Introduction

Despite their widespread use in biomedical research and drug screening, 2D culture-based methods have inherent limitations that hinder their ability to accurately represent the complexity of in vivo environments1,2. In tumor tissues, cells often exhibit different patterns of proliferation and differentiation compared to 2D-grown cell cultures, making it difficult to extrapolate truly reliable results that represent in vivo conditions3.

Alternative platforms, such as 3D cell culture (3DCC), offer solutions to overcome these limitations and improve the translation of preclinical findings to clinical practice. Among 3D models, tumor spheroids present crucial advantages like reproduction of tissue architecture, restoration of cell-to-cell interactions, exhibition of gradients of oxygen and nutrients, as well as factors that mimic in vivo tumor microenvironment1,2,4. Spheroids can also exhibit correlation with tissues from the mechanobiological aspect, such as stiffness gradients5. Additionally, cells within spheroids are constantly undergoing proliferation and differentiation, further mimicking the in vivo environment. However, the use of spheroids and organoids also presents challenges related to their biological variability, which strongly affects researchers’ results. To overcome these limitations, other than the commonly adopted operator-dependent visual inspection, clearly biased by the operator experience, the challenge is to define aspects that can be quantified without the need for expensive and time-consuming analysis. On that note, daily lab routines rely mainly on seeding number counting, which is considered at time 0 only, and diameter measurements, mainly obtained after 2D-based brightfield imaging5. Since the need is to characterize 3D objects, decision making choices based on diameter-only lead to larger variability and broad standard deviation data outputs, specifically when cross-sectional related aspects are not considered. To address these challenges, Cristaldi et al. validated a system for the simultaneous measure of spheroids’ mass density, diameter and weight6. Such biophysical characterization has changed the perception of 3DCC heterogeneity, as well as opened the door for biological correlation on multiple aspects, from permeation monitoring, associated to the mechanobiology7, to extracellular matrix (ECM) growth kinetics and ratio to cell number8, from drug activity8,9 to tissue regeneration10.

Recently, the notable work by Sousa et al.11 extensively described the crucial impact on the biophysical characterization of aggressive prostate cancer for patient derived organoids (PDOs), as well as the possibility to use mass density over size trends to characterize phenotypes diversity and separation between luminal and solid organoids. The W8 biophysical characterization of 3DCC not only defines the novel approach for 3 dimensional samples, but it is clearly projected to become a simple yet crucial step for scientists to adopt during multiple stages of their daily lab workflows. To further strengthening its potential, in this work we disclose several technological implementations of the W8 System as well as the method for the latest improvements in data interpretation while retaining simplicity. Here we prove the impact of the Mass Density, Weight and Size outputs when combined with the commonly adopted Principal Component Analysis (PCA) and Hierarchical Cluster Analysis (HCA).

The PCA is a statistical tool that aims at simplifying the interpretation of multiparametric analysis12,13, whereas HCA allows to simply visualize sample clusterization13,14. PCA-Based studies, other than playing a crucial role for drug delivery- / discovery- related studies12, and large-scale data analysis15, were also adopted for bio-focused purposes16,17,18. HCA has been used in the quantitative analysis of the interaction between environment, neighborhood, and cellular state in 3D spheroids19, as well as for mass spectrometry imaging of human brain tumors20.

Here we propose the PCA-coupled Biophysical Characterization (PCA-BC) method, and demonstrate its simple yet accurate capability to define and measure 3DCC population’s heterogeneity. This will give researchers analytical data to investigate and correlate to their samples distribution, by discriminating clusters and subpopulations. Specifically, it defines what could be the new standard to highlight 3DCC’s variability on multiple layers: (i) within the same population; (ii) between different batches; (iii) and even among different operator (here referred to as inter- and intra- batches and inter- and intra-operator performances). A notable impact on 3DCC standardization and quality control can be perceived, which is further enhanced by the precise subpopulation separation capability for more precise in depth studies down the workflow line.

Results

PCA discriminates different cell lines depending on biophysical characteristics

Spheroids derived from the A549 and LoVo cell lines, herein referred to as A-Sph and L-Sph respectively, were selected based on the statistical differentiation, and rationale (see Figs. S1.1–S1.6, Table S1, Supplementary Material).

As the main initial step, the PCA analysis is presented for the entire selected dataset, encompassing Mass Density, Size, and Weight variables of 90 spheroids (45 per cell line), cultivated by the 3 operators (Fig. 1A). Each operator conducted 3 replicates, resulting in 5 spheroids per repetition. No removal of PCA-related outliers occurred due to the previous statistical analysis (see Methods). Figure 1A illustrates the automated HCA through color separation after running a 2-subgroup analysis in order to attempt the automatic recognition and separation by cell line. To enhance PCA graph interpretation, increasing trends in Mass Density and Size are depicted at the bottom left and right sides in Fig. 1A. Such trends originate from the Loadings Plot graph (Fig. 1B), elaborately described in the Supplementary Materials - S1, and maintain their consistency throughout all PCAs. Oval dotted areas are also included to emphasize the data distribution of each cell line concerning the trend of the variables. The biophysical analysis permitted to obtain the average and standard deviation values for each variable related to the spheroids produced from A549 and LoVo cell lines as 1034 ± 7 fg/mm3 and 1014 ± 2 fg/mm3 in Mass Density, 4488.9 ± 2325.5 ng and 65005.9 ± 36157.6 ng in Weight and 197.5 ± 34.3 mm and 473.0 ± 112.4 mm in Size, respectively.

Fig. 1
figure 1

Visualization of the entire dataset. (A) PCA scores plot and HCA separation; (B) Loadings plots of the variables.

Full size image

PCA-BC can highlight intra and inter-operator variability

Figure 2 displays the data distribution of the samples produced by the operator Y (colored dots), compared to the entire dataset (gray dots). The automatic clusterization was selectively performed on the Y-operator’s dataset and defined for 6 subgroups, 3 replicates for each of the 2 cell lines. The same procedure is conducted for samples produced by Operators X and Z, and further discussed in the supplementary information paragraph 3 (Figs. S3.1 and S3.2, Supplementary Material).

Fig. 2
figure 2

Representative work of operator Y. PCA of the spheroids produced by the operator Y, colored by HCA (6 subgroups), and overlapped on the entire dataset.

Full size image

The distinct data dispersion of A-Sph derived from the efforts of the three operators is shown in Fig. 3. This is highlighted by the isolated color selection for each operator and data overlay on the entire PCA dataset. Specifically, green, blue and orange colors are assigned to isolate operators X, Y and Z respectively (Fig. 3A–C). The 3 repetitions are included for each operator.

Fig. 3
figure 3

Data dispersion for each operator. Visualization of the dispersion related to the A-Sph population, with respect to the entire PCA dataset, for operator X (A), Y (B) and Z (C).

Full size image

PCA-BC exposes the impact of the operator depending on the biophysical heterogeneity of the population

Within Fig. 4, two separate PCAs are presented. Figure 5A focuses on isolated A-Sph data, while Fig. 4B concentrates on L-Sph. The color analysis within each graph emphasizes differences between the automatic clusterization obtained by the HCA and the data correlated to the known operator by the authors. Specifically, samples displayed by Red, Blue, and Green dots are automatically clustered by the 3-groups HCA (one for each operator). Differently, dots included within the Red, Blue, or Green circled areas, regardless the automatic color of the dot, are assigned to the operator who produced them.

Fig. 4
figure 4

Comparison between HCA and operator-based clusterization. Automated clusterization (HCA) and grouping by known operator of A549 (A) and LoVo (B) are shown.

Full size image

As a conclusive step, a final experimental design was conducted to investigate on spheroid’s population viability, comparing prior and post the W8 Biophysical Characterization. Although a previous preliminary study was proposed using the AlamarBlue™ assay21, here we explored the same outcome by adopting for the first time the CellTiter-Glo® 3D Cell Viability Assay, a reference 3D drug testing validated in a previous study by the authors22. The viability results obtained for the A-Sph by the CellTiter-Glo® 3D confirmed the absence of statistically significant differences between prior and post W8 analysis when comparing populations from the same cell line (Fig S4.1). Moreover, the larger viability data dispersion obtained for the L-Sph was observed also in this experiment further highlighting the larger intrinsic heterogeneity of such spheroid population.

Discussion

The Principal Component Analysis outcomes, Score Plot and Loading Plot, derived from the measured biophysical metrics of Mass Density, Size, and Weight (variables), provide a clear overview of the spheroids’ biophysical distribution (Fig. 1A and B). Most importantly, it permits detailed discussions across various aspects, which is further enhanced by the application of the automated color-based clusterization (Hierarchical Cluster Analysis). The influence of each variable with respect to the 2 Principal Components is established by the Loading Plot (Fig. 1B). Concerning PC1, as suggested from the mathematical relation, mass density is opposite to the other two variables, with impact of similar magnitude (in absolute value). On the other hand, all variables differ when compared to PC2 for the chosen conditions (Fig. S1.1, Supplementary Material). Specifically, spheroids with higher Mass Density are shifted toward the upper right region of the Score Plot, whereas the spheroids Diameter and Weight increase counterbalances Mass Density trend by shifting object’s position to the middle-up left region, even though with a slightly different angle with respect to each other (color-gradient dependent variable PCAs depicted in Figs. S1.5 and S1.6, Supplementary Material).

Knowing that each dot in Fig. 1A represents the biophysical footprint of a specific spheroid, a large amount of information can be determined from the dot’s distribution in Fig. 1A, which highlights the impact and simplicity of the proposed PCA-BC method. Based on data positioning only, in the Score Plot, a clear 2-groups separation can be easily observed. This separation was confirmed when performing a 2-group HCA, with the intent to evaluate whether A-Sph and L-Sph would be automatically discriminated. As a result, not only the clusterization is distinctly evident, but the efficiency is outstanding; among 90 spheroids, only 2 A-Sph were clustered within the LoVo group (the two green-circled blue dots in Fig. 1A), and no miss clustering occurred for L-Sph, showcasing an overall data analysis efficiency of 97.8%. It is not secondary to consider that, following the chosen rationale, outliers highlighted from the PCA were not removed, as the statistical analysis was performed prior to the dataset selection for PCA and based on the biophysical characterization. Moreover, although the selection of the best 3 samples per sub-group would have eliminated the 2 wrongly clustered A-Sph, all the 5 produced were maintained. This was decided to make PCA and HCA separation more difficult and to highlight biological heterogeneity for the purpose of the impact of the work. Notably, the initial PCA is already enough detailed to grasp insights also on the biophysics and heterogeneity of both inter- and intra-population. This can be graphically perceived in a simple manner by evaluating the spheroid dots distribution, position and orientation, per cell line, within the Score Plot (Fig. 1A). In detail, the A-Sph distribution (included in the green dotted oval in Fig. 1A), is located on the upper right area of the Score plot, indicating an overall higher Mass Density than L-Sph. From the oval orientation it can also be determined that diameters are more consistent within the close A-Sph cluster, whereas a slightly broader distribution occurs with respect to the Mass Density. Differently, L-Sph distribution is evidently much broader (samples within the blue dotted oval), and presents an almost orthogonal orientation with respect to the A-Sph. This not only indicates an overall lower Mass Density than A-Sph, but demonstrates the larger heterogeneity of L-Sph, which is enhanced for the Diameter and less pronounced in Mass Density. This multiparametric evaluation offers researchers a robust approach to quantify intrinsic inter-cell line heterogeneity. The generated dispersion plot facilitates the visual assessment of each cell line’s tendency to generate homogeneous spheroid subpopulations. This information can be critical during the pre-selection phase of experimental design, guiding the selection of the most suitable cell lines for the further steps (i.e. drug testing assays, genomics, proteomic and so on). To enhance the discriminatory power of the method, a PCA for each of the three operators was performed, followed by automatic clustering (HCA). Results on the work of operator Y were selected as a representative data (Fig. 2). The HCA was set to include all separated procedures related to generating spheroids from two cell lines (A-Sph and L-Sph) in triplicate (each referred to as a, b, or c), thus imposing a 6 groups separation. Such investigation highlights how A-Sph consistently forms a unified cluster across all three repetitions (15 Spheroids in total), except for two objects from repetition c. This is an evident indication of the high biophysical uniformity for the A-Sph line, also suggesting a potential anomaly during repetition c (key indication for QC related information). In contrast, L-Sph forms four distinct clusters. Notably, although englobing spheroids from other repetitions, only spheroids from repetition are entirely included within the same cluster (shown in purple). Differently, spheroids derived from repetitions b and c are scattered across multiple clusters, indicating a significant biophysical heterogeneity for 3DCCs derived from the LoVo cell line. Furthermore, studies on the work of operator X and Z are discussed in AI3 (see Figs. S3.1 and S3.2, Supplementary Material), each one interestingly denoting a different trend. It is crucial to notice that, even in the case of the most broad distribution with respect to the automatic clusterization, which occurs for operator Z, A-Sph and L-Sph are automatically separated, confirming the robustness of the PCA-BC method to distinguish between the two cell lines regardless of the impact of the operator nor the intrinsic heterogeneity of the cell line. The HCA analysis here proposed indeed provide crucial intra-batch information for researchers, highlighting the importance of rigorous quality control to ensure consistent biophysical homogeneity across spheroids generated from multiple batches, different time points, and various laboratories. Finally, since A-Sph demonstrated a more uniform distribution, the biophysical-PCA approach can be adopted to visualize the impact of each operator. Figure 4 directs attention to the individual contributions (depicted as colored dots) within the A-Sph-associated region of the PCA graph. This comparative analysis yields several noteworthy observations. Firstly, spheroids generated by Operator X exhibit a cohesive clustering, with only one marginally deviating from the group (AXb4). Similarly, the majority of samples from Operator Y form a well-defined cluster, although two spheroids stand out distinctly in a different area as previously mentioned, both originating from Rep c (AYc2 and AYc3). Additionally, it is evident that, on average, Operator X produces spheroids with a higher Mass Density compared to those generated by Operator Y. In contrast, Operator Z reveals a divergent trend, characterized by a broader data distribution, indicative of a generation of less reliable batches across the three repetitions. From comparing inter-batch variability with the PCA-BC method, it can be seen that A-Sph permits to maintain a high reproducibility between different batches, which displays high potential towards quality control and batch-to-batch standardization protocols, but also highlights the impact of the operator, which we further investigated. Specifically, we conducted two single-group PCAs, one for A-Shp and one for L-Sph. This approach aimed to elucidate distinct cell line behaviors and correlations between spheroids produced by known operators versus those derived from automated clustering. It is notable how, in the case of A-Sph (Fig. 4A), a discernible trend of consistency emerges, with spheroids from operators Y, Z, and X aligning in a left-to-right progression within the PCA graph. Specifically, regarding operator Y, two spheroids are automatically assigned to Z. Meanwhile, the work of operator Z incorporates two spheroids assigned to Y and three assigned to X, indicating a larger dispersion and thus heterogeneity. Finally, only one spheroid from X is miss clustered and assigned to Z. In contrast, this coherence is not evident in the case of L-Sph, likely attributable to the inherent heterogeneity of the cell line itself (Fig. 4B). Apart from a modest clustering capacity observed in the work of operator X, the data exhibit wide distribution within the graph, impeding the Hierarchical Cluster Analysis (HCA) to autonomously correlate each known operator. It is now evident how the work of the operator plays a crucial role in obtaining a uniform population. Here we demonstrated how the PCA-BC method would be efficient enough to highlight operator-dependent interventions. However, we have verified how the inverse is also true, and the nature of the samples impacts the accuracy of the operator due to intrinsic biological variability.

Conclusions

While 3D tumor models are gaining traction in drug discovery for selecting compounds for animal testing, their high intrinsic variability remains a major hurdle, limiting their potential and hindering the translation of preclinical findings to actual clinical efficacy. Moreover, this research aligns with ongoing efforts to reduce reliance on animal testing by advancing sophisticated 3D in vitro models, reflecting the shift toward more predictive and ethically responsible preclinical testing. In this study, we propose a novel approach to address this challenge by employing Principal Component Analysis (PCA) on three objectively measurable parameters of scaffold-free cultures: mass density, weight, and diameter. By applying this widely used statistical method, we aim to evaluate and quantify the inter- and intra-sample population as well as, the inter- and intra-operator variability within the 3D tumor model. To assess these parameters (mass density, weight, and diameter) across a heterogeneous population of spheroids derived from different commercial tumor cell lines, we exploited the innovative W8 technology and relative method. The combined PCA-BC method analysis successfully offered valuable insights for the researches to understand population behavior, reproducibility and correlation, leading to improve decision-making steps during daily lab routines. Performing PCA-BC analysis before administering drugs could help to verify the model’s consistency, thereby ensuring a more reliable and robust experimental setting. Additionally, when multiple operators perform experiments, PCA-BC allows displaying and clustering the most similar spheroid replicates regardless of the operator, minimizing potential biases arising from manual skills. Moreover, the developed algorithm identified a unique “biophysical fingerprint” for each population of spheroids generated from the characterized commercial cell lines. In particular, mass density to size ratio opens up to data correlations that can involve cell number, or cell density, making it a useful non-invasive metric for assessing spheroid growth and drug response, as well as uncover phenotypic heterogeneity. However, disparities between mass and cell densities very often occurs, permitting correlation with several cross-sectional related variations like extracellular matrix (ECM) content, necrotic core or formation of internal cavities, despite a similar spheroid size. These differences, imperceptible based on size alone, can significantly influence drug and radiation responses, thereby introducing variability in screening assays. In case of drug testing applications, this could significantly impact the drug’s ability to penetrate deeper layers of the spheroid and ultimately affect the observed efficacy data.

Furthermore, by combining the W8 technology ability to separate analyzed populations by structural parameters (size only, mass density only, or both) with the PCA-BC method, researchers can perform studies on structurally controlled 3DCC platforms with greater reliability, minimizing model-related bias and increasing the likelihood of uncovering subtle yet critical insights, from biological and biochemical effects to drug responses, genomics, and proteomics. The present work, performed on commercially available cell lines, aims at setting the basis and open the doors to explore this intriguing biophysical aspect in greater depth, an endeavor particularly crucial for complex patient-derived (PD) 3D cell cultures. This analysis paves the way for the development of a multiparametric algorithm that can rapidly assess the model’s reliability and, consequently, the validity of the generated data. Specifically, the PCA-BC method it brings attention to unique structural heterogeneity and the importance of “per line” and “per protocol” structural characterization assessments, as key factors to improve decision making during 3DCC workflows. Implementing this with future tailored protocol improvements for optimal viability correlations would significantly strengthen the overall accuracy and effectiveness of drug discovery, using 3D tumor models.

Methods

Experimental design

With the aim of better understanding 3DCC biophysical variability, distribution and potential clusterization, a preliminary Principal Component Analysis (PCA) was run using the mass density, weight and size values obtained from the W8 analyses. The following populations of spheroids created from 9 different cell lines, all harvested on the 7th day of culture, where compared: A549, HCT-116, LoVo, MCF7, Panc-1, SKMEL, T47D, U87 and VCAP.

The potential of such approach was immediately evident, so that an initial seeding number comparisons between 3 conditions was performed (See S1, Supplementary Material).

Although some degree of separation between spheroids from different cell lines was already apparent, a more precise rationale was required to create a robust proof of concept for the experiments. To achieve this, spheroids from five cell lines were selected based on two main criteria: (i) those representing a range of reproducibility and homogeneity—classified as good, medium, and poor—ensuring a wide distribution throughout the PCA; (ii) those of heightened biological interest to facilitate meaningful data collection. The 2 and 2 statistical analyses of the population were used as reference metrics to evaluate the separation (Table S1, Supplementary Material).

After this selection, it was noteworthy that, even when all spheroids from the five selected groups were included in the PCA based on raw data, subpopulations are almost perfectly separated by the automatic Hierarchical Cluster Analysis (HCA), with only a minor overlap between a subset of spheroids from MCF-7 and HCT-116 (Fig. S1.4, Supplementary Material). However, when selecting the five spheroids with Mass Density values closest to the mean Mass Density for each population, any overlapping in the PCA analysis is eliminated. Figure 4 shows the PCA distribution (B), as well as the clear automated distinction between the five groups by the HCA analyses (A).

Fig. 5
figure 5

Experimental rationale. (A) Dendrogram of the best 5 spheroids for each of the 5 selected cell lines. (B) PCA of the best 5 spheroids per each of the 5 selected cell lines.

Full size image

Results shown in Fig. 5 led to build the rationale guiding the design of the experiments and methods presented in this study. To allow meaningful comparisons, in-depth comparison analyses where performed on spheroids derived from the LoVo and A549 cell lines, which were chosen based on the following considerations:

  • Representative Rationale: A549 and LoVo reflect contrasting conditions as representative groups for best and worst combination of R² and Q² values (Table S1, Supplementary Material).

  • Hierarchical Clustering Analysis (HCA): They occupy distinct positions in the first grade of separation branches of the HCA (Fig. 5, Y axe values > 7).

  • Principal Component Analysis (PCA): Initial PCA results suggest differences in the mass density-to-size ratio, particularly in diagonal distributions, that warrant further investigation (herein discussed in the results).

  • Biological Significance: Both cell lines are of high biological interest to researchers, offering valuable insights into diverse cellular behaviours.

Based on these observations, this manuscript compares the two selected cell lines after defining a detailed experimental rationale and protocol workflow described in Fig. 6. The protocol focuses at establishing the most elevated standardized condition to operate, not only to compare the two cell lines, but also examine their reproducibility across multiple batches production, replicates and operators. This approach aims to distinguish intrinsic biological diversity from variability caused by external factors. Here, we propose a simple yet robust biophysical method to assess and understand these variations, which can be integrated into daily workflows. The method is schematized as follows:

Fig. 6
figure 6

Experimental design. For the two cell lines, three batches were prepared by three operators in triplicates. Each replicate was seeded in 10 wells of a 384-well plate and incubated for seven days. The harvested samples were then analyzed using the W8 instrument, and the obtained biophysical outputs were elaborated by PCA and HCA methods.

Full size image

W8 analysis

The W8 system

After connecting the centrifuge tube containing the spheroids population to the W8 System, the operator can launch the measurements, which are performed automatically.

The W8 System, equipped with the Libra software, start the measurements by withdrawing individual spheroids sequentially. This is achieved through a combination of spheroid suspension and precise activation of peristaltic pumps, ensuring that spheroids are transported into the flow-chip and visualized by the Lower Magnification (LM) camera.

The selected object is tracked by the optics and an initial size evaluation is performed in order to activate a customized flow-rate for sample handling and minimized shear stress.

The sample is then centered via the described SCCP, in which active flow and sedimentation timing alternate across the coaxial restriction of the channel’s architecture. This phase ensures the spheroid’s location matches the focal plane of the optics leading to a more precise size measurement. The spheroid is then moved in front of the HM camera, which captures 1 frame every 100 ms, and 16 Feret’s diameters, at different angles, are automatically measured for every frame. This enables averaging about 800 measurements of diameter per spheroid. Most importantly, contrarily from a specimen-based brightfield imaging technique, samples within the W8 system are in free-floating condition, and thus free rotations are allowed. This is a crucial aspect that permits often recording images of the same spheroids from different angles, strongly increasing the reliability of the size value for 3D objects, as well as their size standard deviation (grey steady phase in Fig. 7).

Subsequently, the flow is stopped to permit the spheroid to fall, enabling the optical systems to track its terminal velocity under precisely monitored environmental conditions (density, viscosity, and temperature). As a final step, the already validated gravimetric theory, readapted for the specific channel geometry, is used to derive the values of mass density and weight for each object analyzed (green falling phase in Fig. 7). In order to further improve the measurements, and minimize the SD of the results, the falling motion is repeated twice for each sample.

Fig. 7
figure 7

The W8 system. Schematic representation of the main measuring phases and improvements.

Full size image

Biophysical characterization

Spheroids’ populations were resuspended in the W8 Analysis Solution (W8AS) and diluted up to 7 mL into a centrifuge tube. The biophysical characterization of the samples was performed using the W8 System (CellDynamics), which allows the measure of Mass Density, Size, Weight, and relative standard deviation values, of each spheroid, following the here described (Section S2 and Scheme S1, Supplementary Material) improved method previously validated procedure by Cristaldi et al.6. The initial measurements were carried out on 270 spheroids, produced from 9 cell lines, and divided in 3 seeding conditions per cell line. This preliminary study, presented in S1 (Fig. S1.3, Supplementary Material), was the crucial starting point to identify the final two cell lines to be used to grow spheroids for the present work. In order to normalize experimental conditions, the following detailed experimental designs was adopted. Spheroids were produced from A549 and LoVo cell lines by 3 operators working in parallel. Specifically, for each cell line, every operator created three spheroid batches, of 5 spheroids per batch (replicates). A total of 90 spheroids, 45 per cell line, were produced under the normalized conditions described in Methods.

Cell culture and tumor spheroids generation

For the generation of tumor spheroids, adherent cells were detached with Trypsin/EDTA (Corning, New York, NY, USA) and counted using TC-20 counter (Bio-Rad Laboratorie S.r.l. MI, Italy). Then, cell suspension was seeded in a PrimeSurface® 3D culture multiwell: Ultra-low Attachment Plates, 384 well, U bottom, Clear plates (S-bio, Hudson, NH, USA). 500 cells per well were seeded with a total volume of 80 µl/well of cell suspension. Culture media were: EMEM (ATCC, Manassas VA, USA) for SKMEL and U87; EMEM + 0.1% insulin (Merck Chemicals, DA, Germany) for MCF-7; DMEM High Glucose (Euroclone S.p.a. MI, Italy) for VcaP and PANC-1; DMEM/F12 (ATCC, Manassas VA, USA) and DMEM low Glucose (Euroclone S.p.a. MI, Italy) for HCT-116; F12K (ATCC, Manassas VA, USA) for A549 and LoVo; RPMI-1640 (Euroclone S.p.a. MI, Italy) + + 1% L-glutamine (Gibco, Thermo Fisher Scientific, USA) + 0,1% insulin for T47D. All culture media were supplemented with 10% FBS (Euroclone S.p.a. MI, Italy). All the spheroids except for U87 were cultured at 37 °C in an atmosphere of 5% CO2 under normoxia. U87 spheroids were cultured in a hypoxic (1% O2) atmosphere. Experiments were performed with W8 Physical Cytometer (CellDynamics ISRL, BO, Italy) at proper spheroids compaction, after 1 week of culture. Once selected the final two cell lines used for the present work (A549 and LoVo) new series of experiments were carried out. A549 and LoVo were seeded in the same condition described above in three different replicates by three different operators working in parallel. Each operator used a different cell batch deriving from the same original batch.

Statistical analysis

To analyze the distribution of the dataset based on skewness and/or kurtosis, first the Shapiro–Wilk test was performed on the obtained mass density, weight and diameter outputs6. For all the cases of non-normal distribution, the Tukey method was then carried out for identifying and eliminating outliers (K > 1.5). Subsequently, the Shapiro–Wilk approach was reiterated to confirm the normal distribution. Data are presented as mean ± SD. Statistical analysis was performed using two-tailed unpaired Student’s t-test.

Principal components analysis and hierarchical cluster analysis

The biophysical characterization performed with the W8 Physical Cytometer allowed obtaining the output dataset needed to run the Principal Components Analysis (PCA) and the Hierarchical Cluster Analysis (HCA). The PCA and the HCA were performed using the SIMCA software package by Umetrics (SIMCA version 14.1) maintaining the standard settings. The PCA results can be visualized by Objects Plot and/or by Loadings Plot23. The Objects Plot permits the comparison between each spheroid within the population, based on the data obtained by its biophysical characterization. The Loadings Plot permits to visualize the trend and correlation of the biophysical metrics of Mass Density, Size and Weight, used as variables. Differently, the HCA, visualized by dendrogram, was performed by dividing the population into a number of groups based on the purpose of the investigation, i.e. 2 groups when aiming at clustering spheroids from 2 cell lines or 3 groups when aiming at separating 3 operators or 3 repetitions.