IntroductionThe CRISPR/Cas9 system is a widely used gene-editing technology with significant potential for gene therapies. In 2023, health authorities approved the first CRISPR/Cas9-based gene therapy, exagamglogene autotemcel [1], and numerous other CRISPR/Cas9 therapies are currently in development [2]. As clinical applications continue to grow, establishing appropriate safety assessment strategies for CRISPR/Cas9-based therapies has become increasingly critical. A primary safety risk is off-target editing [3], which refers to inadvertent genetic modifications that could lead to unfavorable phenotypical changes such as cell transformation or diminished cell fitness. Several health authorities have issued guidelines addressing this matter [4,5,6], proposing a staggered approach that includes in silico and in vitro detection of off-target modifications, along with an assessment of the biological consequences of off-target editing, such as the potential risk of tumorigenicity. Given the novelty of this technology, standard methodologies have not yet been established, highlighting the need for the development of innovative and exploratory approaches, particularly for tumorigenicity assessment.Tumorigenicity is defined as “the ability of a cell population transplanted into an animal to give rise to malignant or benign tumors by proliferation” [7]. Standard tumorigenicity assessment involves in vivo tumorigenicity studies, which entail the injection of human, genetically modified cells into immunocompromised mice [8], and monitoring tumor formation for six to twelve months, depending on the cell type [9,10,11]. However, the rate of successful human tumor engraftment greatly varies depending on the injected cell type [12,13,14,15], potentially due to the fact that the complex human tumor microenvironment is not sufficiently mimicked in the murine model [8]. Thus, in vitro transformation assays could be relevant alternatives or complementary testing methods that are in line with the 3Rs principle (Replacement, Reduction and Refinement of animal experiments). In this context, the Soft Agar Colony Forming Assay (SACF) and the Growth in Low Attachment Assay (GILA) are two in vitro platforms that could be used as characterization tools during development to support tumorigenicity risk evaluation [16,17,18]. These assays measure the ability to grow anchorage-independently, which is a characteristic of transformed adherent cells [16,17,18]. In SACF, anchorage-independence is assessed by culturing the edited cells in semi-solid agarose, which interferes with cell attachment. After four weeks in culture, emerging colonies are counted as a measurement of the transformation potential. In GILA assay, the ability to grow anchorage independently is evaluated through culturing the cells in ultra-low attachment plates for two weeks, after which the surviving cells are indirectly quantified by measuring ATP.The suitability of both assays for the in vitro transformation assessment of CRISPR/Cas9-edited adherent cells has been previously demonstrated at a single site, including the assessment of suitable cell lines, the establishment of adequate positive and negative controls and the characterization of the limit of detection [18]. However, evaluation across multiple laboratories is still lacking and is critical to support broader use in cell therapy development. Both International Council for Harmonization (ICH) Q2(R2) and the Food and Drug Administration (FDA) guidance on “Analytical Procedures and Methods Validation for Drugs and Biologics” explicitly state that inter- and intra-laboratory variability is a foundational component of assay performance characterization [1920]. This study aimed to address this need by conducting a coordinated multi-site evaluation of SACF and GILA. Organized by the Cell and Gene Therapy - TRAcking, Circulation, & Safety (CGT-TRACS) committee of the Health and Environmental Sciences Institute (HESI Global), experiments were performed at four independent laboratories at AstraZeneca, Bristol-Myers Squibb (BMS), Novartis, and Sanofi. At each site, MCF10A cells with a CRISPR/Cas9-mediated knockout of the known tumor suppressor gene PTPN12 were generated to serve as positive control. These PTPN12-edited cells were subsequently mixed with unedited cells at defined ratios to mimic varying frequencies of transformed cells and tested in SACF and GILA. Both assays were conducted under standardized protocols to evaluate the limits of detection at each site, and Spearman’s rank correlation analysis was used to assess both intra- and inter-laboratory variability.Materials and methodsCulture of MCF10A cell lineMCF10A cells (#CRL-10317, ATCC, United States) were cultured in DMEM/F12 medium (#11330, Gibco, Thermo Fisher ScientificFootnote 1), enriched with 5% (v/v) horse serum (#16050122, Gibco), 20 ng/mL EGF (#PHG0311, Thermo Fisher Scientific), 250 ng/mL hydrocortisone (#H-0888, Sigma), 100 ng/mL cholera toxin (#C8052, Sigma), and 10 μg/mL human insulin (#I9278, Sigma), and incubated at 37 °C in a 5% CO2 atmosphere. Several MCF10A lots were evaluated for this study as a difference in colony formation efficiency was observed between lots (Table 1). The cells were cultured for at least four weeks after thawing before starting the experiments to ensure assay consistency (Supplementary Fig. S2A).Electroporation of MCF10A cells with the ribonucleoprotein (RNP) complexA 44 μM cr:trRNA working solution was prepared by combining 100 μM stock solutions of crRNA (custom, sequence in Table 2) and trRNA (#1072534) with nuclease-free duplex buffer (all Integrated DNA Technologies (IDT)) and heating at 95 °C for 5 min. Alt-R Streptococcus pyogenes Cas9 (#1081058, IDT) was diluted to 36 µM with buffer R of the Neon transfection system (#MPK1096 (kit), MPK5000 (transfection system), Thermo Fisher Scientific) and mixed with the cr:trRNA working solution to a final ribonucleoprotein complex (RNP) concentration of 18 µM (contains 18 µM Cas9 and 22 µM of the cr:trRNA duplex).Table 1 Different MCF10A lots evaluated for this study.Full size tableTable 2 CRISPR/Cas9 target sequence used in this study.Full size tableA total of 600,000 MCF10A cells in logarithmic growth phase were resuspended in buffer R. Electroporation was performed in batches of 150,000 cells, with each batch mixed with 9.5 pmol of the RNP complex. The electroporation was carried out with the Neon transfection system using 10-μL tips and applying a single pulse of 1700 V and a pulse width of 20 ms. Electroporated cells were pooled into 6 mL of MCF10A culture medium and evenly distributed across two wells of a 6-well plate. The cells were cultured for 48 h prior to proceeding with SACF, GILA, and cleavage analyses.Spiking of PTPN12-edited with untreated MCF10A cells48 h post electroporation, PTPN12-edited and untreated MCF10A cells were detached and resuspended in culture medium to 100,000 cells/mL, preparing 4 mL in total for edited cells and 10 mL for untreated cells. The cells were then mixed in different ratios to achieve final mixtures containing 100%, 75%, 50%, 25%, 12.5%, 6.3%, 3.1%, 1.6%, 0.8%, and 0% PTPN12-edited cells in a total volume of 1 mL. From each mixture, 200 μL was used for SACF and 250 μL for GILA.SACFSACF was conducted as described previously [18]. Briefly, a 1.2% agarose solution was prepared by dissolving SeaPlaque agarose (#50101, Lonza) in Ultra-pure DNase/RNase-free distilled water, and mixed in a 1:1 ratio with 2× MCF10A culture media containing 2× DMEM/F12 medium (diluted from 10× DMEM/F12, #CAM17-005, GenDEPOT, Katy, TX, USA), 20% horse serum (v/v), 40 ng/mL EGF, 200 ng/mL cholera toxin, 20 μg/mL human insulin, and 500 ng/mL hydrocortisone. 50 μL of this solution was added to each well of a black 96-well flat and clear-bottom plate with a cell-repellent surface (655976-SIN, Greiner Bio-one), excluding the well on the edges, and incubated at 4 °C for 30 min. For the cell layer, a cell suspension of 100 000 cells/mL was combined with 2× MCF10A culture medium, and 1.2% agarose solution in a 1:1:1 ratio. 75 μL of this suspension was added on top of the agar base layer (resulting 2500 cells per well) and solidified at 4 °C for 20 min. Subsequently, 100 μL of MCF10A culture medium was added to each well as top layer. Plates were incubated for four weeks at 37 °C in a 5% CO2 atmosphere, with MCF10A culture medium changes twice per week.For imaging, the top layer medium was carefully removed without disrupting the cell suspension layer, and 50 μL of culture medium containing 25 nM MitoTracker Red CMXRos (#M7512, Thermo Fisher Scientific) and 1 μg/mL Hoechst 33342 (#62249, Thermo Fisher Scientific) was added to each well. The plate was incubated for 1 h at 37 °C and, subsequently, 125 μL of 4.8% Paraformaldehyde solution (#28908, Thermo Fisher Scientific) in PBS were added to each well resulting in a final concentration of 2%. After a 30-minute incubation at room temperature, the wells were washed twice with 100 μL of PBS. Finally, 75 μL of buffer QG (#19063, QIAGEN) was added to each well to solubilize the agar at 37 °C for 1 h. QG buffer incubation was limited to 3 h to prevent the solution from becoming cloudy.SACF image acquisitionImaging was performed with the ImageXpress Micro Confocal instrument (Molecular Devices) with a 2× or 5× magnification objective in widefield imaging mode, taking one or four images covering the whole surface of the wells depending on the objective. Analysis was conducted using the MetaXpress cell analysis software (Molecular Devices), with colonies identified based on Hoechst 33342 signal and contaminants excluded using MitoTracker Red CMXRos. First, Hoechst images were slightly blurred to increase the signal homogeneity of the objects and prevent inaccurate object segmentation using the “Open/Close” feature with a circle size value of 5 pixels. The resulting image was then subjected to the “Find Round Objects” algorithm with minimum and maximum width values of 40 and 900 micrometers, respectively, and an intensity above 1500 relative fluorescence units. As the obtained mask recovered not only the colonies of interest but also some artefacts of different shapes and signal intensities (e.g. cell debris and fibers), additional filters were applied. A maximum Ellipse Form Factor of 1.4, a minimum Shape Factor of 0.75 and a maximum length of 1000 µm were used to remove elongated shape artefacts not in accordance with the expected roundness of colonies. To exclude artefacts of small size and showing Hoechst intensities out of the expected range for colonies, a minimum object area of 3600 µm2 and an average Hoechst intensity between 1000 and 15,000 relative fluorescence units were applied as additional filters. To ensure that the colonies consisted of living cells (i.e. cells with functional mitochondria), objects below a MitoTracker average intensity of 1800 relative fluorescence units were excluded. The whole colony population was divided into 3 subpopulations based on their size: small (below 25,000 µm2), medium (between 25,001 and 125,000 µm2) and large (above 125,001 µm2). Of note, all fluorescence intensity thresholds or filters may vary depending on the imaging device and the selected acquisition settings, for example objective, light source intensity and exposure time. Therefore, those parameters were checked and adapted within each lab.For Supplementary Fig. S2C, plates were additionally imaged using the Scientific CellInsight CX7 High Content Screening (HCS) Platform (Thermo Fisher Scientific) with 10× magnification in widefield imaging mode. A well with the highest expected fluorescence intensity was selected, and images were acquired in channel 1 (Ex/Em 386/440) for the Hoechst signal and channel 2 (Ex/Em 549/600) for the MitoTracker signal, respectively, using laser autofocus. The pixel intensity histogram was reviewed to confirm the signal was within the camera’s dynamic range; if out of range, the exposure time was adjusted, and imaging was repeated. Once the exposure time was confirmed, 25 images were captured per well, covering the entire well. Image analysis was performed using Thermo Scientific HCS Studio Cell Analysis Software. Background correction over 255 pixels was applied, followed by signal smoothing and object selection based on size (approximately >1300 pixels²). Artifacts were excluded based on criteria such as shape, average intensity, and signal variation. SACF raw data is available in Supplementary Table 1.GILAGILA was performed as described previously [18]. Briefly, 2500 cells in 100 μL culture medium were seeded in U-bottom, ultra-low attachment plates (#7007, Corning Life Sciences) and incubated for two weeks at 37 °C with 5% CO2. Subsequently, ATP levels were assessed using the ViaLight Plus cell proliferation and cytotoxicity bioassay kit (#LT07-221, Lonza) according to the manufacturer’s instructions. The following plate readers were used across different participating sites: CLARIOstar (BMG Labtech, site A), SpectraMAx i3x (Molecular Devices, site B), EnVision 2104 multilabel reader (Revity, site C), and the Infinite M1000 (Tecan, site D). Wells containing contaminant objects such as plastic fibers were excluded from further analysis as these objects could alter the results. GILA raw data is available in Supplementary Table 1.Statistical analysesP-values were determined using mixed effects models for negative binomial distribution (SACF) or Gaussian distribution (GILA) with post-hoc Holm–Bonferroni adjustment as described previously [18]. These models included the concentration as a fixed effect and the experiment repetition as a random effect. The 0% samples served as the reference condition. Script for the analysis used in this study is available in Supplementary File 1.To evaluate the variability of the SACF and GILA within and between laboratories, a Spearman’s rank correlation coefficient analysis was performed. This method assesses the relative agreement of the mean values for each treatment between sites or between assay repetitions within sites. This non-parametric method determines the strength of the monotonic relationship between different measurements without considering the agreement in absolute values. This is relevant as the GILA assay produces arbitrary values that could differ between measurements and instruments. Moreover, for both GILA and SACF, it is of importance to detect a positive signal as compared to the negative control rather than a certain absolute value. The technical replicates were averaged over the assay repetition for the within-site analysis and overall replicates for the between-site analysis. Treatment concentration was used as the unit of analysis, and Spearman’s rank correlation coefficient was determined to assess the correlations. To account for multiple testing, p-values were adjusted using the Holm method (Supplementary File 2).Determination of mutation/cleavage efficiency via Sanger sequencing and TIDEDNA was isolated from the cells remaining after SACF and GILA preparation, which included ~120,000 edited cells and 270,000 untreated cells. Isolation was performed using the PureLink Genomic DNA Kit (#K1820, Thermo Fisher Scientific) according to the manufacturer’s protocol, with DNA eluted in 30 μL of elution buffer.PCR amplification of the target region was performed using Q5 High-Fidelity 2× Master Mix (#M0492S, New England Biolabs), 1 μM primers (Microsynth, see Table 3 for sequence), and 10–100 ng total DNA in a 25-μL reaction. PCR conditions included 30 s at 98 °C, 35 cycles of 5 s at 98 °C, 30 s at the primer annealing temperature of 65 °C, 20 s at 72 °C, and a final extension for 2 min at 72 °C. The PCR product was purified using the GeneJET PCR purification kit (#K0702, Thermo Fisher Scientific) according to the manufacturer’s instructions.Table 3 Primers used in this study.Full size tableFor Sanger sequencing, purified DNA was diluted to 18 ng/100 bp in 12 μL elution buffer and mixed with 3 μL of 100 μM forward primer. Sanger sequencing was carried out at Azenta Genewiz (site A), Psomagen (site B), Microsynth (site C and D). Chromatograms were analyzed for mutation frequency using Tracking Indels by Decomposition (TIDE) software [21] using the default parameters and the largest possible indel size range to calculate the on-target editing efficiency.ResultsExperimental design of the HESI Global’s CGT-TRACS multi-site study to evaluate SACF and GILA for CRISPR/Cas9-edited cellsThe aim of this HESI multi-site study was to evaluate the robustness and reproducibility of SACF and GILA for the in vitro transformation assessment of CRISPR/Cas9-edited cells. The study was carried out at four independent sites, which are referred to as sites A, B, C, and D. Building on a previous study by Lemmens et al., which characterized these in vitro transformation assays by identifying suitable cell lines, establishing positive and negative controls, and determining assays’ LOD, this study focused on replicating these findings and on the assessment of inter- and intra-laboratory variability.As in the prior study, MCF10A, a well-characterized human non-tumorigenic, spontaneously immortalized breast epithelial cell line, was selected due to its suitability for CRISPR/Cas9 editing and in vitro transformation assessments. However, instead of utilizing a stable Cas9-expressing MCF10A line as previously, wild-type (WT) MCF10A cells were chosen for this study due to their broader availability. Of note, variability in the transformation efficiency was observed across different MCF10A cell lots. To address this, a cell lot with high transformation efficiency was selected and used across all participating sites to ensure consistency (see Table 1 for an overview of the tested cell lots).The original study defined multiple positive controls for SACF and GILA including ERRFI1, NF2, PTEN, and PTPN12 [18]. Among these, PTPN12 was selected as control in this study because it is the most extensively validated and consistently provided the strongest and most reliable induction of anchorage‑independent growth in MCF10A cells across both assays.Also, in line with the previous study, the LOD at each site was evaluated by spiking PTPN12-edited MCF10A cells with WT cells at defined ratios to mimic different frequencies of potentially transformed cells. In detail, the experimental workflow included electroporation of MCF10A cells with a ribonucleoprotein (RNP) complex composed of Cas9, tracrRNA, and crRNA targeting the PTPN12 gene. After a 48-hour recovery period, PTPN12-edited and unedited wild type cells were mixed at various ratios, resulting in samples containing theoretically 100%, 75%, 50%, 25%, 12.5%, 6.3%, 3.1%, 1.6%, 0.8%, and 0% edited cells. To verify PTPN12 editing efficiency, an aliquot of both edited- and unedited cells was kept aside for DNA extraction and cleavage efficiency analysis by Sanger sequencing. The different spike-in samples were then assessed in SACF and GILA. The SACF involved embedding single-cell suspensions in soft agar, followed by a four-week incubation. While non-transformed cells did not survive in soft agar, transformed PTPN12-edited cells formed colonies that were stained with Hoechst dye and MitoTracker, and processed for high-content imaging and quantification by fixation and agar solubilization (Fig. 1A, Supplementary Fig. S1A–D). For the GILA, cells were cultured in ultra-low attachment plates for two weeks, after which cell growth and spheroid formation were assessed via ATP measurement (Fig. 1A, Supplementary Fig. S1E, F). Each assay was repeated three to four times at each participating site, with six technical replicates used per sample for every experiment. Of note, before conducting the experiments, MCF10A cells were kept in culture for at least four weeks after thawing because shorter culturing times resulted in more variable and reduced transformation efficiency (Supplementary Fig. S2A).Fig. 1: Overview of the HESI Global multi-site study to evaluate SACF and GILA for CRISPR/Cas9 edited cells.Full size imageA Schematic overview of the HESI Global multi-site study. Created in BioRender. Lemmens, M. (2025) https://BioRender.com/hfql8q3. On day 0, MCF10A cells were electroporated with an RNP consisting of Cas9, a trRNA and a crRNA targeting the PTPN12 gene. After 48 h, the cells were collected, and an aliquot was retained to evaluate the mutation frequency. The remainder of the cells were utilized for spiking unedited cells with cells edited for PTPN12, at varying ratios. The prepared samples were seeded in SACF and GILA assays. High-content imaging was utilized to count the colonies in the SACF, four weeks post-seeding. ATP levels in GILA were measured at the end of the second week post-seeding. The experiment was replicated 3–4 times at every participating site, with a total of four sites involved in the study. B Mutation frequency 48 h after electroporation. Sanger sequencing was performed on DNA from PTPN12-edited MCF10A cells and an unedited control. The Tracking of Indels by Decomposition (TIDE) software was used to quantify the mutations. Mutation frequency was performed once on each experiment repetition (n = 1). RNP Ribonucleoprotein, SACF Soft Agar Colony Forming Assay, GILA Growth in Low Attachment Assay, PTPN12 Protein Tyrosine Phosphatase Non-receptor type 12, ATP Adenosine triphosphate.PTPN12 editing efficiency was assessed in the allegedly 100% edited sample to evaluate the CRISPR editing efficiency across sites, as variability in editing frequency could alter the final spike-in concentrations and impact comparability between sites and experiments. Editing efficacy was quantified using Sanger sequencing analyzed with the TIDE (Tracking of Indels by Decomposition) tool. TIDE provides a rapid measurement of insertion and deletion (InDel) rates, although it is limited to detecting shorter InDels (100 base pairs) occur in CRISPR/Cas9-edited cells and their frequency correlates positively with smaller InDels [23]. Thus, TIDE measurements may not fully capture the editing complexity introduced by larger deletions but nevertheless provide a useful approximation of editing efficiency. If substantial differences in cleavage efficiencies were detected between experiments and sites, final spike-in concentrations would need to be adjusted accordingly to ensure comparability.Mutation frequency measurements from the participating sites demonstrated high consistency. Across three to four experimental repetitions per site, cleavage efficiencies ranged from 79.2% (site B) to 97.1% (site D), with the majority of experiments (10 out of 13) achieving cleavage efficiencies exceeding 90% (Fig. 1B). Only three experiments exhibited editing efficiencies below 85%. These results highlight that the electroporation protocol employed in this study is highly reliable for achieving robust editing efficiencies in MCF10A cells. Furthermore, due to consistent editing efficiencies observed across experiments and sites, spike-in concentrations in SACF and GILA were not adjusted.Evaluation of variability between and within laboratories for the GILA assayThe PTPN12 KO cells were then used in SACF as well as GILA assay at different spike-in concentrations to assess LOD and intra- and inter-laboratory variability. In the previous study by Lemmens et al, GILA cultivation times from two to four weeks were evaluated, and two weeks was determined as optimal incubation time. To assess if the culture time could be shortened, a one-week incubation was tested at one site (Supplementary Fig. S2B). Two weeks incubation led to higher ATP levels for the concentration 0.8–25%, while there was no difference in ATP levels for the concentration 50% and higher. Thus, for this multi-side study, two weeks incubation time was used for GILA. For both assays, the LOD was determined at each site by identifying the lowest concentration that elicited statistically significant results over 0% controls (p