, Zero Frame) corresponding to an in-focusand stable oocyte image. Afterwards, the oocyte area in each image is computed andthen converted to a relative volume curve (Vt/V0),assuming spherical shape of oocytes. These sequential steps are time-consuming.Recently we developed a new Phyton-based tool which allows us to processimages, extract all the required information, and calculate Vt/V0curves and Pf of each oocyte in an unsupervised manner. For testingand validation of the automatic analysis we compared the script output to the Pftraditional estimation method. We analyzed a typical experimental dataset withvideos obtained at 30 frame per seconds using oocytes injected with incrementalmasses of AQP4-M23 cRNA. To test the method effectiveness, each obtained Vt/V0 curve was compared tothe manual determination. Then, the most accurate automatized curve in each experiment wasselected. In this step we validated as accurate determinations 79.5 ± 3.6 % ofthe total analyzed videos (43 from a total of 54 videos), without anydifference in the obtained Pf at every injected mass. We evaluated differencesin the estimation of ZF (ΔZF), Pf(ΔPf) and R2 between automatizedand traditional analysis. ΔPf of each experimental group didn’t showany difference and the global value remains similar to Pf in nativeoocytes (ΔPf = 0.93 ± 0.15 x10-3 cm.s-1, n=43). Thedifference between the methods in the frame detection is estimated trough ΔZF (ΔZF = 80 ± 15 frames, n =43) and corresponds to a time difference of 2.77 ± 0.35 seconds. This timedifference corresponds to approximately 7% of the entire recorded video. Interesting,determination coefficients from lineal fitting were lower in the automatizedmethod (R2man = 0.95 ± 0.03 vs R2script= 0,92 ± 0.03, n=43, p<0.05). Thisevidence suggests our new automatized method can estimate Pf from awhole data set with minimal error, however the automatically detected ZF valuesproduces a slightly less accurate linear fitting for Pf calculation.