ELDA: real-time functional drug profiling in acute lymphoblastic leukemia

Abstract

Introduction:

Pediatric acute lymphoblastic leukemia (ALL) has cure rates exceeding 80%, yet approximately 20% of patients experience relapse, for whom treatment remains challenging. The biological and clinical heterogeneity of ALL complicates therapeutic decision-making, especially in relapse. A functional, patient-specific tool to identify effective drugs could complement current stratification approaches and potentially optimize therapy selection.

Methods:

We developed ELDA (Ex vivo Leukemia Drug Advisor), a novel drug screening platform based on time-lapse bright-field microscopy. Primary ALL cells were co-cultured with human bone marrow stromal cells in a three-dimensional collagen matrix, replicating a physiologically relevant microenvironment. Candidate drugs were added, and wells were imaged every 30 min for 96 h. A custom image analysis algorithm quantified viable cells by detecting subtle cell membrane movements, enabling real-time monitoring of treatment response.

Results and discussion:

ELDA’s results demonstrated strong concordance with established cytotoxicity assays, including MTT and Calcein-AM, validating its performance in vitro. Drug profiling of 25 pediatric ALL samples against a panel of 50 clinically relevant and experimental agents revealed distinct sensitivity and resistance patterns for each patient. Notably, heterogeneity was observed even among cases sharing the same genetic alterations, underscoring the limitations of genotype-based therapeutic predictions alone. Drugs with similar mechanisms of action exhibited correlated activity profiles, confirming the biological consistency of the assay. Importantly, ELDA-predicted differences in sensitivity to dexamethasone and vincristine were validated in vivo, further supporting its translational relevance. ELDA is a standardized and validated ex vivo drug screening methodology capable of generating individualized drug response profiles for pediatric ALL. Beyond its investigational potential in precision oncology, ELDA offers a platform for preclinical evaluation of novel agents and drug combinations.

1 Introduction

Acute lymphoblastic leukemia (ALL) is the most common malignancy in children and adolescents, accounting for 25% of pediatric cancers (1, 2). ALL originates from T-cell (T-ALL) or B-cell (B-ALL) precursors, comprising approximately 15% or 85% of cases, respectively. The presence of genetic abnormalities such as aneuploidies, chromosomal translocations, and certain mutations is instrumental for the classification of ALL into several “molecular subtypes,” which may be associated with different clinical responses (3, 4).

Treatment of pediatric ALL is based on chemotherapy with corticosteroids (prednisolone, dexamethasone), antimetabolites (methotrexate, L-asparginase, 6-mercaptopurine, 6-thioguanine, cytarabine), anthracyclines (daunorubicin, doxorubicin), vincristine, and cyclophosphamide (5, 6). Despite the high cure rate for childhood ALL, exceeding 80% in developed countries, relapse, which affects 20% of patients, remains the leading cause of cancer-related death in children and adolescents worldwide (7, 8), and its treatment remains a challenge (9–12).

Intrinsic resistance of leukemic cells to the chemotherapy drugs used in the treatment is the main cause of relapse. This concept is confirmed by the strong association between the delay in the elimination of leukemic cells during the first months of treatment and the occurrence of relapses. It is also confirmed by in vitro experiments, in which ALL cells harvested at diagnosis are evaluated for resistance to increasing doses of drugs used in the clinic: dexamethasone, prednisolone, L-asparaginase, and vincristine. Patients whose leukemic cells are more resistant to these drugs are more likely to experience disease relapse (13–17) even when considering only patients from a defined subgroup, such as those at high risk (18) and those with the t(9;22)(BCR/ABL1) (19).

In certain studies, the in vitro resistance profile to prednisolone, vincristine, and L-asparaginase (PVA) was an independent prognostic factor for predicting relapses (20–22), although not for late relapses occurring after 2.5 years from diagnosis (21). However, this prognostic strategy was abandoned in favor of minimal residual disease (MRD) assessment, which performs a similar evaluation conducted in vivo (23, 24). Nonetheless, in vitro assays retain value in identifying potential therapeutic agents for relapsed or refractory ALL. Indeed, in vitro drug screening assays have played a fundamental role in the discovery of novel drugs for specific subtypes of ALL. For instance, TCF3-HLF-positive ALL has been shown to be sensitive to venetoclax (25, 26), while pre-B-ALL exhibits sensitivity to dasatinib (27) and T-ALL to dasatinib and temsirolimus (28, 29).

Methods for in vitro drug screening in ALL are mostly based on regular liquid cell culture of ALL cells, with or without co-culture on a bone marrow stromal cell (BMSC) layer. Cell viability is measured at pre-determined time points, usually after 72 or 96 h of drug exposure. In monoculture assays, ALL cell viability can be evaluated using the MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) reagent, which measures mitochondrial activity and is indicative of cell viability (29). In co-culture assays, viable ALL cells can be labeled with Calcein-AM and differentiated from MSC by image analysis (25) or flow cytometry (30, 31). These are endpoint methods, usually run with 100,000 or 20,000 ALL cells per well in 96- or 384-well plates, respectively.

Here, we present a new method for drug screening in primary ALL cells, based on the co-culture of ALL cells and stromal cells into a collagen matrix. Viability of ALL cells, and discrimination from stromal cells, is obtained by analyzing refringence and membrane movements in time-lapse microscopy. Results obtained by this method, named ELDA for Ex vivo Leukemia Drug Advisor, were comparable to those obtained with MTT and Calcein-AM, but it uses only 4,000 ALL cells per well (384-plate), needs no cellular dye, and measures cellular viability in real time.

2 Materials and methods

2.1 Patient samples

Primary human ALL cells were obtained from cryopreserved bone marrow aspirates of pediatric patients treated at Centro Infantil Boldrini. Informed consent was obtained from the patients and their guardians. The study was approved by the Institutional Research Ethics Committee under CAAE 34601120.7.0000.5376. ALL cells were expanded in NSG-immunodeficient mice, and the patient-derived xenografts (PDX), at passage 1, were cryopreserved in liquid nitrogen. Frozen PDX samples were thawed and used for the experiments. The use of animals for PDX generation and in vivo experiments (see below) was approved by the Boldrini Animal Care and Use Committee under protocol number 0011/2020.

2.2 TERT-immortalized mesenchymal stem cells

TERT-immortalized mesenchymal stem cells (T-MSCs) were obtained from Dr. Dario Campana (32). T-MSCs were cultured in RPMI-1640 medium (Cultilab, Campinas, Brazil) with 10% fetal bovine serum (FBS). Semi-confluent T-MSC cultures were trypsinized using trypsin/EDTA (Sigma-Aldrich, St. Louis, MO), and cells were washed twice with FBS. T-MSCs were used for the experiment from the third to the tenth passage.

2.3 ELDA assay

One and a half million ALL cells (4,000 cells/well) and 40,000 T-MSCs (100 cells/well) in AIM-V culture medium (Invitrogen, Waltham, MA) supplemented with 1.4 mg/mL of type 1 collagen (PureCol^® - Advanced Biomatrix, San Diego, CA) were seeded in a 384-well plate (9 µL/well). The plate was spun down and kept 1 h at 37 °C in a CO₂ incubator for gelification, then 73 µL of AIM-V with 10% FBS was added to each well, and the plate was returned to the incubator overnight. Subsequently, using an EpMotion 5075 pipetting robot (Eppendorf, Hamburg, Germany), drugs were added at five different concentrations with a dilution factor of 1:5, in duplicate (technical replicates). The experimental setup included negative controls with vehicle (DMSO) and positive controls for each drug, represented by the RS4;11 cell line at the highest drug concentration. RS4;11 cells were seeded at 1,300 cells per well, together with 34 T-MSCs per well.

The plate was placed in an EVOS M7000 Cell Imaging Microscope (Thermo Fisher Scientific, Waltham, MA) to capture images of each well every 30 min over a 96-h period. Images were acquired in bright-field and phase-contrast modes using a 5× objective (Olympus, Tokyo, Japan). Each image has a resolution of 2,048 × 1,536. Throughout the imaging process, an attached incubator maintained the cells at a constant temperature of 37°C, with a CO₂ concentration of 5% and humidity levels exceeding 80%. This controlled environment ensured optimal conditions for cell viability and growth throughout the duration of the experiment.

For digital image analysis, we used the EMMA software created by one of the authors, as described (33–35). In essence, this software distinguishes between living and dead cells by analyzing the membrane movement of ALL cells across consecutive images. EMMA is based on ImageJ and incorporates the TurboReg plugin (36) for alignment of bright-field image stacks acquired at 30-min intervals for the individual wells, correcting translational motion such as plate drift and vibration. The ImageJ background subtraction function (37) was then applied, rendering cells as bright objects against a uniform dark background. The RunningZProjector plugin (gvondassow.com/Research_Site/Methods_files/Running_ZProjector2.java) was used to detect motion and small variations in cell membranes. It detects the maximum pixel intensity across a six-image (3-h) sequence. The analyzed image was then subtracted from the maximum pixel intensity projection, resulting in an image where actively moving membranes appeared as bright rings. ImageJ’s Gaussian blur filter was used to convert these bright rings into spots that cover the entire cell and to produce the overlaid images (red color) shown in this article. Although individual cells were detected, viability was quantified using the sum of their pixel area rather than cell number. Given that the signal is always normalized by the initial time point of the same well, the variation in the initial number of cells between wells, as long as it is not too large (e.g., too many cells in a well would lead to over confluency, etc.), would not affect the readings of the experiment. By applying this algorithm to all images acquired for each well, it becomes feasible to quantify the effects of drug concentration and exposure time in a non-destructive manner, without the need to separate T-MSCs from ALL cells. At the conclusion of the 96-h period, a total of 192 images was captured per well, amounting to 73,728 images per 384-well plate. For more information about the analysis scripts, refer to Khin et al. (33) and Silva et al. (34) The analysis was executed on a system equipped with an Intel Xeon Gold Processor 5118 CPU @ 2.30 GHz, with 256 GB RAM, requiring 13 h per plate (31 drugs).

Cell viability values underwent two normalization steps: 1) The pixel value corresponding to living cells at a given time point for every individual well is divided by the pixel value of the same well at the beginning of the experiment, which is considered 100% viability. The software uses a sequence of six images to assign whether a cell is alive or dead; thus, the first viability value (in pixels) corresponds to time point 6 (3 h after imaging initiation). This first normalization accounts for variations in the number of cells actually photographed in each well (while the same number of cells/well are seeded, final distribution of cells into the well varies quite considerably); 2) the viability at a given time point for each experimental condition is divided by the mean viability of the negative control (vehicle only) replicates at the same time point. The second normalization accounts for drug-independent cell death or cell proliferation.

Primary ALL cells have a very short viability in culture (31, 38, 39). Quality control consisted of three steps: 1) visual inspection of time-lapse videos, 2) analysis of the control viability curve, and 3) analysis of drug-response curves. Experiments in which control ALL cells initiated time-lapse acquisition, largely non-viable (i.e., static), were considered invalid. The viability curve of the control wells may appear acceptable; however, this can result from highly motile stromal cells displacing nearby dead ALL cells, thereby confounding normalized viability estimates. Visual inspection of time-lapse recordings was therefore mandatory for quality control. When control viability was >60% at 60 h, the assay was considered valid. When viability was <50%, the assay was discarded, as it was deemed poorly representative of the initially seeded cell population. When control viability ranged between 50% and 60%, drug-response viability curves were evaluated, and the assay was considered valid provided that an appropriate dose–response pattern was observed (Supplementary Figure 1).

2.4 MTT assay

A total of 100,000 PDX ALL/well was cultured with RPMI-1640 in 96-well plates. Cell viability was analyzed using the MTT reduction assay (0.5 mg/mL final concentration, 6 h of incubation) after 72 h of treatment (cytarabine, vincristine, and dexamethasone). The formazan dye formed by the viable cells was dissolved by the addition of acid sodium dodecyl sulfate solution (10% SDS, 0.01 mol/L HCl) and overnight incubation at 37°C in the CO₂ incubator. Absorbance was measured at 570 nm in a Synergy H1 equipment (BioTek, Winooski, VT).

2.5 Calcein-AM

To determine cell viability by an orthogonal method, viable cells were discriminated using the CyQUANT Direct Cell Proliferation Assay (Thermo Fisher Scientific) or an in-house labeling with Calcein-AM (1 µg/mL) and propidium iodide (1 µg/mL). Following the completion of an ELDA test, the equipment was halted, but the plate was kept in the plate holder, inside the incubator chamber, to ensure minimal deviation in terms of position and focus with respect to the last ELDA imaging. The CyQUANT reagent was added to the wells, and the plate was incubated for 1 h at 37 °C and 5% CO₂. Subsequently, wells were photographed using the GFP filter on the EVOS M7000 microscope, using the same settings as the last ELDA imaging. Similar experiments were performed by adding a mixture of propidium iodide and Calcein-AM instead of CyQUANT, in which case the plate was incubated for only 30 min at 37 °C, and imaging was performed using both GFP and RFP filters on the EVOS M7000 microscope. The discrimination between ALL cells and T-MSCs was achieved based on differences in size and shape, and the quantification of green (live) and red (dead) cells was performed using the CellProfiler software (Broad Institute, Boston, MA) (40) with Intel Xeon Gold Processor 5118 CPU @ 2.30 GHz, 2.29 GHz, 256 GB RAM.

2.6 Drugs selected for the screening

An FDA-approved anticancer drug library consisting of 166 drugs (kindly provided by the Developmental Therapeutics Program, NCI) was initially screened across eight distinct primary ALL samples. Each drug was tested at six varying concentrations, employing a 10-fold serial dilution starting at 5 µM. The screening assays were conducted as endpoint assessments, with a duration of 72 h. This pilot screening was performed under identical conditions to the ELDA assay (4,000 ALL cells and 100 T-MSCs per well, with collagen and in AIMV supplemented with 10% FBS); however, viable cells were distinguished using Calcein-AM labeling for live cells and propidium iodide staining for dead cells. The CellProfiler software (Broad Institute, Boston, MA) (38) was used to quantify the number of living cells (green fluorescence) and dead cells (red fluorescence) within each well. Drugs showing no activity in any ALL sample at the highest concentration tested were discarded (57 drugs in total). Drugs with activity at ≤0.5 µM in at least one patient sample were pre-selected (49 drugs discarded, 60 drugs pre-selected). Then, the most active 50 drugs were prioritized, covering also various mechanisms of action, as well as drugs of clinical significance in ALL or those involved in promising clinical trials (e.g., imatinib, vorinostat, thioguanine, and ruxolitinib). Detailed information regarding the selected drugs can be found in Supplementary Table 1.

2.7 In vivo drug testing

Ten million PDX ALL cells were injected into NSG mice. Leukemia engraftment was monitored by flow cytometry. When hCD45⁺ cells over total CD45⁺ (mCD45⁺hCD45) cells reached a median average of 0.5%, mice were randomized into the control, dexamethasone (5 mg/kg, intraperitoneal, from Monday to Friday), and vincristine (0.15 mg/kg, intraperitoneal, every Monday) groups. A total of n = 7 mice per treatment group were included in each experiment. Treatment lasted 4 weeks. The percentage of hCD45 in the peripheral blood was monitored weekly. CD45 values between two time points were interpolated linearly based on the straight line connecting the measurements. For survival analysis, a death event was defined as the mouse reaching 25% hCD45 in peripheral blood or exhibiting leukemia-related morbidity (weight loss, lethargy, ruffled fur), whichever occurred first.

2.8 Data analysis

Statistical analysis was done with GraphPad Prism V6.0 (GraphPad Software, La Jolla, CA) using Student’s t-test, two-way ANOVA with Dunnett’s multiple comparison test, or Pearson’s correlation analysis. p-values <0.05 were considered statistically significant and marked as *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. The area under the curve (AUC) and the lethal dose to kill 50% of the cells (LD₅₀) were provided by the EMMA software (35).

3 Results

The ELDA assay is an adaptation of the EMMA method developed for multiple myeloma (33–35) and comprises four key execution steps (Figure 1A). Firstly, a co-culture is established in a black 384-well plate, combining primary ALL cells, T-MSCs, a type 1 collagen matrix, and AIM-V culture medium supplemented with 10% FBS. The second step occurs 16 h later, with the addition of drugs at five different concentrations. Each 384-well plate accommodates 31 drugs in duplicate, alongside negative control wells comprising vehicle (DMSO) and positive control wells where a cell line is treated with the maximal drug concentration in order to validate drug stability and assay performance.

Figure 1

Following the addition of drugs, the plate undergoes time-lapse microscopy, wherein each well is photographed in phase contrast every 30 min for a duration of 80 to 96 h, using the EVOS M7000 microscope (Thermo Fisher Scientific). At the conclusion of this 96-h period, images undergo analysis using a software developed by Silva et al. (2017) (35), which discerns between living and dead cells based on dynamic changes in cell membrane movement over sequential images. The software is capable of distinguishing leukemia cells from stromal cells based on an optical artifact formed by non-adherent cells in phase contrast: a lens-like structure (similar to the appearance used to distinguish live from dead cells in a hemacytometer), in which the cytoplasm appears bright, and the membrane is outlined as a dark ring. In the initial image captured at time zero, ALL cells are identified and pseudo-colored in red (Figure 1B). As time progresses, living cells exhibit membrane movements, which alter the quantification of pixels at their specific XY coordinates in sequential images; these locations are consistently colored red by the software. Conversely, cells undergoing death show no membrane movement, resulting in an unchanged spatial pixel count relative to previous images, and are therefore rendered as non-red (Figure 1C). Representative videos showing ALL cells that remain viable (Supplementary Video 1) or that undergo progressive loss of viability over the course of the experiment (Supplementary Video 2) are provided as supplementary material. Applying this algorithm to all consecutive images from each well of the plate, the software can ascertain the added viability of ALL cells over time in a non-destructive manner, eliminating the necessity to add viability reagents or separate stromal from ALL cells. Consequently, dose–response graphs can be generated based on concentration and exposure time, facilitating the calculation of LD₅₀ and AUC values for each drug tested (Figure 1D).

3.1 ELDA standardization

To determine the minimal number of ALL cells measurable by the image analysis system, while addressing its reproducibility, we cultured primary ALL at different cell densities, ranging from 62 to 16,000 cells per well, in 10 technical replicates. After collagen gelification, the plate was placed in the onstage incubator of the EVOS system and imaged every 30 min for 4 h. Cell numbers were measured using the EMMA software. As shown in Figure 2A, the system could not reliably distinguish between 1,000 from 500 or 62 cells. A geometric progression was apparent between 1,000 and 8,000. At the highest cell density tested (16,000), however, the trend seemed to diminish its slope. Reproducibility remained consistent across replicates.

Figure 2

It is well-established that co-culturing of ALL onto a layer of MSCs prolongs the survival of ALL in vitro (38, 41). However, in ELDA, the addition of too many MSCs is not feasible, as the movement of MSCs can also move surrounding ALL cells in the collagen matrix, which may or may not be living ALL cells, flagging dead cells as live cells. When testing the co-culture of ALL with 2,000 MSCs, excessive movement of the stromal cells caused significant overestimation of ALL cell viability. A similar issue occurred using 800 MSCs per well. Therefore, we systematically reduced the number of MSCs until we identified the optimal number of 100 MSCs per well (Figure 2B), which preserved stromal support while minimizing motion-related artifacts.

The culture medium plays a crucial role in providing essential nutrients and energy sources and maintaining optimal pH and osmolarity levels for cells in vitro. Traditionally, ALL cell lines are cultured using RPMI supplemented with FBS. Primary ALL cells are co-cultured onto bone marrow stromal cells using the serum-free AIM-V medium (25, 42, 43). AIM-V is a nutrient-rich medium formulation based on DMEM and Ham’s F12, enriched with human serum albumin, insulin, transferrin, and a blend of purified non-cytokine factors (44).

The viability of ALL cells cultured in serum-free AIM-V was lower than in RPMI supplemented with 20% FBS. However, when 20% FBS was added to AIM-V, the viability of ALL cells from two out of four patients (50%) was higher than that with RPMI-20% FBS (Figure 2C). Subsequent experiments revealed that comparable results could be achieved with AIM-V supplemented with only 10% FBS (Figure 2D, Supplementary Figure 2).

After performing some ELDA tests, it became evident that ALL cells, with very rare exceptions, did not proliferate in vitro, and some lost viability very rapidly, in less than 48 h. Since freezing and thawing may result in dead cells, we tested whether the removal of apoptotic and dead cells using magnetic beads, before cell seeding, would have an impact on ELDA results. However, no differences were observed by removing apoptotic/dead cells, either under drug-free conditions (Figure 2E, Supplementary Figure 3) or with respect to the LD₅₀ and AUC values for doxorubicin treatment (Figures 2F, G). These results indicate that dead cells do not interfere with ELDA results. As cellular viability is normalized to the number of living cells (in pixels) at the beginning of time-lapse microscopy, i.e., more than 14 h after cell seeding, those cells that were prone to die had already died, thus making no difference in the experiment, provided that enough number of living cells were available to record the effects of the drugs.

3.2 Validation of ELDA results by orthogonal methodologies

The ELDA assay categorizes a cell as viable when it exhibits membrane movements, whereas cessation of movement results in the classification as dead. To validate that movements are indeed specific to living cells and the lack of movements for the dead ones, we used fluorescent dyes to label live and dead cells.

First, the RS4;11 cell line was cultured in the ELDA system with 31 drugs, in duplicate wells, for a duration of 80 h. Upon completion of the ELDA, the CyQUANT Direct Cell Proliferation Assay kit was used to label viable cells, and the plate was reimaged using a GFP filter. The final phase-contrast image obtained from the ELDA analysis was then merged with the GFP-filtered (green fluorescent) image captured after CyQUANT. While ELDA pseudo-colored live cells in red, CyQUANT labeled live cells in green. Consequently, when both methods agreed that the cells were alive, overlapping regions of red and green resulted in shades of yellow. Conversely, when both methods confirmed cell death, no color was observed. Regions colored red only denoted instances where only ELDA classified cells as viable, while green indicated cells solely identified as viable by CyQUANT. The 62 images generated from the analysis of the 31 drugs, conducted in duplicates, were individually assessed. The level of agreement between ELDA and CyQUANT, i.e., the number of discolored plus yellow cells over the total number of cells, was 78% on average. In nearly 90% of the analyzed images, the level of agreement between ELDA and CyQUANT exceeded or was equal to 70%. Only 13% of the images exhibited an agreement of less than 70% regarding the identification of live and dead cells. Moreover, a robust correlation (r = 0.9; p < 0.0001) was observed when comparing the cell viability data obtained concurrently through ELDA and CyQUANT analysis across the different images (Figure 3A). Additionally, a strong correlation (r = 0.84; p < 0.0001) was evident between the total area of green fluorescent cells (in pixels) following CyQUANT labeling and the total area of live cells (in pixels) identified by ELDA (Figure 3B).

Figure 3

In a similar experimental setup, primary ALL cells were subjected to the analysis. Five primary ALL samples, from different genetic subtypes, were seeded into the ELDA system and exposed to six different drugs—cytarabine, vincristine, dexamethasone, idarubicin, doxorubicin, and daunorubicin—at varying concentrations. Following a 72-h incubation period, ELDA imaging was concluded, and the live and dead cells were distinguished by the addition of Calcein-AM and propidium iodide to the plates. Subsequently, the plates were photographed using appropriate fluorescence filters, and cell viability was assessed using the CellProfiler software, which differentiated between ALL and T-MSC cells based on their size disparity. Consistent with previous findings, a robust correlation (r = 0.9; p < 0.0001) was observed between cellular viability outcomes obtained via ELDA and those derived from the Calcein-AM labeling method (Figure 3C).

Lastly, we conducted a comparative analysis between ELDA and the MTT assay across three primary ALL samples, evaluating five different concentrations of cytarabine, vincristine, and dexamethasone. While a positive correlation between ELDA and MTT viability results was observed, the ALL cells exhibited slightly greater resistance to the tested drugs in the ELDA as compared to MTT (Figure 4). To investigate whether the collagen matrix utilized in ELDA could contribute to increased drug resistance in ALL cells, we performed an MTT viability assay with and without collagen. Viability remained highly similar in the presence and absence of collagen (Supplementary Figure 4), dismissing any apparent interference of collagen in the differences observed between the ELDA and MTT assays. In line with existing literature reports (25, 46), it is possible that the inclusion of T-MSCs in ELDA, absent in the MTT assay, confers a protective effect on ALL cells. Other possible explanations for this difference are the higher number of ALL cells per well in the MTT assay and the fact that ELDA measures cell movements while MTT measures mitochondrial activity as a means of assessing the numbers of living cells. In conclusion, even though ELDA has been validated for the discrimination of live and dead cells, the results obtained may not perfectly translate to other drug profiling techniques.

Figure 4

45). ELDA was done using 4,000 cells/well in 384-well plates. Results collected at 72 h are shown.

3.3 Drug screening in 25 pediatric ALL patient samples

To further evaluate the potential of ELDA, we analyzed PDX-derived samples from 31 pediatric ALL cases. The biological and clinical features of these ALL cases are detailed in Supplementary Table 2. As mentioned previously, six samples (19%) failed to meet predefined viability quality criteria (negative control viability ≥ 50%) or were already mostly dead at the start of time-lapse recording, and were therefore excluded from further analysis. The remaining 25 samples were uniformly screened against 50 preclinical and clinical compounds (Supplementary Table 3). For each drug, we utilized five doses optimized from an initial six-point screen (Supplementary Table 1). Notably, the majority of the compounds tested did not exhibit cytotoxic effects on T-MSCs at concentrations lethal to ALL cells (Supplementary Video 3), indicating selective drug activity. Bortezomib, carfilzomib, fedratinib, ixabepilone, omacetaxine, panobinostat, and vincristine elicited cytotoxic effects on T-MSCs at the highest concentration (Supplementary Table 3, Supplementary Video 4).

Compounds such as bortezomib, carfilzomib, and romidepsin demonstrated efficacy at low LD₅₀ values (in nM) and exhibited a narrow LD₅₀ range in most ALL cases. Conversely, certain conventional cytotoxic agents, such as dexamethasone, mitoxantrone, vincristine, idarubicin, and dactinomycin, as well as panobinostat and omacetaxine, displayed responses distributed over a wider concentration range, with pronounced activity observed in the nanomolar range. On the other hand, most tyrosine kinase inhibitors (TKIs), at the concentrations used (5 µM maximal dose), demonstrated limited efficacy across most ALL cases, and when any efficacy was observed, the LD₅₀ values were typically in the micromolar (µM) range (Figure 5).

Figure 5

A positive correlation was observed between the response to drugs of the same class. For instance, histone deacetylase inhibitors such as belinostat, panobinostat, and vorinostat exhibited a positive correlation in drug response. Similarly, among the proteasome inhibitors, including bortezomib, ixazomib, and carfilzomib, a positive correlation in drug response was evident. Furthermore, a positive correlation was observed among the proliferation inhibitors, such as idarubicin and mitoxantrone (Figure 6A). These findings suggest a high level of accuracy of the method for predicting drug responses.

Figure 6

The use of PDX-derived leukemia samples may introduce potential biological biases, including clonal selection during engraftment, adaptation to the murine microenvironment, and partial loss of intrapatient heterogeneity. To directly address the concordance between fresh patient samples and their corresponding PDX, we compared drug response profiles from eight freshly isolated leukemia samples and their matched PDX counterparts across the 50-drug panel. While some quantitative differences in AUC magnitude were noted for individual compounds, a significant positive correlation was observed between AUC values obtained from fresh patient samples and their matched PDX-derived cells (Figure 6B). Thus, despite the potential biological shifts inherent to xenograft propagation, it does not fundamentally distort the drug response landscape captured by ELDA, as the overall drug response patterns were largely preserved. This finding supports the suitability of PDX-expanded material for standardized ex vivo drug profiling.

3.4 In vivo validation of ELDA

To further validate the ELDA results, we analyzed in vivo responses from pairs of patients sharing the same genetic subtype but exhibiting different drug sensitivities (Figure 7). For example, xenograft PT210018 (i-AMP21) was highly sensitive to standard chemotherapy agents, including dexamethasone (LD₅₀ = 0.2 nM) and vincristine (LD₅₀ = 1.6 nM), whereas xenograft PT210321 (i-AMP21) was resistant to both drugs, showing no effect at the highest concentrations tested. Consistent with ELDA predictions, dexamethasone or vincristine treatment significantly prolonged the survival of mice transplanted with PT210018 leukemia (Figure 7A;Supplementary Figure 5), while no survival benefit was observed in PT210321 (Figure 7;Supplementary Figure 5). Clinically, patient PT210018 achieved a good response to induction therapy (MRD-negative at day 33), whereas patient PT210321 had a high MRD level (1.32 × 10⁻²) at the end of induction.

Figure 7

Similar concordance between ELDA and in vivo outcomes was observed in additional patient pairs: PT230353 vs. PT220026 (TCF3-PBX1) and PT230230 vs. PT220233 (B-other), for both dexamethasone (Figure 7) and vincristine (Supplementary Figure 5). For dexamethasone, the delay in leukemia progression in vivo showed a strong correlation with ELDA-derived LD₅₀ values (Spearman r = −0.98, p = 0.0056; Figure 7B).

4 Discussion

Here, we present the standardization and validation of ELDA, a novel drug screening method based on multi-image bright-field microscopy. Parallel validation experiments of ELDA and other techniques, such as Calcein-AM and MTT assays, were performed. A strong correlation between the results was found, while several advantages of ELDA became evident. ELDA offers a significant reduction in the number of cells required for the assay, as it enables the screening of 31 drugs, at five concentrations, in duplicate, using only 1.5 million ALL cells. In ELDA, drug sensitivity is assessed under conditions that more closely resemble the physiological conditions, as ALL cells are cultured in the presence of stromal cells and in a collagen matrix that mimics the extracellular matrix in the bone marrow niche. ELDA is a non-invasive assay based on bright-field microscopy, requiring no cell viability marker substance, which may be toxic to cells or interfere with drug metabolism. Finally, unlike most cell viability assays, which are inherently destructive and limited to specific time points (25, 29, 37, 47, 48), ELDA monitors and measures the survival of cells continuously over time, providing a better evaluation of the kinetics of cell responses to drugs.

ELDA was also validated in vivo using two different PDX ALL cases of the same genetic background (i-AMP21) but with different responses to dexamethasone and vincristine. As expected, the higher the drug sensitivity in ELDA, the higher the survival of mice treated with the corresponding drug, supporting the translational relevance of the assay.

We tested 50 drugs on 25 ALL samples in the ELDA model. In general, compounds such as HDAC and proteasome inhibitors demonstrated efficacy at low LD₅₀ values for most samples. Cytotoxic agents, such as dexamethasone, mitoxantrone, vincristine, dactinomycin, and venetoclax, displayed responses distributed over a wider concentration range, while most TKIs had no effect at the highest tested dose (5 μM) or demonstrated limited efficacy across ALL cases. These results seemed comparable to other drug screening studies in ALL (25, 29). One study, including 60 pediatric ALL samples, revealed that bortezomib exhibited effectiveness at low (IC₅₀ < 10 nM) and narrow IC₅₀ ranges, while venetoclax, dexamethasone, topotecan, and cytarabine displayed a wider concentration range, with varying degrees of activity observed in the nanomolar range across different samples. Additionally, some TKIs such as dasatinib and imatinib achieved either high or no IC₅₀ values across most of the tested samples (25). Another study, with 805 pediatric ALL samples, demonstrated consistent efficacy of bortezomib and panobinostat in almost all samples within the nanomolar range. Conversely, compounds like venetoclax and cytarabine exhibited a more heterogeneous response, and dasatinib was either ineffective or displayed high IC₅₀ values in the majority of samples tested, except for BCR-ABL1 and TCF3-PBX1 ALL cases (29).

Unfortunately, primary ALL cells tend to die when in culture, and this was not different in ELDA. In our experience, a quarter of ALL cases do not survive more than 48 h in vitro, and their drug response curves are not reliable. Preliminary experiment with different grow factors (IL3, IL7, SCF, FLT3L, IGFBP7, IGF1), adhesion molecules (CD40L, VCAM), SPITE Medium Supplement (insulin, transferrin, selenium, pyruvate, and ethanolamine, from Sigma-Aldrich S5666), and a lipid mixture 1 (non-animal derived fatty acids, cholesterol, Tween-80, tocopherol acetate, and Pluronic F-68, from Sigma-Aldrich L0288) render no difference, with the only exception of IL7 (data not shown). It is worth mentioning that stromal cells are able to move dead ALL cells, which may be a confounding effect, especially when ALL cells died too quickly. We recommend checking the time-lapse movies to identify this confounding effect. The ELDA software is under improvement to better discriminate cell death events using changes in cell attributes (49).

Recently, there has been a resurgence in drug profiling for primary ALL (25, 27, 29, 39, 50). Advances in PDX models of ALL and its validation provided sufficient number of ALL cells for the assays. In vitro drug testing using patient-derived or PDX cells has enabled the identification of novel therapeutic agents for ALL (25, 26). Large studies analyzing more than 600 ALL cases have demonstrated an association between in vitro drug-resistance profiles and clinical response (29, 51, 52), highlighting the clinical relevance of functional pharmacotypic profiling. Early approaches evaluated drug resistance in leukemic cells cultured in isolation, with viability determined using metabolic markers such as MTT, and these methods remain in use (29). More automated alternatives, such as fluorescein diacetate (53), and flow cytometry-based platforms enabling simultaneous phenotypic characterization in co-culture systems have further advanced the field (54–56). More recently, 3D co-culture systems incorporating mesenchymal and endothelial cells have been developed, revealing differences between 2D and 3D drug responses (25, 57, 58). However, these platforms do not provide real-time assessment of cell viability based on dynamic cellular behavior.

In this context, ELDA integrates stromal co-culture within a 3D matrix and defines viability through preserved cellular integrity (cellular refringence in bright-field microscopy) and sustained membrane motion over time. Because loss of membrane motion is an early event preceding apoptosis (33), ELDA may detect early apoptotic changes, although it may not fully distinguish intermediate states such as quiescence or sublethal drug effects. Our data further indicate that ELDA-derived drug–response profiles are preserved between freshly isolated patient samples and their matched PDX counterparts, despite potential biological shifts associated with xenograft propagation. ELDA therefore represents a promising and scalable functional drug profiling platform that may complement genomic classification and contribute to pharmacotyping strategies in pediatric ALL. To our knowledge, no independent studies evaluating ELDA have yet been published. The primary objective of this work was to establish and validate ELDA as a robust functional assay and translational research platform. Prospective trials integrating ELDA-guided interventions will be necessary to determine its clinical utility.

References

• 1

  WardEDeSantisCRobbinsAKohlerBJemalA. Childhood and adolescent cancer statistics, 2014. CA Cancer J Clin. (2014) 64:83–103. doi: 10.3322/caac.21219

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  BhojwaniDYangJJPuiCH. Biology of childhood acute lymphoblastic leukemia. Pediatr Clin North Am. (2015) 62:47–60. doi: 10.1016/j.pcl.2014.09.004

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  RaetzEATeacheyDT. T-cell acute lymphoblastic leukemia. Hematol Am Soc Hematol Educ Program. (2016) 2016:580–8. doi: 10.1182/asheducation-2016.1.580

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  InabaHMullighanCG. Pediatric acute lymphoblastic leukemia. Haematologica. (2020) 105:2524–39. doi: 10.3324/haematol.2020.247031

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  KatoMManabeA. Treatment and biology of pediatric acute lymphoblastic leukemia. Pediatr Int. (2018) 60:4–12. doi: 10.1111/ped.13457

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  VroomanLMSilvermanLB. Treatment of childhood acute lymphoblastic leukemia: Prognostic factors and clinical advances. Curr Hematol Malig Rep. (2016) 11:385–94. doi: 10.1007/s11899-016-0337-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  HungerSPMullighanCG. Acute lymphoblastic leukemia in children. N Engl J Med. (2015) 373:1541–52. doi: 10.1056/NEJMra1400972

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  BaileyLCLangeBJRheingoldSRBuninNJ. Bone marrow relapse in paediatric acute lymphoblastic leukaemia. Lancet Oncol. (2008) 9:873–83. doi: 10.1016/S1470-2045(08)70229-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  LocatelliFSchrappeMBernardoMERutellaS. How I treat relapsed childhood acute lymphoblastic leukemia. Blood. (2012) 120:2807–16. doi: 10.1182/blood-2012-02-265884

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  HungerSPRaetzEA. How I treat relapsed acute lymphoblastic leukemia in the pediatric population. Blood. (2020) 136:1803–12. doi: 10.1182/blood.2019004043

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  Graiqevci-UkaVBehluliESpahiuLLiehrTTemajG. Targeted treatment and immunotherapy in high-risk and relapsed/ refractory pediatric acute lymphoblastic leukemia. Curr Pediatr Rev. (2023) 19:150–6. doi: 10.2174/1573396318666220901165247

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  BuchmannSSchrappeMBaruchelABiondiABorowitzMCampbellMet al. Remission, treatment failure, and relapse in pediatric ALL: An international consensus of the Ponte-di-Legno Consortium. Blood. (2022) 139:1785–93. doi: 10.1182/blood.2021012328

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  PietersRKaspersGJKlumperEVeermanAJ. Clinical relevance of in vitro drug resistance testing in childhood acute lymphoblastic leukemia: The state of the art. Med Pediatr Oncol. (1994) 22:299–308. doi: 10.1002/mpo.2950220502

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  HongoTYajimaSSakuraiMHorikoshiYHanadaR. In vitro drug sensitivity testing can predict induction failure and early relapse of childhood acute lymphoblastic leukemia. Blood. (1997) 89:2959–65. doi: 10.1182/blood.V89.8.2959

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  KaspersGJVeermanAJPietersRVan ZantwijkCHSmetsLAVan WeringERet al. In vitro cellular drug resistance and prognosis in newly diagnosed childhood acute lymphoblastic leukemia. Blood. (1997) 90:2723–9. doi: 10.1182/blood.V90.7.2723

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  SrinivasGKusumakumaryPJosephTPillaiMR. In vitro drug sensitivity and apoptosis induction in newly diagnosed pediatric acute lymphoblastic leukemia: Correlation with overall survival. Pediatr Hematol Oncol. (2004) 21:465–73. doi: 10.1080/08880010490477239

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  LönnerholmGThörnISundströmCFrostBMFlaegstadTHeymanMet al. In vitro cellular drug resistance adds prognostic information to other known risk-factors in childhood acute lymphoblastic leukemia. Leukemia Res. (2011) 35:472–8. doi: 10.1016/j.leukres.2010.09.001

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  FrostBMNygrenPGustafssonGForestierEJonssonOGKanervaJet al. Increased in vitro cellular drug resistance is related to poor outcome in high-risk childhood acute lymphoblastic leukaemia. Br J Haematol. (2003) 122:376–85. doi: 10.1046/j.1365-2141.2003.04442.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  HongoTOkadaSInoueNYamadaSYajimaSWatanabeCet al. Two groups of Philadelphia chromosome-positive childhood acute lymphoblastic leukemia classified by pretreatment multidrug sensitivity or resistance in in vitro testing. Int J Hematol. (2002) 76:251–9. doi: 10.1007/BF02982795

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  Janka-SchaubGEHarmsDOden BoerMLVeermanAJPietersR. Die In-vitro-Zytostatika-Resistenz als unabhängiger prognostischer Faktor in Studie COALL-05–92 zur Behandlung der akuten lymphoblastischen Leukämie im Kindesalter: Basis für eine Doppelstratifizierung nach traditionellen Risikokriterien und Resistenzprofil in Studie COALL-06-97. COALL Studiengruppe [In vitro drug resistance as independent prognostic factor in the study COALL-O5–92 Treatment of childhood acute lymphoblastic leukemia; two-tiered classification of treatments based on accepted risk criteria and drug sensitivity profiles in study COALL-06-97. Klin Padiatr. (1999) 211:233–8. doi: 10.1055/s-2008-1043794

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  Den BoerMLHarmsDOPietersRKazemierKMGobelUKörholzDet al. Patient stratification based on prednisolone-vincristine-asparaginase resistance profiles in children with acute lymphoblastic leukemia. J Clin Oncol. (2003) 21:3262–8. doi: 10.1200/JCO.2003.11.031

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  StyczynskiJWysockiM. Is the in vitro drug resistance profile the strongest prognostic factor in childhood acute lymphoblastic leukemia? J Clin Oncol. (2004) 22:963–4. doi: 10.1200/JCO.2004.99.274

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  Van DongenJJSeriuTPanzer-GrümayerERBiondiAPongers-WillemseMJCorralLet al. Prognostic value of minimal residual disease in acute lymphoblastic leukaemia in childhood. Lancet. (1998) 352:1731–8. doi: 10.1016/S0140-6736(98)04058-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  KruseAAbdel-AzimNKimHNRuanYPhanVOganaHet al. Minimal residual disease detection in acute lymphoblastic leukemia. Int J Mol Sci. (2020) 21:1054. doi: 10.3390/ijms21031054

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  FrismantasVDobayMPRinaldiATchindaJDunnSHKunzJet al. Ex vivo drug response profiling detects recurrent sensitivity patterns in drug-resistant acute lymphoblastic leukemia. Blood. (2017) 129:e26–37. doi: 10.1182/blood-2016-09-738070

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  FischerUForsterMRinaldiARischTSungaleeSWarnatzHJet al. Genomics and drug profiling of fatal TCF3-HLF-positive acute lymphoblastic leukemia identifies recurrent mutation patterns and therapeutic options. Nat Genet. (2015) 47:1020–9. doi: 10.1038/ng.3362

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  GengHHurtzCLenzKBChenZBaumjohannDThompsonSet al. Self-enforcing feedback activation between BCL6 and pre-B cell receptor signaling defines a distinct subtype of acute lymphoblastic leukemia. Cancer Cell. (2015) 27:409–25. doi: 10.1016/j.ccell.2015.02.003

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  LaukkanenSVelosoAYanCOksaLAlpertEJDoDet al. Therapeutic targeting of LCK tyrosine kinase and mTOR signaling in T-cell acute lymphoblastic leukemia. Blood. (2022) 140:1891–906. doi: 10.1182/blood.2021015106

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  LeeSHRYangWGochoYJohnARowlandLSmartBet al. Pharmacotypes across the genomic landscape of pediatric acute lymphoblastic leukemia and impact on treatment response. Nat Med. (2023) 29:170–9. doi: 10.1038/s41591-022-02112-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  Dal BelloRPasanisiJJoudinaudRDuchmannMPardieuBAyakaPet al. A multiparametric niche-like drug screening platform in acute myeloid leukemia. Blood Cancer J. (2022) 12:95. doi: 10.1038/s41408-022-00689-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  PerezDRNicklCKWallerADelgado-MartinCWoodsTSharmaNDet al. High-throughput flow cytometry identifies small-molecule inhibitors for drug repurposing in T-ALL. Slas Discov. (2018) 23:732–41. doi: 10.1177/2472555218774248

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  MiharaKImaiCCoustan-SmithEDomeJSDominiciMVaninEet al. Development and functional characterization of human bone marrow mesenchymal cells immortalized by enforced expression of telomerase. Br J Haematol. (2003) 120:846–9. doi: 10.1046/j.1365-2141.2003.04217.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  KhinZPRibeiroMLJacobsonTHazlehurstLPerezLBazRet al. A preclinical assay for chemosensitivity in multiple myeloma. Cancer Res. (2014) 74:56–67. doi: 10.1158/0008-5472.CAN-13-2397

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  SilvaAJacobsonTMeadsMDistlerAShainK. An organotypic high throughput system for characterization of drug sensitivity of primary multiple myeloma cells. J Vis Exp. (2015) e53070. doi: 10.3791/53070

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  SilvaASilvaMCSudalaguntaPDistlerAJacobsonTCollinsAet al. An ex vivo platform for the prediction of clinical response in multiple myeloma. Cancer Res. (2017) 77:3336–51. doi: 10.1158/0008-5472.CAN-17-0502

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  ThévenazPRuttimannUEUnserM. A pyramid approach to subpixel registration based on intensity. IEEE Trans Image Process. (1998) 7:27–41. doi: 10.1109/83.650848

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  SternbergSR. Biomedical image processing. Computer. (1983) 16:22–34. doi: 10.1109/MC.1983.1654163

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  ManabeACoustan-SmithEBehmFGRaimondiSCCampanaD. Bone marrow-derived stromal cells prevent apoptotic cell death in B-lineage acute lymphoblastic leukemia. Blood. (1992) 79:2370–7. doi: 10.1182/blood.V79.9.2370.2370

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  SbranaFVPinosRBarbaglioFRibezziDScagnoliFScarfòLet al. 3D bioprinting allows the establishment of long-term 3D culture model for chronic lymphocytic leukemia cells. Front Immunol. (2021) 12:639572. doi: 10.3389/fimmu.2021.639572

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  StirlingDRSwain-BowdenMJLucasAMCarpenterAECiminiBAGoodmanA. CellProfiler 4: Improvements in speed, utility and usability. BMC Bioinf. (2021) 22:433. doi: 10.1186/s12859-021-04344-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  CampanaDCoustan-SmithEKumagaiMAManabeA. Growth requirements of normal and leukemic human B cell progenitors. Leuk Lymphoma. (1994) 13:359–71. doi: 10.3109/10428199409049624

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 42

  WangLFortneyJEGibsonLF. Stromal cell protection of B-lineage acute lymphoblastic leukemic cells during chemotherapy requires active Akt. Leuk Res. (2004) 28:733–42. doi: 10.1016/j.leukres.2003.10.033

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 43

  XuHWangNCaoWHuangLZhouJShengL. Influence of various medium environment to in vitro human T cell culture. In Vitro Cell Dev Biol Anim. (2018) 54:559–66. doi: 10.1007/s11626-018-0273-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 44

  McMillinDWNegriJMMitsiadesCS. The role of tumour-stromal interactions in modifying drug response: Challenges and opportunities. Nat Rev Drug Discov. (2013) 12:217–28. doi: 10.1038/nrd3870

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 45

  HollemanACheokMHden BoerMLYangWVeermanAJKazemierKMet al. Gene-expression patterns in drug-resistant acute lymphoblastic leukemia cells and response to treatment. N Engl J Med. (2004) 351:533–42. doi: 10.1056/NEJMoa033513

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 46

  HelinskiEHBielatKLOvakGMPaulyJL. Long-term cultivation of functional human macrophages in Teflon dishes with serum-free media. J Leukoc Biol. (1988) 44:111–21. doi: 10.1002/jlb.44.2.111

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 47

  KarásziEJakabKHomolyaLSzakácsGHollóZTelekBet al. Calcein assay for multidrug resistance reliably predicts therapy response and survival rate in acute myeloid leukaemia. Br J Haematol. (2001) 112:308–14. doi: 10.1046/j.1365-2141.2001.02554.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 48

  HayonTDvilanskyAShpilbergONathanI. Appraisal of the MTT-based assay as a useful tool for predicting drug chemosensitivity in leukemia. Leuk Lymphoma. (2003) 44:1957–62. doi: 10.1080/1042819031000116607

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 49

  JiangQSudalaguntaPSilvaMCCanevaroloRRZhaoXAhmedKTet al. CancerCellTracker: A brightfield time-lapse microscopy framework for cancer drug sensitivity estimation. Bioinformatics. (2022) 38:4002–10. doi: 10.1093/bioinformatics/btac417

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 50

  BrideKLHuHTikhonovaAFullerTJVincentTLShraimRet al. Rational drug combinations with CDK4/6 inhibitors in acute lymphoblastic leukemia. Haematologica. (2022) 107:1746–57. doi: 10.3324/haematol.2021.279410

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 51

  LiZZhaoHYangWMaillardMYoshimuraSHsiaoYCet al. Molecular and pharmacological heterogeneity of ETV6::RUNX1 acute lymphoblastic leukemia. Nat Commun. (2025) 16:1153. doi: 10.1038/s41467-025-56229-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 52

  EnbladAPKraliOGezeliusHLundmarkABlomKAnderssonCet al. Ex vivo drug responses and molecular profiles of 597 pediatric acute lymphoblastic leukemia patients. Hemasphere. (2025) 9:e70176. doi: 10.1002/hem3.70176

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 53

  LindhagenENygrenPLarssonR. The fluorometric microculture cytotoxicity assay. Nat Protoc. (2008) 3:1364–9. doi: 10.1038/nprot.2008.114

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 54

  Groth-PedersenLChenYHFaberMValentinRAlbertsenBKWehnerPSet al. A novel chemosensitivity profiling platform for small acute lymphoblastic leukemia cell populations. Leuk Lymphoma. (2015) 56:2208–11. doi: 10.3109/10428194.2014.996851

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 55

  GochoYLiuJHuJYangWDhariaNVZhangJet al. Network-based systems pharmacology reveals heterogeneity in LCK and BCL2 signaling and therapeutic sensitivity of T-cell acute lymphoblastic leukemia. Nat Cancer. (2021) 2:284–99. doi: 10.1038/s43018-020-00167-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 56

  NiedermayerAStursbergJBergmannAKZimmermannMCarioGBrüggemannMet al. Remission induction in refractory, drug resistant pediatric PICALM::MLLT10+ B-cell acute lymphoblastic leukemia by venetoclax. Leukemia. (2025) 39:1520–5. doi: 10.1038/s41375-025-02591-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 57

  RowlandLSmartBBrownADettorreGMGochoYHuntJet al. Ex vivo drug sensitivity imaging-based platform for primary acute lymphoblastic leukemia cells. Bio Protoc. (2023) 13:e4731. doi: 10.21769/BioProtoc.4731

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 58

  KanariMJimenez GarciaISteffenFDKrattigerLABataclanCLiuWet al. A three-dimensional ex vivo model recapitulates in vivo features and drug resistance phenotypes in childhood acute lymphoblastic leukemia. Leukemia. (2025) 39:2881–94. doi: 10.1038/s41375-025-02739-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article