A new feature for monitoring the enzymatic harvesting process of adherent cell cultures based on lensless imaging

LFI configuration

A lensless imaging device (LFI).19mounted inside a standard cell culture incubator (37 °C, 5% CO2 and 95% humidity, Thermo Fisher Scientific) with a 3D printed stage for the 6-well plate holder (Fig. 1), was used for continuous monitoring of the cell harvesting process. As the acquisition speed might be a limiting factor due to the heating of the image sensor, the temperature response was monitored for a high acquisition speed, as described in Supplementary Fig. S1.

Figure 1
Figure 1

Lensless imaging. (a) Schematic of lensless imaging setup (figure adapted from 31). (b) Image of the LFI microscope, mounted inside a standard cell culture incubator with a 3D printed stage for plate support. (c) Reconstructed phase image of the detaching cells, acquired with the LFI set-up. The black scale bar represents 100 µm.

Cell expansion

Human periosteum-derived mesenchymal stem cells (hPDCs) were isolated from periosteal biopsies of seven different donors, as described by Roberts et al..20. The procedures were approved by the Ethics Committee for Human Medical Research (KU Leuven) and the patient informed consent forms were obtained. The hPDC donors were expanded (5700 cells/cm2) at 37°C, 5% CO2 and 95% RH in Dulbecco’s GlutaMAX™ Modified Eagle High Glucose Medium (DMEM; Life Technologies, UK) containing 1 × 10-3 m sodium pyruvate and supplemented with 10% irradiated fetal bovine serum (HyClone FBS; Thermo Scientific, USA) and 1% antibiotic-antifungal (100 units/ml penicillin, 100 mg/ml streptomycin, and 0.25 mg /ml of amphotericin B; Invitrogen). The medium was changed every 3-4 days. At all passages, cells were harvested with TrypLE™ Express 1× (Life Technologies, UK). In this study, all experiments and methods involving these cells were performed in accordance with relevant guidelines and regulations.

Collecting essay

A collection assay involving different donors, cell densities, and collection solutions was performed. Seven different hPDC donors were seeded in 6-well plates (5700 cells/cm2, 5 wells per donor) in 3 ml of growth medium (DMEM-C), which was completely refreshed after 3 days. In order to study the effect of culture time (related to cell density and matrix formation) and the dilution of the enzyme solution on the shedding rate of each donor, cells were cultured for 3, 5 and 7 days and harvested enzymatically with a standard (day 3, 5 and 7) and diluted TrypLE Express solution (5x in PBS, day 5 and 7). The following protocol was used for each well: (i) Growth medium was aspirated and cells were washed 1× with phosphate-buffered saline (PBS, 1 mL). (ii) The LFI was calibrated to the cells (in PBS) and a time-lapse experiment was started with an interval of 20 s. (iii) PBS was aspirated and the “dry” well placed under the LFI. Reference images were taken before TrypLE was added. (iv) Between two acquisitions, 1 mL of TrypLE solution was added to the well plate (opened) and the incubator was closed. The collection process was monitored for 20 minutes and all images were acquired without plate coverage.

Feature engineering and image reconstruction

LFI phase and intensity images were reconstructed, equivalent to conventional phase-contrast and bright-field microscopy, respectively19. Different reconstruction parameters, such as the reconstruction method [3L or iterative phase recovery (IPR)], iteration count, IPR threshold (80-120), and focus level were varied for the intensity image to identify a new feature for measuring cell detachment. Based on the visual inspection, the following reconstruction parameters were chosen: temperature (37°C), clipping (no), phase rotation (auto). More specifically, the phase image was reconstructed using the 3L method (number of iterations: 5) and the intensity image with the IPR method (IPR threshold: 90, number of iterations: 10). For each dataset, the reconstruction depth was determined manually on cells attached around the image center. The raw image has been cropped with the following pixel parameters: [836:2236, 1198:2898].

Image processing and feature extraction

A software tool has been developed and implemented in MATLAB© 2019a (MathWorks, MA, USA) to extract the new feature, defined as the ratio of detached cell regions to the total number of cell pixels, from the reconstructed phase and intensity images. In Fig. 2 the segmentation procedure was visualized. The phase image was segmented using a random forest classifier, trained with the “Pixel Classification” workflow in Ilastik21. The standard feature variables for color/intensity, edge, and texture were considered with a sigma value of 0.7, 1, 1.6, 3.5, 5, and 10. In total, 36 features were used for classification. The classifier was trained on several images of different donors, ranging from attached to detached cells. Batch classification of all phase images was performed in Python using Ilastik’s “headless” mode. The next steps were explained in more detail in Fig. 2. Furthermore, the cell masks in Fig. 2B were used to extract the cell circularity, defined as \(Circularity= \frac{4*Area*\pi }{{Perimeter}^{2}}\)and mediated on the cell population.

figure 2
figure 2

Automatic feature extraction procedure. (a) Raw phase image. (b) The cell mask obtained with the Ilastik software, post-processed by filling holes smaller than 200 pixels (4-connectivity) and subsequently removing objects with an area smaller than 50 pixels (8-connectivity). (c) Segmented cells superimposed on the raw phase image. (d) Raw intensity image. (and) High intensity pixels (threshold 0.65) of the phase image were segmented. (f) Pixels/regions of detached cells (threshold 0.35) were segmented from the intensity image. (g) Both masks were combined (OR operation), holes smaller than 1000 pixels (4-connectivity) were filled, regions were matched (AND operation) with the phase mask (b) and regions smaller than 30 pixels (8-connectivity) have been filtered out. (h) The remaining regions were used as seeds and dilated with a circular disk of 5 pixels. Subsequently, these dilated regions were matched (AND operation) with the cells detected in (b). The resulting detached cell mask was subtracted from the cell mask in (b) and regions smaller than 50 pixels (8-connectivity) were added to the separate cell mask (OR operation). The final mask of the detached cell regions was obtained. (I) Regions of detached cells superimposed on the intensity image. (j) Segmented cells overlaid on the phase image, with attached and detached cell regions indicated by a blue mask and a green mask, respectively.

Quantitative evaluation

The detachment function was validated quantitatively for three different culture times (after 3, 5 and 7 days of culture, using the undiluted datasets). For day 3, 401 × 401 pixel images were cropped around the center of the reconstructed images, randomly selected over time for different donors. For days 5 and 7, the same procedure was followed, but images of 301 × 301 pixels were used due to the higher cell densities. In total, each validation dataset consisted of 16 images. Using the brush tool in the Image Labeler app in MATLAB© 2019a (MathWorks, MA, USA), attached and detached cell regions were manually labeled for each phase image. Next, the percentage of detached cell regions was calculated by dividing the number of detached pixels by the total number of labeled (i.e., detached and attached) pixels. To validate the method, the absolute percentage error was calculated according to the formula (1):

$$Absolute \, error (\%)= {abs(Disconnected\, cell \%}_{man}-{Disconnected\, cell \%}_{auto})$$


With “man” and “auto” referring to the manual and automated extraction of the percentage of detached cell regions, respectively. For each culture time, absolute percent errors were represented as mean ± std.

Harvest time

For all conditions, the optimal time point to inhibit the reaction was determined by applying a fixed threshold of 92.5%. This threshold was confirmed by visual inspection of the detached cell cultures at the respective times of inhibition. Additionally, supplementary videos (4-7) were constructed to visualize the response to detachment and highlight the optimal time point for inhibition. In case the threshold was not reached within the 20 min period, the final percentage of detached cellular regions was indicated.

Add a Comment

Your email address will not be published. Required fields are marked *