Multi-assay assessment of cytotoxicity reveals multiple mechanisms of action in 3D microtissues

Zein-Sabatto, Ahbid; St. Angelo, Katerina; Madnick, Samantha J.; Hoffman-Kim, Diane; Morgan, Jeffrey R.; Lee, Jonghwan

doi:10.1038/s41598-025-86792-4

Download PDF

Article
Open access
Published: 24 January 2025

Multi-assay assessment of cytotoxicity reveals multiple mechanisms of action in 3D microtissues

Ahbid Zein-Sabatto^1,2,
Katerina St. Angelo³,
Samantha J. Madnick³,
Diane Hoffman-Kim^2,4,5,6,
Jeffrey R. Morgan^1,2,3,6 &
…
Jonghwan Lee^1,2,5,6

Scientific Reports volume 15, Article number: 3090 (2025) Cite this article

3139 Accesses
1 Citations
Metrics details

Subjects

Abstract

Cell viability assays are an integral component of toxicology and high-throughput drug screening studies; however, many assays rely on a single biomarker of cell death which provides an incomplete assessment of cell viability. Here, we introduce an innovative approach that combines data from multiple assays using a linear mixed effects regression model and principal component analysis. We explored the cytotoxic response of various assay-treatment combinations using four assays with distinct mechanisms of action and seven different treatments across three types of microtissue cultures. The multi-assay data revealed the presence of multifaceted cellular injuries which highlight the need for multimodal approaches to better understand complex disruptions to viability. By incorporating outputs from the four assays, we introduced a new lethal concentration threshold that captures changes from different cellular injuries to provide a more comprehensive evaluation of cytotoxicity. Overall, the proposed approach provides a unique opportunity to analyze data from multiple assays in a holistic manner to improve the predictive power of drug screening and toxicology studies.

Comprehensive analysis of cellular metrics: From proliferation to mitochondrial membrane potential and cell death in a single sample

Article Open access 24 March 2025

Towards a comprehensive approach for characterizing cell activity in bright-field microscopic images

Article Open access 07 October 2022

High throughput screening of mesenchymal stem cell lines using deep learning

Article Open access 20 October 2022

Introduction

Analysis of cytotoxicity is a key endpoint for toxicology and pharmacological in vitro studies. Identifying the main pathway of cell death elicited by a pharmaceutical provides key information regarding the compound’s pharmacodynamics and ultimate clinical success. For example, tumor necrosis has been associated with poor patient prognosis^1,2,3,4,5; therefore, selecting treatments that target alternative cell death pathways may circumvent inflammation-induced metastasis, immunosuppression, and pro-tumor cytokine release that are associated with drug-induced tumor necrosis^6,7,8. Studies have also identified tumors with apoptosis resistance that contribute to multidrug-resistant strains of cancer, highlighting the need for detailed investigation of the cytotoxicity induced by pharmaceuticals^9,10,11.

Many commercially available assays target specific biomarkers through enzymatic interaction or target-marker binding affinity¹². Consequently, an assay quantifies cell death by measuring changes to its specific biomarker. Cell death, however, may involve overlapping biological mechanisms that have various impacts on biomarkers associated with viability assays^13,14. For instance, assays measuring metabolic function may capture changes from accidental cell death, regulated cell death, autophagy, senescence, and other cellular perturbations^{14,15,16,17,18}. This necessitates the need to run a carefully curated array of assays to compile a more complete understanding of cell viability within a sample. Ultimately, selecting the most appropriate “gold-standard” assays is crucial considering the complexity of cell death that often favors a multimodal approach.

Analyzing and interpreting results from multiple assays remain challenging. Common statistical tests, such as the ANOVA, are conducted independently for each assay. This approach makes it difficult to obtain comprehensive results, including inter-assay responses, based on the collective outputs of all assays within a study. Also, fulfilling all the prerequisites for a monolithic statistical analysis in a high-throughput screening (HTS) study is often challenging when the normality, homoscedasticity, and number of replicates for a study are inadequate^19,20,21.

In this study, we demonstrated the analysis of multimodal cell viability data through a multivariate approach using a linear mixed effects regression model and principal component analysis (PCA). PCA is a multivariate approach that can be effectively used for dimensionality reduction of larger datasets²². PCA assumes the overall data structure can be reduced to a linear combination of its constituents and performs well with many replicates^22,23. Notably, the multivariate approach we utilized provided a unique opportunity to simultaneously analyze the results from four independent assays. This functionality enabled the direct assessment of inter-assay responses that capture various aspects of cellular injury under similar treatment conditions. The demonstrated results will inform researchers which assay-treatment combinations produce the most substantial cytotoxicity responses. The applicability of this approach to novel drug discovery and toxicology was strengthened by incorporating multiple tissue culture types including self-aggregating 3D tumor and primary neonatal rat cortical microtissues (also referred to as neurospheroids in this study). Previous studies have established the improved in vivo relevance and clinical translatability of the 3D in vitro model compared to traditional 2D cultures^24,25,26,27. In addition, the 3D self-assembling platform used in this study is easily scalable for high-throughput applications, making it a powerful in vitro model.

Results

ATP assay determines LC values for 2DG and oligomycin A treated microtissues

Cell death caused by metabolic inhibition was accomplished by treating microtissues with 2DG (2-Deoxy-D-glucose), a glucose analog that inhibits glycolysis, or oligomycin A, an ATP synthase blocker that inhibits oxidative phosphorylation^28,29. Dose-response curves were generated by plotting the change in ATP content measured by the CellTiter-Glo 3D assay (Fig. 1). The assay outputs for HepG2 and neurospheroids were normalized by the cross-sectional area of the microtissues, then all data was normalized by its control group to express viability as a percent change. The data was then fitted to a logistic function using a nonlinear mixed effects (NLME) model using Eq. (1) (black lines, Fig. 1a, c)

$$f\left( x \right) = f_{0} + \frac{{1 - f_{0} }}{{1 + e^{{k\left( {x - x_{0} } \right)}} }}$$

(1)

where f is the fractional output, x is the drug concentration in log, f₀ denotes the mean output of the blank sample (i.e., assay reagent with media alone), k represents the steepness, and x₀ is the midpoint concentration of the logistic function in the log scale. Passage-specific random effects were incorporated in the model. The resultant logistic fits were used to derive the final lethal concentrations (LC) for each tissue culture type from which the LCs elicited 25, 50, and 75% cell death relative to controls for both compounds (Fig. 1b, d). All LCs were calculated and listed in Table 1, except for oligomycin A treated HepG2 microtissues which could not elicit a LC50 and LC75 without exceeding solubility recommendations.

Live/Dead assay determines LC values for Triton X-100 and melittin treated microtissues

Loss of membrane integrity is a commonly attributed characteristic of cell death. This perturbation was accomplished by using Triton X-100, a nonionic surfactant, and melittin, a pore forming toxin from honeybee venom^30,31. Changes in cell viability were captured in-gel by the Live/Dead Viability/Cytotoxicity assay kit and calculated by Eq. (2)

$$Live\;Intensity\;Fraction = \frac{{I_{{live}} }}{{I_{{live}} + I_{{dead}} }}$$

(2)

where I_live is the mean pixel intensity of the segmented spheroid in the live channel, and I_dead is the mean pixel intensity in the dead channel (Fig. 2a). NLME was used to fit the dose-response curves to a logistic function using Eq. (3) (black lines, Fig. 2b, d)

$$f\left( x \right) = \frac{1}{{1 + e^{{k\left( {x - x_{0} } \right)}} }}$$

(3)

where f is the fractional output, x is the concentration in log, k is the steepness of the slope, and x₀ is the midpoint concentration of the logistic function in the log scale. Passage-specific random effects were also allowed for the midpoint concentration. Data was normalized to the control and reported as a percent change in viability (Fig. 2b, d). LC values were then extracted from the fitted lines for all tissue culture types (Fig. 2c, e, and Table 1).

Caspase assay determines LC values for cisplatin and melphalan treated microtissues

Cisplatin and melphalan are alkylating agents used to treat a variety of cancers^32,33. Their mechanism of action involves crosslinking DNA which inhibits replication, transcription, or translation^32,33. The accumulation of DNA damage within the cells ultimately results in apoptosis which we measured using the Caspase-Glo 3/7 3D assay. Dose-response curves were plotted and fitted to a logistic function (Eq. 1) using NLME (black lines, Fig. 3a, c). The resulting fits were then used to extract the designated LC values (Fig. 3b, d, and Table 1). Only HepG2 and neurospheroid data were normalized by the cross-sectional area of the microtissues.

EdU proliferation assay determines LC values for paclitaxel treated microtissues

Paclitaxel is classified as a taxane and over-stabilizes microtubules in the cell resulting in the cessation of G2 and M phases in mitosis³⁴. We captured the LC values of paclitaxel treated microtissues using the Click-iT EdU (5-ethynyl-2’-deoxyuridine) cell proliferation assay which labels newly synthesized DNA (Fig. 4a). The proliferative capacity was calculated by Eq. (4)

$$Proliferative\;capacity = \frac{{I_{{EdU}} }}{{I_{{Hoechst}} }}$$

(4)

where I_EdU is the mean brightness of pixels representing newly synthesized DNA tagged with EdU, and I_Hoechst is the mean brightness of pixels from all nuclei tagged by Hoechst staining (Fig. 4a). Data was normalized to the controls and expressed as a percent change. NLME was used to fit the dose-response curves using Eq. (3) (black lines, Fig. 4b). LC values were generated for each tissue culture type except for neurospheroids which could not capture paclitaxel’s LC50 or LC75 using recommended solubility guidelines (Fig. 4c; Table 1).

Multimodal data reveals inter-assay response and sensitivity

Thus far, we derived LCs from dose-response curves obtained by pairing treatments with their “gold-standard” assays (e.g., Live/Dead assay for melittin). However, this conventional approach does not investigate a compound’s off-target effects (e.g., how melittin affects cellular metabolism, apoptosis, or proliferation beyond its targeted effects on membrane integrity). To investigate multimodal interactions in assay sensitivity, we tested all combinations of treatments and assays at the LC values previously derived using the “gold-standard” assays (Figs. 1, 2, 3 and 4). This generated a total of 8,758 datapoints across all assay-treatment pairs (Supplementary data, Figs. S1–S4). Figure 5 displays the rearranged figures for HepG2 microtissues as an example dataset. Since the previously extracted LC values approximately captured the linear region of the dose-response curves, we fitted linear regressions to the multimodal dataset using a linear mixed effects (LME) model to capture assay-treatment sensitivity.

From this linear regression, we expected the slope to be -1.0 if the viability captured by an assay decreases proportionally to the increase in LC value (e.g., a 25% increase in LC results in a 25% decrease in viability). This expected slope was reproduced when the “gold-standard” assay-treatment pairs were used, except for ATP-oligomycin and Caspase standard pairs (Fig. 5). Oligomycin was not potent in HepG2 and required a high concentration to achieve its LC25 (242 µM) which pushed the subsequent LCs beyond the practical solubility limit. Meanwhile, the Caspase assay produced a positive slope since Caspase 3/7 activity increases with the accumulation of apoptotic cell death. In some instances, assay output sharply declined then plateaued after the LC25 (e.g., cisplatin treated HepG2 microtissues with the EdU assay, Fig. 5). In these cases, we excluded larger LC datapoints for more accurate fitting (see Methods for more detail). Not all assays were expected to elicit a significant response; therefore, we considered a response to be substantial (shaded by a dark gray region in Fig. 5) when the p-value of the LME fit was p < 0.05 and the magnitude of the slope was greater than 0.1 (i.e., 10% of the ideal slope of 1.0 for “best fit” assays).

Figure 6 presents the slopes from the linear regressions over all assay-treatment pairs and tissue culture types, where the slopes from the Caspase assay were converted for consistent visual comparison across different assays (see Methods for more detail). This conversion made the average slope of the standard-pair compounds (cisplatin and melphalan) to be −1; however, individual slopes were not −1 since the two compounds lead to different levels of apoptosis at the highest tested dose.

This multimodal data revealed the various biological mechanisms that occur simultaneously within a sample. The metabolic inhibitors 2DG and oligomycin A elicited substantial responses from both the ATP as well as the EdU proliferation assays (red and cyan bars, respectively, Fig. 6). Membrane disrupting compounds like Triton X-100 and melittin not only produced Live/Dead assay regressions with a slope near −1.0 as expected (blue bars, Fig. 6), but also substantially impacted ATP content within the samples (red bars, Fig. 6). Meanwhile, the alkylating agents particularly impacted Caspase 3/7 activity and proliferation (yellow and cyan bars, respectively, Fig. 6). The effects of paclitaxel in the tumor microtissues primarily disrupted the proliferation of cells (cyan bars, Fig. 6); however, the metabolic and apoptotic responses were more impacted in neurospheroids (red and yellow bars, respectively, Fig. 6).

While the analysis on the regressions’ slopes provided comprehensive insight regarding assay sensitivity and compound off-target effects, its results remain convoluted. This approach does not provide data that collectively represents the combinatory effects of different mechanisms of cytotoxicity that simultaneously occur within a sample. Here, we propose a method to determine a novel and comprehensive lethal concentration that incorporates multiple mechanisms of action by analyzing multi-assay data with PCA. PCA allows for the reduction of a dataset’s dimensionality to (n-1) arbitrary components that best explain the data’s variance. A simulation of multiple dose-response curves validated the PCA’s approach as a useful tool to analyze the multimodal datasets (Supplementary data, Fig. S5). Prior to running the PCA, all data was normalized by its control group and Caspase data was converted to maintain directionality (see Methods for more detail). In melittin treated HepG2 data, principal component one (PC1) accounted for 65.0% of the data variance and was able to sort the data based on LC (Fig. 7a). Principal component two, which accounted for 29.7% of the data variance, did not produce the same sorting effect.

Results for all other tissue-treatment combinations are presented in Supplementary data, Figs. S6–S8. For this study, we only focused on PC1 and fitted the data using a second order polynomial (Fig. 7b). A new comprehensive threshold, termed multi-mechanism lethal concentration 50 (MMLC50), was derived from this line of best fit (black line, Fig. 7b). The MMLC50 was determined by calculating two-thirds of the data’s range and subtracting that value from the PC value at the control then projecting that point onto the x axis (gray dotted lines, Fig. 7b).

HepG2 microtissues treated with oligomycin and paclitaxel and neurospheroids treated with paclitaxel did not produce a LC75 and were excluded. All LC and MMLC50 values were summarized in Table 1.

Table 1 Lethal concentrations (LC) resulting in 25, 50, and 75% cell death utilizing the “gold-standard” assays were reported alongside the multi-mechanism lethal concentration 50 (MMLC50). The MMLC50 thresholds were derived from principal component analysis run on the multimodal dataset. BPL: beyond practical limit, NA: not applicable, L/D: Live/Dead assay, neuro: Neurospheroid.

Full size table

PCA loadings provide an alternative method to compare inter-assay responses and compound cytotoxicity (Fig. 8). Loadings were calculated by Eq. (5).

$$Loadings = \frac{{Eigenvector*\sqrt {Eigenvalue} }}{{Standard\;Deviation}}$$

(5)

Assays with loadings of greater magnitude are more correlated with PC1 while positive or negative values indicate whether the assay results were directly or inversely proportional to PC1, respectively. Overall, multiple assays contributed to PC1 across treatments and tissue types with a few exceptions. Assay-tissue-treatment combinations with loadings less than 0.25 included EdU-HepG2-melittin, ATP-MCF7-cisplatin, and EdU-Neurospheroid-paclitaxel (Fig. 8). These findings were consistent with the LME regression analysis (Supplementary data, Figs. S1–S4).

Discussion

Dose-response curves are commonly used to profile a compound’s biological impact by plotting the change in activity over increasing concentration. In this study, we generated the typical sigmoidal curves of four widely used assays by testing 8–10 logarithmically spaced concentrations of seven compounds with distinct mechanisms of action on three tissue culture types (Figs. 1, 2, 3 and 4). Results from the dose-response curves provided information regarding the cytotoxicity of different compounds using a single matched assay. From this data, we identified that metabolic activity of HepG2 microtissues was resistant to oligomycin A treatment since very high concentrations were needed to elicit a LC25 and all subsequent LCs were not achieved. Similarly, the proliferation of neurospheroids was not greatly impacted by treatment with paclitaxel. The LC values in Table 1 captured the vulnerability of certain tissue culture types to different mechanisms of cellular injury. For instance, neurospheroids were more sensitive to metabolic inhibition involving oxidative phosphorylation compared to the tumor microtissues. The ability of tumors to rely on glycolysis to maintain metabolic demand is a well-documented phenomenon and explains the observed trend^35,36. Among membrane disrupting compounds, melittin had a greater impact on neurospheroids than in tumor microtissues, but Triton X-100 had a similar potency across the tissue types. Previous studies have shown that tumors overexpress annexin which consequently increases their membrane repair capabilities³⁷. Other mechanisms such as membrane blebbing, shedding, and endocytosis could also be upregulated in tumors to counteract the effects of melittin³⁸. Among the microtissues treated with alkylating agents, neurospheroids were most vulnerable to cisplatin and melphalan. Despite an overlap in mechanism of action (i.e., the crosslinking of guanine in DNA^32,39), cisplatin was more potent in HepG2 while melphalan was more potent in MCF7 microtissues (Table 1). This suggested that other biological mechanisms contribute to the discriminative cytotoxicity of the compounds in the two tissue types. Paclitaxel’s ability to over-stabilize microtubules ultimately results in the cell’s inability to undergo replication³⁴. HepG2 microtissues required a significantly larger dose of paclitaxel treatment to elicit its LC25 and LC50 compared to MCF7 microtissues which may allude to the presence of drug resistance pathways^40,41; whereas the slowly proliferating neurospheroids did not respond as substantially to paclitaxel treatment (Table 1).

Traditional LC values provide limited information regarding the potency of a compound to a single biological perturbation. This vastly underestimates the overall cytotoxicity since it fails to account for “off-target” effects of a treatment. To address this, we collected multimodal data from all combinations of seven treatments, four assays, and three tissue culture types. This provided an opportunity to observe the sensitivity of different assays to similar treatment conditions. The steepness of the slopes derived from the LME model can be attributed to the sensitivity of each assay (Figs. 5 and 6). This approach allowed us to identify the presence of multiple mechanisms of action occurring simultaneously within the sample. For instance, oligomycin A treated microtissues resulted in significant changes to not only ATP content, as hypothesized, but to proliferation as well (Fig. 6). Another example includes melittin treated microtissues which exhibited metabolic inhibition, membrane disruption, and reduced proliferation as concurrent biological perturbations (Fig. 6).

In parallel to LME fitted regressions, PCA provided an approach to compare inter-assay response. PCA is more robust against deviations in normality and homoscedasticity when sample sizes are sufficiently large^42,43. The applicability of the PCA approach was reflected by the simulation data, magnitudes of the loadings, and Chi-square values from the Bartlett’s test (Fig. 8, Supplementary data, Fig. S5 and Table S1). Though arbitrary in nature, thresholding the loadings could provide a method to sort treatment combinations that had statistically significant impacts on viability. PCA loadings, unlike the discrete slopes extracted in Fig. 6, correlated each assay to an arbitrary component that is dependent on input variables (Fig. 8). In general, PC1 can be broadly associated with cell viability but should not be over interpreted. PCA loadings with smaller magnitudes indicated that a specific assay was not sensitive to the cytotoxic response elicited by a treatment for a given tissue type. For instance, cisplatin treated MCF7 microtissues had a decreased ATP assay loading response compared to the other three assays which highlighted its lower sensitivity. This can be confirmed by cross-referencing with the LME fittings (Supplementary data, Figs. S1–S4). Collectively, the multimodal results revealed that most treatment combinations have at least two or more overlapping markers of cell death present, except for MCF7 microtissues treated with paclitaxel (Figs. 6 and 8). Longitudinal data collected after 12, 24, and 48 h of exposure to paclitaxel in MCF7 microtissues revealed highly “noisy” data from the ATP and Caspase assays, low sensitivity from the Live/Dead assay, and increasing EdU assay sensitivity with exposure time (Supplementary data, Fig. S9). These findings confirm the critical contribution exposure time has on dose-response curves and its impact on the multimodal data analysis. Future studies may consider optimizing temporal factors in their experimental design and should only interpret their results within the context of the time courses tested.

In this study, we showed that multi-assay approaches help simultaneously analyze different mechanisms of cellular injury, and their results can be intuitively visualized using methods like LME and PCA. The MMLC50 is another method to analyze multi-assay data; however, it is dependent on the types of assays selected. Since the MMLC50 is derived from PCA, input assays that carry the most variance will drive the data’s fit (Supplementary data, Fig. S5). The presence of input data with large margins of error, as was observed in the ATP and Caspase assay data from Supplementary Fig. S9, cause significant deviations in MMLC50. Similarly, if an input dose-response curve with a sufficient change in output plateaus at a smaller lethal concentration (i.e., assay output reaches its maximal effect at a lower dose than the initial standard assay-treatment pair), then the PC1 fit is significantly impacted. This is observed by a leftward shift in MMLC50 relative to the LC50 indicating that there may exist a more sensitive assay that is detecting changes to viability at a lower dose. For example, Triton X-100 treatment in HepG2 microtissues not only changed Live/Dead as expected (Fig. 6, blue bar) but also resulted in plateaus in ATP and EdU assays (Supplemental data, Figs. S1 and S4). MMLC50 close to the LC50 suggests that the compound, at minimum, causes disruption to the standard-assay while potentially affecting other assay outputs to a similar or lesser degree. Under such circumstances, PCA loadings can be used to identify whether multiple injury mechanisms are occurring simultaneously, as was the case with HepG2 microtissues treated with melittin (Fig. 8 and Supplemental data, S6). Finally, if an input dose-response curve with a sufficient change in output maintains a plateau at lower doses followed by a sudden steep change (i.e., assay output only begins to change at a higher dose than the initial standard assay-treatment pair), then the MMLC50 is also affected. This is observed by a rightward shift in MMLC50 relative to the LC50 which may suggest the presence of a more sensitive assay; however, the cellular injury captured by this assay requires a higher treatment dose to manifest. As previously mentioned, outputs from assays that have large variances but do not respond significantly to dose can be incorrectly analyzed by PCA MMLC50 as a change in sensitivity. One potential occurrence of this scenario involves MCF7 microtissues treated with paclitaxel which observed a rightward shift in MMLC50 (Table 1). Despite this shift, LME regression fits and PCA loadings confirm that the EdU proliferation assay was the most sensitive assay (Fig. 8 and Supplemental data, Figs. S1–S4).

Finally, this study aimed to incorporate a multimodal approach using experimental designs that are applicable to HTS studies. Technological advances in automation, culturing techniques, and analytical tools are improving the accessibility of HTS studies and their implementation in various fields. The platform used in this study utilized self-aggregating microtissues that reliably created 3D spheroids in 96-microwell arrays using cells from diverse origins. Other low-adhesion, multi-well plate formats have also become commercially available. Regardless of the selected platform, 3D cultures form reproducible samples that more closely recapitulate the in vivo microenvironment and maintain the clinical relevance of the model. These samples are often used in conjunction with simple, “add-mix-wait” assays for cost-effective outputs that are easily scalable for HTS studies. The analytical approaches demonstrated in this study, based on LME and PCA, can be readily applied to larger studies; furthermore, their outputs enable the quantifiable assessment of multi-assay responses in an intuitive and comprehensive manner. This will allow researchers to reveal complex, multifaceted pathways of cytotoxicity that may be overlooked by single assay analysis. For example, the multimodal approach may reveal unanticipated mechanistic effects of drugs leading to more accurate pharmacological and toxicological profiling resulting in the enhancement of clinical translatability and reduction of late-stage failures during drug development. Similarly, shifts in MMLC50 could be leveraged to identify which assays, and therefore which cellular perturbations, could be optimized in a study. The proposed analytical framework could also be implemented in personalized medicine screens on patient-derived tumor cultures to help select patient-specific therapies. Both LME and PCA outputs are capable of distinguishing between compound treatment effectiveness, and therefore effectively eliminate nonresponsive treatments from a larger screen. Ultimately, the proposed approach will allow researchers to answer hypothesis driven questions stemming from multimodal cytotoxicity data that could not be previously accomplished using standard approaches.

Methods

3D HepG2 tumor cell culture

Human hepatocellular carcinoma (HepG2, HB-8065) cells were purchased from the ATCC and expanded in T-75 culture flasks using Eagle’s Minimum Essential Media supplemented with 10% fetal bovine serum (FBS) and 1% Penicillin-Streptomycin. Cells were fed every other day and passaged weekly. Cells were only seeded from flasks between the fourth and tenth passage. Once ready for seeding, flasks containing HepG2 cells were washed twice with 2 mL of 0.25% trypsin in phosphate-buffered saline (PBS) and then incubated with 3 mL of 0.25% trypsin for 7 min. Once the cell layer was detached, 12 mL of warmed complete media was added to neutralize the trypsin, and the flask’s bottom was rinsed twice. The cell suspension was transferred to a 50 mL conical tube and centrifuged at 125 G for 6 min. The supernatant was removed, and the cell pellet was disaggregated by firmly grasping the conical tube with one hand and shaking the tube using the other hand. Once the pellet was disaggregated, 4 mL of complete media was added, and the cell suspension was mixed using a 5 mL serological pipette by repetitively pipetting six times against the bottom of the tube and six times against the tube wall. After cell counting, a sterile flask was seeded at 1.25 million cells in 15 mL of warmed complete media. To ensure uniform seeding, the cell suspension was mixed 3–4 times using a 1 mL stripette prior to adding to the flask media. The flask was then gently rocked and mixed to disperse the cells evenly prior to returning to the incubator.

3D HepG2 microtissues were made using Microtissues Inc. 24–96 small spheroid molds. Gels were made by pouring molten 2% ultrapure agarose dissolved in PBS into each mold. After the agarose solidified, the gels were extracted and placed in a 24-well plate. The wells containing gels were filled with 1 mL incomplete media (i.e., without FBS) and placed in the incubator. Gels were equilibrated by exchanging incomplete media three times every 24 h. Once ready for seeding, the appropriate dilution was made to seed every gel with 144,000 cells using a final volume of 70 µL per gel. Once the gels were seeded, the suspension was allowed to settle for at least 10 min in the biosafety cabinet with the lights off and then 20 min in the incubator before adding 1 mL of warmed complete media to every well. The cells were incubated at 37 °C and 5% CO₂ and cultured for three days without media changes before receiving treatment.

3D MCF7 tumor cell culture

Human breast adenocarcinoma (MCF7, HTB-22) cells were also purchased from the ATCC and expanded in T-75 culture flasks using Eagle’s Minimum Essential Media supplemented with 10% FBS, 1% Penicillin-Streptomycin, and 0.01 mg/mL human recombinant insulin. Passage and culture protocols were similar to those used for HepG2 cells with a few exceptions. Once ready for seeding, flasks containing MCF7 cells were washed once with 2 mL 0.25% trypsin and incubated with 2 mL of 0.25% trypsin for 7 min. After the cell layer was detached, 6 mL of complete media was added to the flask, and the flask bottom was rinsed twice. The cell suspension was transferred to a 50 mL conical tube and centrifuged at 125 G for 6 min. The supernatant was removed, and the cell pellet was disaggregated by firmly grasping the conical tube with one hand and shaking the tube using the other hand. Once the pellet was disaggregated, 4 mL of complete media was added, and the cell suspension was mixed using a 5 mL serological pipette by repetitively pipetting six times against the bottom of the tube and six times against the tube wall. After cell counting, a sterile flask was seeded at 6 million cells in 15 mL of warmed complete media. To ensure uniform seeding, the cell suspension was mixed 3–4 times using a 1 mL stripette prior to adding to the flask media. The flask was then gently rocked and mixed to disperse the cells evenly prior to returning to the incubator. Methods to create the agarose gels were identical to the previous section, except that gels were seeded at 24,000 cells per gel.

3D primary neonatal rat cortical cell culture

Primary neonatal rat cortical microtissues, also referred to as neurospheroids, were derived from primary cortical tissue extracted from neonatal CD rats (Charles River). The complete protocol and characterization of the model can be referenced from a previous study⁴⁴. All animal procedures were performed in accordance with relevant guidelines and regulations under a protocol approved by the Institutional Animal Care and Use Committee (IACUC) at Brown University. Neonatal rats were euthanized following the American Veterinary Medical Association (AVMA) Guidelines for the Euthanasia of Animals (2020). All methods are reported in compliance with the ARRIVE guidelines.

In brief, two male and two female neonatal pups were anesthetized using hypothermia. The effect of anesthesia was confirmed with a paw pinch, and the pups were euthanized via decapitation using a pair of sharp scissors. The cortices from the 0–2-day old rat pups of mixed sex (1:1 male-to-female) were dissected, cut into small pieces, and placed into 2 mg/mL papain solution dissolved in Hibernate-A without calcium chloride (Gibco) for 30 min at 30 °C. The papain solution was removed and replaced with 1.5 mL of regular chilled Hibernate-A buffer supplemented with 1x B-27 Plus (Gibco) and 0.5 mM GlutaMax (Gibco). The tissue was then triturated with a fire-polished Pasteur pipette to create a single cell suspension. The cell suspension was filtered using a 40 μm nylon mesh strainer and the filtrate was centrifuged at 150 G for 5 min. The supernatant was then carefully removed, and the cell pellet was resuspended in warmed complete media made from Neurobasal Plus (Gibco) supplemented with 1x B-27 Plus, 0.5 mM GlutaMax, and 1x Penicillin-Streptomycin. The cell suspension was filtered and centrifuged for a second time then resuspended in 5 mL warmed complete media. The resuspended solution was centrifuged again, and the pellet was resuspended in 10 mL of warmed complete media prior to cell counting.

Methods to create the agarose gels were similar to the previous sections; however, the respective media for neurospheroid culture was used. Unlike the previous protocols, cells were seeded at 384,000 cells per gel, and the cell suspension was allowed to settle in the gels for at least 30 min inside the incubator with intermittent gentle mixing every 10 min prior to the addition of 1 mL of warmed complete media. Media was changed 24 h after seeding and then every 2–3 days. Neurospheroids were incubated at 37 °C and 5% CO₂ and were cultured for seven days prior to treatment.

Compound treatment

Compounds that target four unique biological mechanisms to elicit cell death were utilized in the study. Oligomycin A (75351, Sigma Aldrich) and 2-Deoxy-D-glucose (D8375, Sigma Aldrich) were selected to inhibit metabolism through ATP synthase and glycolysis, respectively; Triton X-100 (X100, Sigma Aldrich) and melittin (M2272, Sigma Aldrich) were selected to compromise membrane integrity; cisplatin (P4394, Sigma Aldrich) and melphalan (M2011, Sigma Aldrich) were selected to crosslink DNA; and paclitaxel (T7191, Cell Signaling Technologies) was selected to over stabilize microtubules. Stock solutions were made using the following dilutions: 100 mg 2DG per 1 mL of warmed complete media, 5 mg oligomycin A in 93.28 µL dimethyl sulfoxide (D2650, Sigma Aldrich), 10 µL Triton X-100 into 990 µL warmed complete media, 5 mg melittin in 1.743 mL of ultrapure water, 25 mg cisplatin in 13.887 mL of 0.9% saline in ultrapure water, 100 mg melphalan in 16.717 mL of 200 proof ethanol with 300 µL of HCl, and 1 mg paclitaxel in 41.82 µL dimethyl sulfoxide. All stock solutions were stored in appropriate conditions and used within two months. Working solutions were made through serial dilutions using appropriate vehicles then adding equal volumes of the various working solutions to warmed complete media. A compound’s treatment was deemed beyond the practical limit if the percent of dimethyl sulfoxide were to exceed 0.5% of total volume. Media from each well in a 24-well plate was aspirated then 1 mL of warm, treated media was added to the appropriate wells. Final drug calculations accounted for the volume displacement caused by the agarose gel.

CellTiter-Glo 3D ATP content assay

The protocol for the luminescence-based CellTiter-Glo 3D assay (Promega) was followed based on manufacturer recommendations. To conduct this assay, we treated gels with 2DG or oligomycin A using serial dilutions for 24 h. After treatment, gels containing HepG2 and cortical microtissues were removed from the 24-well plate and placed in a 35 mm petri dish containing 3 mL of warmed complete media. Individual microtissues were removed from the gels by gently flushing media over the wells using a P-1000 pipette tip. Once an adequate number of microtissues were removed, a trimmed P-200 pipette tip was used to place eight randomly selected microtissues with 90 µL of media into eight separate wells in a white-walled, clear bottom 96-well plate. Note that the P-200 pipette tip was trimmed to prevent the shearing of the microtissues. Once the 96-well plate was loaded, the cross-sectional area of each microtissue was imaged using a phase-contrast microscope (TC100 Eclipse, Nikon). Assay reagent was thawed to room temperature prior to use and added at a 1:1 ratio to media to each well of the 96-well plate using a multichannel pipette. Plates treated with CellTiter-Glo 3D were mixed for 5 min at room temperature using an orbital shaker and incubated an additional 25 min at room temperature. A Cytation3 (BioTek) plate reader was used to measure the luminescent output from the assay using the default gain of 125 and 1 s integration time. Read height was adjusted to a control well for each plate run. Unlike neurospheroids and HepG2 microtissues, MCF7 cells formed loosely bound microtissues that cannot be flushed from their wells without disaggregation. Consequently, the CellTiter-Glo 3D assay was conducted on whole gels by transferring treated gels into an opaque, white-walled 24-well plate. Once transferred, 0.5 mL of warmed complete media was added to each well followed by 0.5 mL of room temperature assay reagent. The 24-well plates containing the MCF7 microtissues were incubated and imaged similarly to the other tissue culture types. Whole gels were not imaged for cross-sectional area.

Caspase-Glo 3/7 3D apoptotic assay

The protocols for the cisplatin and melphalan treated microtissues closely followed those used for the CellTiter-Glo 3D assay in the previous section. Gels were treated with either compound using serial dilutions for 24 h. After treatment, gels containing HepG2 and cortical microtissues were flushed, and eight randomly selected microtissues were placed in individual wells of a white-walled, clear bottom 96-well plate alongside 90 µL of media. The cross-sectional area of the HepG2 and cortical microtissues were also captured using phase-contrast microscopy. MCF7 microtissues were prepared similarly to those analyzed using the CellTiter-Glo 3D assay. Room temperature Caspase-Glo 3/7 3D (Promega) reagent was added at a 1:1 ratio for all tissue culture types and mixed for only 30 s using an orbital shaker. After shaking, the plates were incubated for 30 min at room temperature prior to analysis using the Cytation3. Identical reading parameters were used for both the Caspase-Glo 3/7 3D and CellTiter-Glo 3D assays.

Live/Dead Viability/Cytotoxicity assay

The mammalian Live/Dead Viability/Cytotoxicity assay (Invitrogen) was conducted on whole gels after treatment with Triton X-100 or melittin for 24 h. After treatment, gels were transferred to a sterile 24-well plate. Once transferred, 1 mL of warmed complete media premixed with 2 µL/mL ethidium homodimer and 0.5 µL/mL Calcein AM was added to each well while being protected from light. The cells were then incubated at 37 °C and 5% CO₂ for 30 min while protected from light. Gels were imaged using a Zeiss Axiovert 100 inverted fluorescence microscope using a 10X objective. A total of 12 (4 × 3) tiles were taken and stitched using Zeiss’s ZenPro software. Images were taken using brightfield, EGFP, and Alexa Fluor 568 channel presets. All images from the same experimental replicate and tissue type (i.e., passage) were captured at identical exposure times and excitation intensity. Minor adjustments were made to exposure time to keep image intensity consistent between replicates; however, this had a negligible effect on the statistical analysis since our signal processing involved normalizing the assay outputs by the control groups within each replicate and LME is capable of handling random effects related to different replicates. Additionally, prior to analysis, microtissues were segmented and background noise levels were adjusted to ensure consistent baseline readings.

Click-iT EdU proliferation assay

The Click-iT EdU proliferation assay (Invitrogen) was conducted by treating microtissues with paclitaxel for 24 h. After treatment, the wells were washed twice with warmed complete media for 15 min each to remove the treatment. The EdU component was incubated with the cultures at 37 °C and 5% CO₂ at a final concentration of 5 µM for 6 h for MCF7 and HepG2 cultures and 24 h for neurospheroids. Microtissues were then fixed in 4% paraformaldehyde 8% sucrose in PBS overnight. Fixed microtissues were rinsed three times with PBS for 30 min each and stored at 4 °C until they were stained. The microtissues were permeabilized by adding 1 mL 0.2% Triton X-100 in PBS twice for two hours each. Wells were washed with PBS twice for 15 min each to remove the permeabilization solution. Following the manufacturer’s protocol, 1 mL of Click-iT reagent cocktail was added to each well while protected from light. After one hour, the wells were rinsed with rinse buffer for 15 min. The supplied Hoechst was diluted to 1X in PBS, and 1 mL was added to each well for one hour. Wells were washed twice with PBS for 15 min each to remove the Hoechst solution. Finally, the stained samples were wrapped in parafilm and stored at 4 °C, protected from light, until they were imaged.

Prior to imaging, the gel walls were cut using a scalpel blade to expose the portion of the gel containing the microwells. In a 24-well plate, a drop of PBS was added to the center of each well, and the cut gel was inverted over the PBS drop. The PBS drop was used to ensure no air bubbles were produced when the gel was flipped but was not large enough to cause the gel to float. Warmed 2% agarose was added over the gel to secure it to the bottom of the well, and 1 mL of PBS was added to keep the well hydrated. The well plate was then imaged using a Zeiss Axiovert 100 inverted fluorescence microscope using a 10X objective. A total of 12 (4 × 3) tiles were taken and stitched using Zeiss’s ZenPro software. Images were taken in the brightfield, Cy5, and DAPI channel presets. Image acquisition, exposure adjustment, and signal processing were conducted similarly to the approach for Live/Dead fluorescence imaging.

Statistical analysis

Dose-response curves were generated for each treatment and “gold-standard” assay combination by plotting the normalized assay output against the concentration of treatment used in the log scale. HepG2 and cortical microtissue data from the CellTiter-Glo 3D assay were extracted from eight microtissues per condition. All HepG2 and cortical microtissue assay outputs were normalized by their cross-sectional area. MCF7 microtissue data was extracted from three intact gels per condition. All data was then normalized to the control groups and fitted to a NLME model in MATLAB (MathWorks) using Eq. 1. The 95% confidence intervals were reported for each experiment and line fit. Caspase-Glo 3/7 3D assay data was analyzed similarly to the CellTiter-Glo 3D assay data. Caspase assay data was only normalized to cross-sectional area (for HepG2 and cortical microtissues alone) and not controls.

Images acquired from the Live/Dead assay captured a total of 30–40 microtissues per field of view from which the automated segmentation protocol extracted 20 microtissues per condition with a cross-sectional area closest to the median. The unbiased segmentation utilized circle detection, adaptive thresholding-based binarization, and morphological opening/closing to select spheroids for analysis. Image processing was conducted in MATLAB using the live intensity fraction as described by Eq. (2) where live intensity was calculated from Calcein in the EGFP channel, and dead intensity was calculated from ethidium homodimer using the Alexa Fluor 568 channel. The live intensity fraction was then normalized to the controls and fitted to a NLME model using Eq. (3). The 95% confidence intervals were calculated for each experiment and line fit. Image acquisition and analysis for the EdU proliferation assay was similar to that performed for the Live/Dead assay. In brief, the proliferative capacity was calculated with Eq. (4) using the Cy5 and DAPI channel presets to measure the intensity of tagged EdU and Hoechst, respectively. A total of 20 microtissues per condition were analyzed using the same segmentation protocol used for the Live/Dead assay. The data was normalized to the controls and fitted to a NLME model using Eq. (3). The 95% confidence intervals were also provided for the EdU assay data.

The LC25, LC50, and LC75 were extracted from the NLME model regressions at which 25, 50, and 75% cell death was achieved relative to controls. These concentrations were then used for the multi-assay study which tested all assay-tissue-treatment combinations. The data from the ATP and Caspase assays collected from HepG2 and cortical microtissues provided results from eight microtissues per condition and two independent experiments. Data from MCF7 microtissues produced results from three gels per condition and two independent experiments. The Live/Dead and EdU assays produced results from 20 microtissues per condition and two independent experiments (only HepG2 microtissues analyzed by the Live/Dead assay had three independent experiments). A total of 8,758 datapoints were collected and not the anticipated 9,408 since not all compounds could elicit every LC value without exceeding solubility guidelines. After all the LC data was plotted and normalized to its respective controls (with the exception of Caspase assay data), the data was fitted to a linear regression using a LME model with 95% confidence intervals. A fit was deemed significant if it had a p < 0.05 and slope greater than 0.1 (i.e., had a slope greater than 10% the expected slope of 1.0). In order to maintain linearity assumptions for all LME model fits, we implemented an exclusion criterion. In some instances, assay output plateaued at higher treatment concentrations. To determine if the LC75 was to be excluded from the model fit, we ran a single-tailed t-test between the LC75 and LC50 datapoints when both passed normality and variance assumptions. If either dataset failed the assumptions, we ran the statistical analysis on the bootstrapped data. LC75 datapoints were excluded from the fit if they were not significantly different from the LC50 (p > 0.05). If LC75 data was excluded, then the same exclusion criterion was repeated between the LC50 and LC25 datasets. The slopes from the linear regressions were then collected and graphed. Slope values for most assays were expected to be negative since increased treatment concentration resulted in decreased viability output. The Caspase assay had a positive slope since increased treatment concentration resulted in increased Caspase 3/7 activity which is indicative of more apoptotic activity. For an integrative slope comparison across assays, all Caspase assay slopes were normalized by the mean of the slopes of the standard-pair drugs (cisplatin and melphalan) and then multiplied by -1.0, such that the average slope of the standard-pair drugs becomes −1.0.

PCA and its subsequent statistical tests were conducted in JMP Pro 16 (JMP Statistical Discovery LLC). The same multi-assay dataset (N = 8,758) used for the LME model fits were used for PCA. Prior to PCA, Live/Dead and EdU assay datasets were truncated to match the sample sizes of the ATP and Caspase assays. To achieve this, eight out of 20 microtissues per condition were randomly selected for HepG2 and cortical microtissues analyzed by the Live/Dead and EdU assays (n = 8 microtissues per condition; two independent experiments; N = 64 datapoint maximum depending on available LCs). Similarly, MCF7 datasets analyzed by Live/Dead and EdU assays were truncated by randomly selecting three out of 20 datapoints (n = 3 datapoints per condition; two independent experiments; N = 24 datapoint maximum depending on available LCs). Prior to conducting the PCA, all data was normalized to its control with the exception of Caspase assay data which was normalized by taking 1-x/mean(LC75). PCA was conducted on the covariances and the eigenvalues, eigenvectors, percentage of variance explained, and results from the Bartlett’s test of sphericity (Chi-square < 0.05 considered significant) for the first principal component were extracted. In addition, PC1-PC2 data transformations for all datapoints were saved, and their scatterplots were made in SigmaPlot 15.0 (Systat Software Inc.). Next, the PC1 transformed data was plotted against the lethal concentration of any given treatment and fitted to a second order polynomial in SigmaPlot 15.0. The new MMLC50 was derived by taking two-thirds of the data range and subtracting it from the value of the PC1 at the control then projecting that value onto the x axis (gray lines, Fig. 7b). The MMLC50 lethal concentration was converted to molar concentration in Table 1 by solving the equation of the line extracted from the NLME fits from Figs. 1, 2, 3 and 4 and solving for molar concentration. PCA loadings were calculated by Eq. (5) and plotted as well.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

Richards, C. H., Mohammed, Z., Qayyum, T., Horgan, P. G. & McMillan, D. C. The prognostic value of histological tumor necrosis in solid organ malignant disease: A systematic review. Future Oncol. 7, 1223–1235 (2011).
Article CAS PubMed Google Scholar
Langner, C. et al. Tumor necrosis as prognostic indicator in transitional cell carcinoma of the upper urinary tract. J. Urol. 176, 910–914 (2006).
Article PubMed MATH Google Scholar
Chen, J. et al. Prognostic value of tumor necrosis based on the evaluation of frequency in invasive breast cancer. BMC Cancer 23, 530 (2023).
Article CAS PubMed PubMed Central MATH Google Scholar
Ling, Y. et al. Tumor necrosis as a poor prognostic predictor on postoperative survival of patients with solitary small hepatocellular carcinoma. BMC Cancer 20, 607 (2020).
Article CAS PubMed PubMed Central MATH Google Scholar
Sengupta, S. et al. Histologic coagulative tumor necrosis as a prognostic indicator of renal cell carcinoma aggressiveness. Cancer 104, 511–520 (2005).
Article PubMed MATH Google Scholar
Condeelis, J. & Pollard, J. W. Macrophages obligate partners for tumor cell migration, invasion, and metastasis. Cell 124, 263–266 (2006).
Article CAS PubMed Google Scholar
Wojtowicz-Praga, S. Reversal of tumor-induced immunosuppression: A new approach to cancer therapyy. J. Immunother. 20, 165 (1997).
Article CAS PubMed MATH Google Scholar
Jang, J. W. et al. Serum interleukin-6 and C-reactive protein as a prognostic indicator in hepatocellular carcinoma. Cytokine 60, 686–693 (2012).
Article CAS PubMed MATH Google Scholar
Kulbay, M., Paimboeuf, A., Ozdemir, D. & Bernier, J. Review of cancer cell resistance mechanisms to apoptosis and actual targeted therapies. J. Cell. Biochem. 123, 1736–1761 (2022).
Article CAS PubMed Google Scholar
Todaro, M. et al. Apoptosis resistance in epithelial tumors is mediated by tumor-cell-derived interleukin-4. Cell. Death Differ. 15, 762–772 (2008).
Article CAS PubMed MATH Google Scholar
Neophytou, C. M., Trougakos, I. P., Erin, N. & Papageorgis, P. Apoptosis deregulation and the development of cancer multi-drug resistance. Cancers 13, 4363 (2021).
Article CAS PubMed PubMed Central Google Scholar
Riss, T. L. et al. Cell viability assays. In Assay Guidance Manual (ed Markossian, S.) (Eli Lilly & Company and The National Center for Advancing Translational Sciences, 2004).
Google Scholar
Ouyang, L. et al. Programmed cell death pathways in cancer: A review of apoptosis, autophagy and programmed necrosis. Cell. Prolif. 45, 487–498 (2012).
Article CAS PubMed PubMed Central MATH Google Scholar
Méry, B. et al. In vitro cell death determination for drug discovery: A landscape review of real issues. J. Cell. Death 10, 1179670717691251 (2017).
Article PubMed PubMed Central MATH Google Scholar
Galluzzi, L. et al. Guidelines for the use and interpretation of assays for monitoring cell death in higher eukaryotes. Cell. Death Differ. 16, 1093–1107 (2009).
Article CAS PubMed MATH Google Scholar
Tasdemir, E. et al. Regulation of autophagy by cytoplasmic p53. Nat. Cell. Biol. 10, 676–687 (2008).
Article CAS PubMed PubMed Central MATH Google Scholar
Dörr, J. R. et al. Synthetic lethal metabolic targeting of cellular senescence in cancer therapy. Nature 501, 421–425 (2013).
Article ADS PubMed MATH Google Scholar
Stepanenko, A. A. & Dmitrenko, V. V. Pitfalls of the MTT assay: Direct and off-target effects of inhibitors can result in over/underestimation of cell viability. Gene 574, 193–203 (2015).
Article CAS PubMed MATH Google Scholar
Malo, N., Hanley, J. A., Cerquozzi, S., Pelletier, J. & Nadon, R. Statistical practice in high-throughput screening data analysis. Nat. Biotechnol. 24, 167–175 (2006).
Article CAS PubMed Google Scholar
Bradley, J. V. The complexity of nonrobustness effects. Bull. Psychon. Soc. 22, 250–253 (1984).
Article MATH Google Scholar
Lix, L. M., Keselman, J. C. & Keselman, H. J. Consequences of assumption violations revisited: A quantitative review of alternatives to the one-way analysis of variance F test. Rev. Educ. Res. 66, 579–619 (1996).
MATH Google Scholar
Jolliffe, I. T. & Cadima, J. Principal component analysis: A review and recent developments. Philos. Trans. Math. Phys. Eng. Sci. 374, 20150202 (2016).
ADS MathSciNet MATH Google Scholar
Björklund, M. Be careful with your principal components. Evolution 73, 2151–2158 (2019).
Article PubMed MATH Google Scholar
Nagle, P. W., Plukker, J. T. M., Muijs, C. T., van Luijk, P. & Coppes, R. P. Patient-derived tumor organoids for prediction of cancer treatment response. Semin. Cancer Biol. 53, 258–264 (2018).
Article CAS PubMed Google Scholar
Dekkers, J. F. et al. Characterizing responses to CFTR-modulating drugs using rectal organoids derived from subjects with cystic fibrosis. Sci. Transl. Med. 8, 344ra84–344ra84 (2016).
Article PubMed Google Scholar
Huang, L. et al. Ductal pancreatic cancer modeling and drug screening using human pluripotent stem cell– and patient-derived tumor organoids. Nat. Med. 21, 1364–1371 (2015).
Article CAS PubMed PubMed Central MATH Google Scholar
Melissaridou, S. et al. The effect of 2D and 3D cell cultures on treatment response, EMT profile and stem cell features in head and neck cancer. Cancer Cell. Int. 19, 16 (2019).
Article PubMed PubMed Central Google Scholar
Pajak, B. et al. 2-Deoxy-d-glucose and its analogs: From diagnostic to therapeutic agents. Int. J. Mol. Sci. 21, 234 (2019).
Article PubMed PubMed Central MATH Google Scholar
Lardy, H. A., Johnson, D. & McMurray, W. C. Antibiotics as tools for metabolic studies. I. A survey of toxic antibiotics in respiratory, phosphorylative and glycolytic systems. Arch. Biochem. Biophys. 78, 587–597 (1958).
Article CAS PubMed Google Scholar
Mattei, B., Lira, R. B., Perez, K. R. & Riske, K. A. Membrane permeabilization induced by Triton X-100: The role of membrane phase state and edge tension. Chem. Phys. Lipids 202, 28–37 (2017).
Article CAS PubMed Google Scholar
van den Bogaart, G., Guzmán, J. V., Mika, J. T. & Poolman, B. On the mechanism of pore formation by melittin. J. Biol. Chem. 283, 33854–33857 (2008).
Article PubMed PubMed Central Google Scholar
Dasari, S. & Tchounwou, P. B. Cisplatin in cancer therapy: Molecular mechanisms of action. Eur. J. Pharmacol. 740, 364–378 (2014).
Article CAS PubMed PubMed Central MATH Google Scholar
Falco, P. et al. Melphalan and its role in the management of patients with multiple myeloma. Expert Rev. Anticancer Ther. 7, 945–957 (2007).
Article CAS PubMed MATH Google Scholar
Schiff, P. B., Fant, J. & Horwitz, S. B. Promotion of microtubule assembly in vitro by taxol. Nature 277, 665–667 (1979).
Article ADS CAS PubMed MATH Google Scholar
Warburg, O., Wind, F. & Negelein, E. The metabolism of tumors in the body. J. Gen. Physiol. 8, 519–530 (1927).
Article CAS PubMed PubMed Central MATH Google Scholar
Vander Heiden, M. G., Cantley, L. C. & Thompson, C. B. Understanding the Warburg Effect: The metabolic requirements of cell proliferation. Science 324, 1029–1033 (2009).
Article ADS CAS PubMed PubMed Central MATH Google Scholar
Lauritzen, S. P., Boye, T. L. & Nylandsted, J. Annexins are instrumental for efficient plasma membrane repair in cancer cells. Semin. Cell. Dev. Biol. 45, 32–38 (2015).
Article CAS PubMed MATH Google Scholar
Brito, C., Cabanes, D., Sarmento Mesquita, F. & Sousa, S. Mechanisms protecting host cells against bacterial pore-forming toxins. Cell. Mol. Life Sci. 76, 1319–1339 (2019).
Article CAS PubMed Google Scholar
Osborne, M. R. & Lawley, P. D. Alkylation of DNA by melphalan with special reference to adenine derivatives and adenine-guanine cross-linking. Chem. Biol. Interact. 89, 49–60 (1993).
Article CAS PubMed MATH Google Scholar
Takano, M. et al. Paclitaxel-resistance conferred by altered expression of efflux and influx transporters for paclitaxel in the human hepatoma cell line, HepG2. Drug Metab. Pharmacokinet. 24, 418–427 (2009).
Article CAS PubMed Google Scholar
Chae, S., Kim, Y. B., Lee, J. S. & Cho, H. Resistance to paclitaxel in hepatoma cells is related to static JNK activation and prohibition into entry of mitosis. Am. J. Physiol. Gastrointest. Liver Physiol. 302, G1016–G1024 (2012).
Article CAS MATH Google Scholar
Hong, D., Balzano, L. & Fessler, J. A. Asymptotic performance of PCA for high-dimensional heteroscedastic data. J. Multivar. Anal. 167, 435–452 (2018).
Article MathSciNet PubMed PubMed Central MATH Google Scholar
Bispo, R. & Marques, F. Stability of principal components under normal and non-normal parent populations and different covariance structures scenarios. J. Stat. Comput. Simul. 93, 1060–1076 (2023).
Article MathSciNet MATH Google Scholar
Dingle, Y. T. L. et al. Three-dimensional neural spheroid culture: An in vitro model for cortical studies. Tissue Eng. Part C Methods 21, 1274–1283 (2015).
Article CAS PubMed PubMed Central MATH Google Scholar

Download references

Acknowledgements

This work is supported by the National Cancer Institute award R01CA268366. We would also like to thank Dr. Eric M. Darling and Dr. Vera C. Fonseca for giving us access to their plate reader and providing training for it.

Author information

Authors and Affiliations

School of Engineering, Brown University, Providence, RI, USA
Ahbid Zein-Sabatto, Jeffrey R. Morgan & Jonghwan Lee
Institute for Biology, Engineering, and Medicine, Brown University, Providence, RI, USA
Ahbid Zein-Sabatto, Diane Hoffman-Kim, Jeffrey R. Morgan & Jonghwan Lee
Department of Pathology and Laboratory Medicine, Brown University, Providence, RI, USA
Katerina St. Angelo, Samantha J. Madnick & Jeffrey R. Morgan
Department of Neuroscience, Brown University, Providence, RI, USA
Diane Hoffman-Kim
Carney Institute of Brain Sciences, Brown University, Providence, RI, USA
Diane Hoffman-Kim & Jonghwan Lee
Center for Alternatives to Animals in Testing, Brown University, Providence, RI, USA
Diane Hoffman-Kim, Jeffrey R. Morgan & Jonghwan Lee

Authors

Ahbid Zein-Sabatto
View author publications
Search author on:PubMed Google Scholar
Katerina St. Angelo
View author publications
Search author on:PubMed Google Scholar
Samantha J. Madnick
View author publications
Search author on:PubMed Google Scholar
Diane Hoffman-Kim
View author publications
Search author on:PubMed Google Scholar
Jeffrey R. Morgan
View author publications
Search author on:PubMed Google Scholar
Jonghwan Lee
View author publications
Search author on:PubMed Google Scholar

Contributions

A.Z.S. and J.L. designed the study and conducted all statistical analysis. J.L. prepared all MATLAB codes. A.Z.S. prepared all experiments, collected all data, prepared all neurospheroid culture, and wrote the manuscript. K.S.A. and S.J.M. prepared and maintained all HepG2 and MCF7 cultures. J.L., D.H.K., and J.R.M. supervised the study and reviewed overall results. All authors revised the manuscript.

Corresponding author

Correspondence to Jonghwan Lee.

Ethics declarations

Competing interests

J.R.M. has an equity interest in Microtissues, Inc. and XM Therapeutics, Inc. These relationships have been reviewed and are managed by Brown University in accordance with its conflict of interest policies. All other authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Zein-Sabatto, A., St. Angelo, K., Madnick, S.J. et al. Multi-assay assessment of cytotoxicity reveals multiple mechanisms of action in 3D microtissues. Sci Rep 15, 3090 (2025). https://doi.org/10.1038/s41598-025-86792-4

Download citation

Received: 21 April 2024
Accepted: 14 January 2025
Published: 24 January 2025
DOI: https://doi.org/10.1038/s41598-025-86792-4