The colorectal cancer (CRC) cell line HT-29 (ATCC, RRID: CVCL_0320) and the human dermal fibroblast cell line HDF (PELOBiotech GmbH, RRID: CVCL_DP66) were cultivated at 37°C in a 5% CO₂ atmosphere. The used culture medium consisted of Dulbecco’s Modified Eagle’s Medium (DMEM (1X), Thermo Fisher Scientific Inc.), 10% v/v Fetal Bovine Serum (FBS, heat-inactivated, Thermo Fisher Scientific Inc.), 1% v/v Gibco™ Penicillin Streptomycin (Pen Strep, Thermo Fisher Scientific Inc.) and 1% v/v Gibco™ L-glutamine 200 mM (Thermo Fisher Scientific Inc.). We changed the cell culture medium every 2 days and passaged the cells at 80% confluency.

Cell-cell and cell-extracellular matrix (ECM) protein interactions are essential for proper tissue development, maintenance, and regeneration. Thus, in vitro cells adhere, grow, and perform more physiologically when adhering to surfaces coated with ECM proteins or specific adhesion molecules (Cruz Walma and Yamada, 2020) Both Poly-L-lysine, which is commonly used in neuroscientific applications (Eickenscheidt and Zeck, 2014; Stutzki et al., 2016; Cojocaru et al., 2022) and Collagen type I as the most abundant ECM protein in our body (Murata et al., 1986; Katsuda et al., 1992; Hohenester and Engel, 2002) ensure tight tissue adhesion (Vitale et al., 2018). We therefore addressed the biocompatibility of the cited ECM biomolecules (Prockop and Kivirikko, 1995; Vitale et al., 2018) by analyzing the viability of the cell line HT-29 cultivated in 96-well plates and on CMOS MEAs coated with Poly-L-lysine or Collagen type I.

We evaluated the effect of 5-Fluorouracil (5-FU, Sigma-Aldrich GmbH) on HT-29 and HDF. For that, the cell lines were cultivated in 96-well plates and on CMOS MEAs coated with Collagen type I. Recent studies indicate that 5-FU exerts a cytotoxic effect on HT-29 cells (Fohlen et al., 2021), leading to cell apoptosis (Vallicelli, Fary, et al., 2018). We therefore assumed that apoptotic cells (i) generate a lower OD during the CCK-8 assay with lower Cell Viability and (ii) detach from the ECM coating, causing fewer sensors where cells are adhered to, which results in a smaller area of the CMOS MEA sensor array covered by cells.

For these experiments, we coated the CMOS MEAs with 10% (v/v) Collagen type I and seeded both cell lines in separate experiments. HT-29 was seeded at a concentration of 20,000 cells/chip as described before (2.2) and HDF was seeded at a concentration of 2,000 cells/chip. Both cell lines were seeded on the uncoated 96-well plates at a concentration of 10,000 cells/well. After overnight incubation for cell adhesion, we replaced the cell culture medium with a medium containing 0.25 mg/mL of 5-FU. The cells were kept in the 5-FU-supplemented medium for the remaining experimental time (72 h). For the untreated control group, we exchanged the medium at the same time when the drug treatment started. The 5-FU-containing medium was prepared freshly before each treatment with a 1 mg/mL stock solution.

We performed the CCK-8 assay and the CMOS MEA recordings at 0 h, 24 h, 48 h, and 72 h after overnight adhesion. For the statistical analysis (t-test and ANOVA), we used the GraphPad Prism 10 software with ns, not significant, *, p ≤ 0.05, **, p ≤ 0.01, ***, p ≤ 0.001, ****, p ≤ 0.0001.

The CMOS MEA biosensing platform CAN-Q Chip comprised 256 × 384 capacitive recording sites with a pitch of 5.6 µm × 6.5 µm covering an active area of 1.6 mm × 2.5 mm as described in (Vallicelli, De Matteis, et al., 2018; Vallicelli, Fary, et al., 2018). The sensor array is covered with a 30 nm ALD-TiO₂ top oxide layer. A Perspex culture chamber is glued on the chip, exposing the recording site arrays to cell culture and medium (Zeitler and Fromherz, 2013). The recording transistor features an open gate between the source and drain contacts, which is insulated from the electrolyte by an oxide layer to prevent electric or ionic charge transfer (Figure 1A, top) (Zeitler et al., 2011; Zeitler and Fromherz, 2013; Vallicelli, Fary, et al., 2018). A local voltage change in the electrolyte-filled cleft between the cell and the recording site modulates the source-drain current. Prior studies on neurons (Voelker and Fromherz, 2006; Zeitler et al., 2011; Zeitler and Fromherz, 2013) demonstrated how the cleft resistance R_J enhances the voltage noise V_J (Figure 1A, bottom).

We calibrated the CMOS MEA with the CAN-Q Acquisition software (Venneos GmbH) with default settings using an Ag/AgCl electrode (VWR International GmbH) as an external reference electrode in PBS (1X). The CAN-Q Acquisition software assessed the spectral power density S_V at frequencies between 1 kHz and 450 kHz. The CMOS MEA recordings with PBS were calibrated to characterize the sensors’ background noise.

For the cell proliferation studies, we recorded the cells after the initial overnight adhesion time to the ECM protein (as described in 2.2) at 0 h, 24 h, 48 h, and 72 h. For the drug treatment studies, we changed the cell culture medium with a 5-FU-supplemented medium and recorded at 0 h, 24 h, 48 h, and 72 h (described in 2.2 and 2.3).

The viability of the cells was additionally quantified using the CASY Cell Counter and Analyzer (OLS OMNI Life Science GmbH & Co. KG). CASY counts and analyzes the cells’ viability using Electronic Current Exclusion (Lindl et al., 2005) and Pulse Field Analysis. The cell size and viability are detected based on the electrical resistance of the cell membrane (Weinreich et al., 2014; Grintzalis et al., 2017).

We collected two samples for each chip: 1) cells attached to the chip and 2) cells floating in the culture medium (i.e., supernatant). The first cellular fraction was detached from the chip using 100 µL Accutase (Sigma-Aldrich GmbH) as a mild-acting enzyme (Quan et al., 2013; Chen et al., 2015; Skog et al., 2019; Nowak-Terpiłowska et al., 2021; Lai et al., 2022) after 10 min incubation at room temperature. The second cellular fraction was isolated from the medium. For that, we collected the chips medium and centrifuged the samples at 800 rpm for 5 min. Then, we removed the supernatant and resuspended the pellet in a 100 μL cell culture medium. To do the CASY measurements, 100 μL of each sample was diluted with 10 mL of CASYton buffer and loaded for measurement. A sample consisting only of medium without cells served as blank. The data were collected with a custom CASY^® Excel program (Innovatis AG) and analyzed with GraphPad Prism 10 as described above.

The S_V was analyzed as an electrical image with image processing techniques using the OpenCV Library via custom Python scripts. Firstly, we subtracted the S_V of the bare electrodes (i.e., background noise) from the proliferation and drug treatment recordings to extract the cell adhesion noise (CAN), assuming uncorrelated noise sources (Zeitler et al., 2011). Secondly, the electrical image was grey-scaled and blurred with a Gaussian Blur filter to reduce the amount of image noise (i.e., single-pixel errors) to enhance the identification of objects of interest (i.e., cells and cell clusters) (Haralick and Shapiro, 1992). Thirdly, Otsu’s Binarization method determined the threshold for cell segmentation (Otsu, 1979). Finally, morphological operations (i.e., opening, closing) were performed to fine-tune the object detection and recognition. For drug treatment studies, we assumed that unhealthy or dead cells detach from the MEA (Frisch and Francis, 1994; Frisch and Screaton, 2001). This leads to a decline in adhesion noise that does not differ from the background noise. After these imaging processing steps, the number of sensors detecting cells estimated the cell-covered area.

Brightfield images with an upright light microscope (Zeiss Axioplan, ×10 objective, Carl Zeiss AG) related the electrically estimated cell positions to ground truth. For the CMOS MEA imaging, we reconstructed the whole microscopic image from individual image sections with the Fiji (Fiji is just ImageJ, version 1.54f, RRID: SCR_002285) plugin “Stitching” described in (Preibisch et al., 2009). Afterwards, we segmented the cell-covered area with the plugin “Trainable Weka Segmentation.” Brightfield images of 96-well plates were taken with an inverted microscope (Zeiss Axiovert 35, Carl Zeiss AG) with a ×10 objective.

To identify adherent cells, we recorded the voltage generated in the cleft between the cell membrane and the planar oxide surface. Previous studies demonstrated that the noise of the cleft can be used to infer the presence of adherent cells (Voelker and Fromherz, 2006; Zeitler et al., 2011; Zeitler and Fromherz, 2013). Details of the detection technique are described in the method section. We assessed voltage fluctuations in the adhesion area of non-electrogenic cells (like the CRC cell line HT-29 and HDFs) cultured on the MEA recording sites (i.e., electrolyte-oxide-silicon field-effect transistors) (Figure 1A). The resistive cleft below non-electrogenic cells gives rise to the adhesion voltage noise, allowing for distinguishing this value from that of a bare sensor site (Rocha et al., 2016; Ell, Zeitler, et al., 2023) (Figure 1B). Adherent cells are detected by estimating the S_V spectrum. Comparing spectra from uncovered with cell-covered sensor sites, S_V exhibits elevated values with cells attached, which is attributed to the resistive cleft. To assess the values of the resistive cleft from the voltage noise spectral power density (S_V), we compared the estimate of electrical imaging to brightfield microscopy by overlay (Figure 1C). Figure 1B illustrates that cells adhered to sensor sites give rise to an increased S_V across the entire frequency spectrum. Subtracting the S_V of a bare sensor from the estimated S_V in culture (i.e., sensors exposed to the electrolyte) provides the adhesion noise ΔS_V (Figure 1B inset), assuming that the two noise sources are independent (Zeitler et al., 2011). Here, we identified the frequency range with a flat adhesion noise spectrum indicating a frequency-independent (i.e., resistive) cleft from 100 kHz to 450 kHz (Voelker and Fromherz, 2006). We selected a frequency value in the middle of this frequency range, i.e., 300 kHz, for all future analyses of ΔS_V. From the adhesion noise plateau with an average amplitude of ΔS_V = 0.027 µV²/Hz, we estimated the cleft resistance using Eq. 1 with a thermal energy k_BT at incubation temperature (Zeitler and Fromherz, 2013). R c l e f t = ∆ S V / 4 k B T (1)

Equation 1: Cleft resistance of the cell-transistor interface derived from noise spectrum. ΔS_V denotes the cell adhesion noise, k_BT refers to the thermal energy with the Boltzmann constant k_B and incubation temperature T (Zeitler et al., 2011).

The cleft resistance R_cleft = 1.58 MΩ aligns with previous studies on neurons (Voelker and Fromherz, 2006). 1/f noise dominated the CAN for the S_V data below 10 kHz frequency, which is attributed to the intrinsic setup noise and not used for further analysis here (Haartman and Östling, 2007; Viswam et al., 2018; Lausen et al., 2022). Undetected small structures on the sensor surface indicate poor cell attachment or dead cells. The exemplary Figure 1C and parts of Figure 3 demonstrate that adhesion voltage noise is a valuable parameter for the label-free detection of adherent cells. In the following, we assess to what degree ΔS_V changes upon application of 5-FU to the cell culture.

After extracting the cell adhesion noise (CAN) ΔS_V from the total voltage noise, we compared ΔS_V across sensors, time, cell types, and different CMOS MEAs employed at untreated and drug-treated (5-FU) states. Therefore, we (i) identified adherent cells by their ΔS_V, (ii) calculated the mean ΔS_V averaged over thousands of cell-identifying sensors (Figures 2A, B, grey dashed dots), (iii) plotted the ΔS_V-traces of each CMOS MEA for the cultivation time of 72 h (Figures 2A, B, grey dashed lines), and (iv) estimated the mean and the standard deviation of all chips employed for both cell types (i.e., HT-29 CRC cells and HDFs) (Figures 2A, B, mean as bold green and pink line, standard deviation as error bars). Figure 2 compares the mean ΔS_V (at 300 kHz frequency) of adherent untreated cell networks (top) to the mean spectrum of 5-FU-treated ones (bottom).

Untreated and 5-FU-exposed HT-29 cells exhibited a mean ΔS_V between 0.009 µV²/Hz and 0.014 µV²/Hz with slightly higher variations of treated cells (Figure 2A). The ΔS_V mean for untreated and 5-FU-treated HDFs remained constant until 48 h of cultivation between 0.012 µV²/Hz and 0.014 µV²/Hz (Figure 2B). After 72 h cultivation, ΔS_V for untreated HDFs went up to 0.020 µV²/Hz with a statistical difference of p ≤ 0.01 (Figure 2C) while remaining constant for cells in 5-FU (Figures 2B, C). This longitudinal comparison demonstrates a stable interfacing for 72 h of the two cell cultures under test. We could not detect any significant differences except for HT-29 cells at 0 h and 72 h, which will be addressed in the discussion section. Given that ΔS_V remains stable for adherent cells, we simplify our analysis and assign sensors with adherent cells as “positive” sensors and with no cells as “negative.” The total fraction of “positive” sensors is used in the following to quantify cell proliferation on a coated CMOS MEA surface.

Since the values of ΔS_V do not change over time upon application of 5-FU (Figure 2), we assessed an additional feature provided by the cell-detecting sensors on the CMOS MEA: the cell-covered area. If the sensor area covered with adherent cells changes in control conditions in a different way than the cell-covered area upon treatment, we would have identified a potential indicator of the proliferation status. In the first experiment, we sought coating conditions where the cell-covered area increases over time.

We analyzed the Collagen and PLL effects on the viability and proliferation capacity of the CRC cell line HT-29 (Prockop and Kivirikko, 1995; Vitale et al., 2018). For that, the HT-29 cells were cultivated in a 2D monolayer in Collagen and PLL-coated 96-well plates and on CMOS MEA. Data from CCK-8 proliferation assay and CAN spectroscopy CMOS MEAs were collected. Cell Viability in %, optical density (OD), and the cell detection positive area per chip were quantitatively examined, as shown in Figure 3. Electrical imaging (red contours) was overlayed with brightfield microscopy imaging (grey background) (Figure 3A) and Figure 3B shows brightfield microscopy images of the cells in 96-well plates. After 2 days of cultivation, the HT-29 cells detached from the PLL-coated sensors while forming clusters and proliferating on Collagen-coated sensors, the same for cells in 96-well plates. We could draw two conclusions from the estimated cell detection positive area (Figure 3C): (i) initial HT-29 cell attachment is better in Collagen-coated plates (cell-covered area at 0 h: 47.5%) than in PLL-coated ones (cell-covered area at 0 h: 12.5%) and (ii) cell attachment dropped for cells growing on PLL after 2 days of cultivation (cell-covered area decreased at 48 h to 5.2%) whereas Collagen enhanced cell attachment and proliferation (cell-covered area at 48 h: 77.0%).

The CCK-8 assay allows us to evaluate the % of metabolic active cells in culture. Dehydrogenases in alive cells can reduce WST-8 [2- (2- methoxy-4-nitrophenyl)-3-(4-nitrophenyl)-5-(2, 4-disulfophenyl)-2H-tetrazolium, monosodium salt] to a water-soluble orange dye known as formazan. The amount of formazan produced correlates with the number of viable cells and can be measured with an absorbance microplate reader at 450 nm (Cai et al., 2019). In Figure 3D, it is shown that the OD increased by 1.9-fold from 0.85 at 0 h to 1.63 OD after 48 h of cultivation on Collagen. For cells growing on PLL, the OD dropped not significantly by 1.9-fold from 0.96 at 0 h to 0.63 OD after 48 h (Figures 3B, D).

HT-29 cells growing on Collagen showed constant ΔS_V values over time, but the cells’ ΔS_V showed variations when growing on PLL-coated MEAs (Supplementary Figure S1). Supplementary Figure S2 relates the two coatings (i.e., Collagen and PLL) to uncoated 96-well plates as control group with statistical significance p ≤ 0.01 at initial cell attachment (0 h) and p ≤ 0.0001 after 48 h of cultivation. Supplementary Figure S3 compares cell segmentation of electrical imaging with brightfield microscopy imaging with non-significant differences in the cell-covered area of the CMOS MEAs but slightly higher rates for microscopic images.

Based on the positive adhesion results, we used Collagen-I to evaluate the response of colorectal cancer cells HT-29 and non-cancerous HDF cells to the anti-cancer drug 5-FU. 5-FU has a cytotoxic effect on HT-29 cells, leading to cell apoptosis (Mhaidat et al., 2014). Consequently, cells undergoing cell death metabolize less or no WST-8 dye, which results in lower optical density (OD) for CCK-8 assay. Similarly, these cells detach from the MEA and will not be detected by the sensors. Hence, we correlated cell viability and apoptosis with OD and the area that is covered by adherent cells (Ell, Bui, et al., 2023). CASY recorded the cell viability of cells attached to the CMOS MEA surface and cells floating in the cell culture medium.

Untreated HT-29 cells growing on the Collagen-coated MEA proliferated and the area covered by adherent cells doubled after 72 h in culture (Figure 4A-i). However, after treating the HT-29 cells with 5-FU, approximately half of the cells detached from the chip and the cell-covered CMOS MEA area declined by 1.8-fold after 72 h exposure time (Figure 4A-ii). Figure 4A-iii shows statistical significance p ≤ 0.01 for untreated and 5-FU-treated HT-29 cells after 72 h with less cell-covered area (54%–22%), while a shorter exposure time to 5-FU for 24 h led to a larger cell-covered area (28%–40%). The ODs for the HT-29 cells matched the CAN-based areas: untreated HT-29 cells showed a 2.5-fold higher OD at 72 h compared with 24 h (Figure 4C-i), whereas HT-29 cells exposed to 5-FU showed a decline in OD by 2.2-fold (Figure 4C-ii). The agreement between CAN and OD value is, however, detected only after 72 h for HT-29 cells.

While 5-FU caused cell detachment on the CMOS MEA and a lowered metabolic activity in the CKK-8 assay for the HT-29 cells, the HDFs’ proliferation status was altered differently: the CMOS MEA’s positive cell detection area for untreated HDFs was on a constant level (Figure 4B-i) and also the OD did not change (Figure 4D-i). For the 5-FU-treated HDFs, the cell-covered area remained constant over 72 h (Figure 4B-ii). The OD decreased within the first 24 h of 5-FU exposure and stagnated for the remaining culture time (Figure 4D-ii). Untreated HDFs showed a slight decrease in cell-covered area (Figure 4B-i), while treated ones remained on a constant coverage over time (Figure 4B-ii). Figure 4B-iii indicates a statistical difference of p ≤ 0.001 for 24 h and p ≤ 0.01 for 48 h, but no significance for 72 h by analyzing the cell-covered area. The cell viability lowered upon 5-FU application to 74% after 24 h, 57% after 48 h, and 58% after 72 h drug exposure time with p ≤ 0.0001 (Figure 4D-iii).

CASY results in Figure 4E-i show elevated cell viability for untreated HT-29 cells compared with 5-FU-treated cells in the supernatant. HT-29 cells adhered to the MEA exhibited significantly higher cell viability than in the supernatant with p ≤ 0.01 (untreated) and p ≤ 0.05 (treated). CASY recorded higher viability for untreated HDFs in the supernatant than for 5-FU-exposed ones with p ≤ 0.01 (Figure 4E-ii). Untreated as well as treated CMOS MEA-adherent HDFs remained at the same viability level.