A 3D vasculature-like model for ALL patient-derived xenografts

To evaluate the role of the non-hematopoietic BM in the leukemic ecosystem, we developed a 3D model that mimics blood vessel-like structures, enabling functional studies of their interactions with leukemic cells. This was achieved by co-culturing stromal, endothelial and leukemic cells in 96-well imaging plates containing pre-cast, optically transparent synthetic hydrogels. These hydrogels are based PEG-peptide bioconjugates and contain adhesion and degradation motives supporting cell spreading, MMP-mediated gel degradation and formation of cell-produced ECM. Moreover, an in-depth surface density gradient allows the characterization of motility patterns for different cell types (Fig. 1A) [23,24,25,26,27]. The primary human BM MSCs are necessary for endothelial spreading, acting as a supporting scaffold (Supplementary Fig. 1A), while no “vascularization” occurs in their absence (Supplementary Fig. 1B). Interestingly, the hTERT-MSC (human telomerase reverse transcriptase-MSC) cell line is not capable of supporting this “vascularization”, highlighting the importance of using primary MSCs for a blood-vessel mimic (Supplementary Fig. 1C). 3D segmentation and characterization of the structures revealed consistent volumes and surfaces of network amongst the control, an MSC-HUVEC co-culture, and MSC-HUVEC-ALL co-cultures established with 10 PDXs, thus validating the robustness of the model across multiple conditions (Fig. 1B–D).

Fig. 1: Description of the 3D vasculature-like model for leukemic PDXs.

A Experimental set-up. The 3DProSeed® Hydrogel Well Plate was used for the establishment of the ex vivo model. Primary MSCs are cultured for 24 h, followed by the addition of HUVECs, MSCs and ALL PDXs. B Representative confocal fluorescence images of the co-cultures in maximum intensity projections using confocal z-stacks (z-step: 10 μm) showing MSCs (blue), HUVECs (green), and ALL cells (red). C, D A vasculature-like network is detected in a similar manner in all 11 conditions tested (MSC-HUVEC controls, 10 PDXs co-cultures with MSC and HUVECS). Depicted are the (C) mean volume and (D) mean surface of the HUVECs (“vasculature”) in triplicates. Controls were quantified in triplicates in three separate experiments (statistical analysis: unpaired t test, significant p value < 0.05).

Leukemic cells exhibit enhanced motility upon immediate interaction with the supporting cells

To analyze the influence of the microenvironment on leukemic cell behavior, we performed timelapse confocal imaging directly after seeding for 30 h with a timestep of 5 min, comparing 3D ALL mono-cultures and ALL co-cultures with MSCs and ECs. It is evident that the leukemic PDXs directly interact with the supporting cells as they form multi-cellular clusters over time. Cell-to-cell distance analysis by single cell segmentation quantification revealed enhanced leukemic aggregation over time, with decreasing ALL to ALL cell distances in a 12-h course (Fig. 2A). Significantly higher speed and displacement in the XY plane was observed for the leukemic cells in co-culture compared to mono-culture, as measured in the first 6 h of the timelapse imaging (Fig. 2B, C). The movement of the ALL cells in the z-axis was also influenced by the presence of supporting cells. ALL cells in mono-culture remained concentrated around the median position, with decreasing standard deviation from 0 to 30 h, indicating no substantial migration (Fig. 2D). In contrast, in co-culture we observed an increased migratory behavior of the leukemia cells with a subpopulation exhibiting increased penetration capacity, as indicated by the increased values of both delta median and standard deviation at time 30 compared to time 0 (Fig. 2E).

Fig. 2: ALL PDX cells exhibit distinct phenotypes upon immediate interaction with microenvironment cells.

A Maximum intensity projections of confocal z-stacks (z-step: 7.8 μm) showing the immediate formation of multicellular clusters consisting of MSCs (blue), HUVECs (green) and ALL cells (red) at 0 h, 6 h and 12 h upon seeding. The leukemic cells follow the clustering behavior of MSCs and HUVECs towards the network formation (12 h, 5 min interval). The corresponding bar plots show decreased ALL cell-to-cell distances, quantified as the minimum 3D distances between two leukemic cells (3D single-cell analysis, bin center = 5). B, C Motility analysis of segmented ALL single cells: (B) speed (C) displacement (bin center = 2, statistical analysis: Kolmogorov-Smirnov unpaired t tests with p < 0.0001). D, E Image-based single-cell segmentation and quantification of the vertical positioning (z-axis) of ALL cells, normalization based on the median position of each timepoint. Data obtained by 30-h live cell imaging with a 5 min interval. Depicted is the Δ in median z-position from timepoint 0 to 30 based on the normalized data (and actual values in μm), as well as the standard deviation (SD) at 0 (T0) and 30 (T30) hours. F 3D segmentation of all cell types at 72 h enables proximity analysis of ALL cells relative to MSCs or HUVECS, dH: minimum ALL distance to HUVEC, dM: minimum ALL distance to MSC. G Percentage of ALL cells located within <10 μm of MSC and HUVEC (statistical analysis: paired Wilcoxon signed rank test with p < 0.05, n = 3 per condition, 10 different PDXs as biological replicates). H, I Topological analysis of ALL cells stratified by molecular subtype. H B-ALL PDXs are significantly closer to MSCs than T-ALL PDXs, while (I) no subtype-dependent difference is observed in proximity to HUVECs (unpaired t-test with Welch’s correction, p < 0.05; n = 3 per condition).

Furthermore, the positioning of the leukemic cells towards MSCs and HUVECs was evaluated after 72 h, where the network formation is completed. Full-volume segmentation and proximity analysis of all cell types was performed and the minimum distance of each ALL to the nearest respective cell type (dH: ALL distance to HUVEC, dM: ALL distance to MSC) was calculated (Fig. 2F). Increased 3D proximity to MSCs compared to HUVECs was observed for 9/10 of the PDXs analyzed, with the exception of one T-ALL PDX residing closer to HUVECs, while the localization towards the two cell types was not directly correlated (Fig. 2G). When stratified by molecular subtype, B-ALL PDXs showed markedly closer proximity to MSCs than T-ALL PDXs, while no significant differences were observed in ALL cell distances to HUVECs (Fig. 2H,I). These findings suggest that B-ALL cells exhibit a subtype-specific topological bias toward mesenchymal stromal cells within the leukemic niche model. No significant difference in spatial localization could be detected between molecular subtypes across the B-ALL and T-ALL PDXs (Supplementary Fig. 1D, E).

MSCs exhibit multi-lineage differentiation potential in the blood-vessel mimic, resembling in vivo cell fate

To better understand the heterogeneity amongst ALL PDX samples and their associated bone marrow-like microenvironments, we performed single-cell RNA sequencing (scRNA-seq) to decipher the transcriptomic signatures of these distinct phenotypes. The conditions included all the cell types (MSC, HUVEC, ALL) cultured alone, an MSC-HUVEC control and 6 PDX leukemic co-cultures (MSC-HUVEC-ALL, referred as PDX1-PDX6 based on the leukemic PDX per condition) of different ALL molecular subtypes. With the use of established cell surface markers (CD19, and CD7 for B- and T-ALL, THY-1 for MSC and CDH5 for HUVEC) for their identification, both supporting and leukemic cells were analyzed by clusters.

Enhanced MSC differentiation was observed in the presence of HUVECs, both in the control (MSC-HUVEC) and leukemic conditions (MSC-HUVEC-ALL) compared to the MSC control, and well-established gene sets were used to identify the different MSC clusters (Fig. 3A, Supplementary Fig. 2A) [3, 28,29,30,31,32,33,34]. Notably, a polarization based on the known MSC-markers PDGFRA (platelet-derived growth factor receptor alpha) and CXCL12 (CXC motif chemokine 12) was observed, consistent with in vivo and human primary MSCs, with PDGFRA-MSCs exhibiting high multipotent capacity and CXCL12-MSCs preferentially committed to osteo-lineages (Fig. 3B) [35,36,37,38]. Unsupervised clustering revealed nine MSC subpopulations, including precursors of osteo-, adipo-, fibro- and chondro-lineages, with some cells characterized by the co-expression of genes related to more than one lineage. Other clusters included cycling MSCs with high MKI67 expression and more committed populations such as perivascular and smooth muscle cells (Fig. 3C, Supplementary Fig. 2B-D). Moreover, increased differentiation towards smooth muscle cells and pre-fibroblasts was observed, potentially due to their capability of producing vascular promoting and supporting factors such as the vascular endothelial growth factor (VEGF) [39, 40]. In the leukemic conditions (MSC-HUVEC-ALL), a modest enrichment (5%) in the fibro-lineage is detected, possibly describing the transition into a cancer-associated fibroblastic phenotype that facilitates the homing of ALL cells close to the network (Fig. 3D) [41]. Gene set enrichment analysis (GSEA) suggests that MSCs support vascularization when co-cultured with endothelial cells, consistent with findings from a previous 3D co-culture leukemia model [14]. Specifically, blood vessel morphogenesis and endothelium development were among the most upregulated pathways, with ECM-remodeling or pro-angiogenic genes such as COL5A3, COL4A1, MMP9, TGFB1, KDR and FLT1 being enriched (Fig. 3E). Notably, when comparing MSCs from the MSC-HUVEC-ALL co-cultures to MSCs cultured alone, pathways related to external encapsulating structure organization are enriched, while not in the MSC-HUVEC control, suggesting leukemia-specific remodeling of the stromal matrix (Supplementary Fig.2E, F). Finally, cell-cell communication network analysis revealed known supporting interactions between MSCs and ALL, with the highest probability being through CD74 and CD44 signaling (Supplementary Fig. 2G).

Fig. 3: MSC multi-lineage differentiation in 3D resembles in vivo and human data.

A UMAP projections showing MSC population overlap based on the culturing condition; left: two controls, i.e. MSC and MSC ( + HUVEC), right: MSC ( + HUVEC + ALL), PDX1-PDX6 indicating the six different ALL PDXs). B Feature plots depicting the expression of CXCL12 (top) and PDGFRA (bottom) across the MSC populations. C Identification of multi-lineage MSC subpopulations based on established lineage-specific gene signatures, projected in UMAP, upon coculturing in 3D. D Proportional distribution of MSC subtypes across conditions, with the increased population percentages in white. Arrows indicating increase or decrease upon co-culture. E GSEA dot plot showing the top 10 enriched pathways in MSCs from the MSC( + HUVEC) and MSC( + HUVEC + ALL) to the MSC control comparison.

GSEA indicates EC-induced ECM reorganization in the presence of leukemic cells

On the other hand, ECs exhibit less complexity but still consist of distinct subpopulations including a mesenchymal-like, a pre-vascular and a vascular cluster (Fig. 4A–C), with a notable transition to the vascular-like phenotype observed in all co-cultures (Fig. 4D, Supplementary Fig. 2H) [28, 42, 43]. Another significant observation is the enrichment of the extracellular matrix degradation pathway, and other ECM-related ones, exclusively in the leukemic conditions (MSC-HUVEC-ALL co-cultures), while in the control (MSC-HUVEC) only enhanced proliferation is observed when comparing to the HUVEC control (Fig. 4E, Supplementary Fig. 2I). This could indicate that endothelial cells degrade and re-organize the ECM upon leukemic infiltration in order to create a favorable niche, as they are already known regulators of ECM dynamics [26].

Fig. 4: Endothelial cells re-organize the microenvironment in the leukemic conditions.

A UMAP projections showing HUVEC population overlap based on the culturing condition; left: two controls, i.e. HUVEC and HUVEC ( + MSC), right: HUVEC ( + MSC + ALL, PDX1-PDX6 indicating the six different ALL PDXs). B Dot plot depicting the expression of representative marker genes used to identify transcriptionally distinct HUVEC subpopulations. C UMAP projection with the corresponding HUVEC subpopulations. D Stacked bar plot depicting the relative abundance of each HUVEC subpopulation per condition. E GSEA dot plot showing degradation of the ECM amongst the top 5 pathways enriched in the HUVECs cultured in MSC-HUVEC-ALL co-cultures.

Leukemic gene expression in the 3D microenvironment mirrors signatures detected in xenografts in vivo

To identify potential differences in the transcriptome of leukemia cells in mono- and in co-culture, we performed scRNA-seq of 6 different ALL PDXs under both conditions. In order to capture potential subtype heterogeneity, 4 different ALL subtypes were selected including two TCF3::HLF positive B-ALLs, two TCF3::PBX1 B-ALLs, one cortical T-ALL and one pre-T-ALL. In addition, these PDXs were also extracted and sequenced directly from the murine bone marrow (referred to as in vivo) with the aim of identifying potential similarities between those conditions (Fig. 5A). All the cells were successfully integrated and clustered based on the ALL subtype (Fig. 5B, Supplementary Fig. 3A). Differential gene expression analysis (DGEA) revealed upregulation of genes that are associated to B- and T-cell development as well as leukemogenesis when comparing co-cultured to mono-cultured ALLs. The corresponding dot plots highlight the similar expression of these genes in vivo and in co-culture, suggesting that the ex vivo model recapitulates essential features for leukemic development (Fig. 5C, D).

Fig. 5: Comparison of leukemic gene expression from 3D co-culture and in vivo leukemia xenografts.

A Sequencing of leukemic PDXs in three different conditions, mono-, co-culture and bone marrow cells from leukemia xenografts in vivo. B UMAP projections of the integrated datasets (RPCA reduction through Seurat), visualized by condition (left) and leukemia subtype (right). Subtype is the leading source of variability, determining the clustering in the UMAP projection. C Expression of leukemia-associated genes for B-ALL (scaled dot plot) across conditions. D Expression of leukemia-associated genes for T-ALL (scaled dot plot) across conditions. E GSEA comparing co-cultured ALL cells to mono-cultured. Depicted are the top 10 upregulated hallmark pathways. EMT is revealed as the most upregulated pathway in co-cultures. F Violin plot showing the expression of the EMT signature expression in B- and T-ALL cells for the co-culture, revealing higher EMT activity in B-ALL. G Stacked bar plots showing the proportion of cycling and non-cycling cells in B- and T-ALL for the two conditions, co-culture and in vivo. H GSEA bar plot depicting the top 10 upregulated pathways comparing non-cycling (G1) to cycling cells (G2M and S). I UMAP projections depicting the B-ALL (top, pdx1-pdx4) and T-ALL (bottom, pdx5-pdx6) PDXs, linked to expression of known ALL cell surface markers.

Further comparison revealed that genes regulating migration, adhesion and ECM-remodeling are amongst the most upregulated genes in co-cultured ALL (Supplementary Fig. 3B). The epithelial-mesenchymal transition (EMT) hallmark, described by some of the previously mentioned genes, was found to be the top pathway enriched, while the overall regulatory mechanism of EMT was consistently upregulated as well (Fig. 5E, Supplementary Fig. 3C). Notably, in recent studies the EMT pathway has been associated with increased aggregation and invasiveness in the context of leukemia, reminiscent of the enhanced motility upon immediate ALL-microenvironment interaction we have observed (Fig. 2A–E) [44,45,46,47,48]. Independent analysis of B- and T-ALL cells validated EMT enrichment in both co-cultured subtypes, while this signature was predominant in B-ALL with approximately 70% overall expression (Fig. 5F, Supplementary Fig. 3D, E). An EMT signature was also detected in PDX that were isolated from the murine bone marrow, more prominently in B-ALL compared to T-ALL (Supplementary Fig. 3F). B-ALL has been described by enhanced microenvironment interactions in vivo compared to T-ALL, in line with its dependency on an EMT-like phenotype [10]. Moreover, association of cell cycle arrest and EMT-like leukemic populations has also been described [47]. Cell cycle analysis revealed that the 3D co-cultures successfully reiterate in vivo subtype-based proliferation heterogeneity with similar proportions of cycling and non-cycling cells; however, GSEA revealed that the EMT pathway is actually predominantly regulated by the non-cycling cells, potentially confirming a connection between cell cycle arrest and the EMT leukemic traits (Fig. 5G,H, Supplementary Fig. 3G). Finally, subpopulation heterogeneity based on known cell surface markers was detected, with cells exhibiting various expression profiles across the same PDX (Fig. 5I).

Preserved cell heterogeneity in co-culture is validated by single cell microscopy and flow cytometry

In order to investigate whether an EMT-like, i.e. aggregated and migratory, phenotype is indeed present in the co-cultures, phenotypic and functional characterization was performed. For that purpose, thirteen PDXs of different subtypes including TCF3::HLF, TCF3::PBX1, B-other and T-ALL were seeded in both mono- and co-culture and analyzed using confocal microscopy and flow cytometry after 72 h (Fig.6A). Single-cell distance analysis revealed increased aggregation in the co-culture, with the majority of the leukemic cells maintaining close distances of approximately 10 μm (Fig. 6B, Supplementary Fig. 4A). Subtype-based analysis revealed enhanced clustering in B-ALL compared to T-ALL, associating B-ALL to the observed EMT increase in our transcriptomic data (Fig. 6C, Fig. 5F). In addition, migration of the ALL cells in the hydrogel was promoted by the non-hematopoietic cells when compared to mono-cultures, where minimum migration is detected and only a subset of T-ALL cells exhibited enhanced migratory capacity, yet again confirming T-ALL versatility (Fig. 6D, E). On the other hand, migration is observed in the case of all co-cultures (Fig. 6F). The portion of leukemic cells, in direct contact with the vasculature-like network, was assessed based on the HUVECs migration capability, with the network settling at a depth of 40% distance from the top (with the highest leukemia cell as position reference) regardless of the ALL subtype used (Fig. 6G). No leukemia cells were found in this depth in mono-cultures, while 20–40% of the co-cultured ALLs were able to migrate closely to the network, irrespective of the subtype (Fig. 6H, Supplementary Fig. 4B, C).

Fig. 6: 3D co-culture-enhanced ALL aggregation, migration and proliferation lead to increased heterogeneity.

A Pie chart showing that 13 PDXs of different subtypes were analyzed in the 3D model. B Immediate cell-cell proximity ( < 10 μm) is enriched in co-cultures, suggesting superior cell communication (n = 3, PDXs used as biological replicates, statistical analysis: Wilcoxon matched-pairs signed rank test, p value < 0.05). C Immediate contact is significantly enriched in B-ALL samples compared to T-ALL (same statistics as B). D Representative 3D reconstruction of B-ALL cells in the hydrogel. E, F Single-cell spatial distribution normalized to the highest ALL z-position per well. Light gray: B-ALL PDXs, Dark gray: T-ALL PDXs. G Dot plot showing the migration percentage of HUVECs, normalized to the highest leukemic cell position detected in the hydrogel (n = 3, statistical analysis: unpaired t test, p value < 0.05). H Migration analysis reveals that approximately 20–40% of ALL cells are in direct contact with the vessel-like structure (n = 3, statistical analysis: Wilcoxon matched-pairs signed rank test, p value < 0.05). Migration percentages are based on the relative position of each cell to the highest localized ALL cell. I Proliferation assessment with CellTrace Violet via Flow Cytometry. The CellTrace intensity reveals a non-cycling (defined by timepoint 0, high CellTrace intensity), a slow cycling (medium CellTrace intensity) and a high cycling population (low CellTrace intensity). J Plots depicting the percentage of cells in each phase across the 13 PDXs per condition. K Immunofluorescent staining with CD19 for lymphocyte (B-ALL) identification. Both CellTrace positive (yellow arrows, non-cycling) and negative cells (white arrows, cycling) are detected. L Stacked bar plots depicting the proportions of non-cycling, slow cycling and high cycling cells in 3D and 2D co-culture respectively on day 7.

Furthermore, the leukemic PDXs were stained with CellTrace Violet and their proliferation was monitored for 7 days (Fig. 6I, Supplementary Fig. 4D). On day 3, no significant difference is observed across the conditions or subtypes, as expected based on the transcriptomic data (Fig. 6J, Supplementary Fig. 4E–G). However, at a later timepoint (day 7), increased proliferation is observed in the co-culture, showing beneficial conditions for leukemia’s growth. Interestingly, a portion of the leukemic cells decrease or pause their cellular growth rate, validating the existence of a non-cycling population as well as proving that the 3D co-culture maintains the cell cycle heterogeneity that is observed in in vivo studies (Fig. 6J). These findings were also validated by immunofluorescence, where upon CD19 staining for the lymphocyte identification, the presence of both CellTrace positive (yellow arrows, non-cycling) and negative cells (white arrows, cycling) was evident (Fig. 6K). Nonetheless, no specific localization based on the cycling state was observed, similarly to our previous in vivo observations where, upon chemotherapy, no selective cell death occurs based on the proliferation phase of the cells (Supplementary Fig. 4H–K) [10]. Finally, comparison of the proliferation rate of the ALL PDXs to the standard 2D hTERT MSC-ALL co-culture system revealed preserved cell state heterogeneity in 3D. Leukemic cells present a distinct proliferation phenotype in 2D with patient samples either strongly proliferating or completely lacking division capacity, while in 3D both cycling and non-cycling populations are detected for all PDXs on day 7, thus highlighting that the 3D model can recapitulate cellular complexity ex vivo when 2D systems fail (Fig. 6L).

Drug response patterns are reproduced, with increased resistance in 3D

Finally, we investigated whether the 3D co-cultures can be utilized for drug response evaluations and if the distinct phenotypes observed alter drug sensitivity. Flow cytometry was performed to evaluate the 3D response. For that purpose, 16 PDXs of various subtypes were treated for 72 h, with compounds of interest specifically for each subgroup, and then enzymatically extracted for evaluation. Flow cytometry analysis was performed by gating for CellTrace positive cells, i.e. lymphocytes, based on unstained controls, followed by propidium iodide gating for the exclusion of dead cells (Fig. 7A). The viability of the ALL cells was compared to 2D controls that were treated in the same manner and analyzed by high throughput microscopy (Fig. 7B).

Fig. 7: Drug screenings in the 3D co-culture reveal increased resistance.

A Gating strategy for flow cytometry. Viable ALL cells were gaited for lymphocytes based on their size, followed by doublet exclusion, cellTrace positivity and PI negativity. B Experimental design. Drug responses in the 3D co-culture system were compared to those in a standardized drug response profiling protocol, where immortalized hTERT-MSCs-ALL co-cultures are treated for 72 h (n = 3). C Bar plots showing the normalized response values (integrated normalized cell counts over two concentrations) for all the 3D and 2D conditions (n = 3, statistical analysis: ratio paired t test, p value < 0.05). D Percentile ranking indicating the response patterns across the conditions, color-coded based on the response in 3D. E–I Dot plots showing the normalized viability of the leukemic cells upon treatment with different compounds. Depicted is the response to 100 and 10,000 nM in the 2D and 3D systems respectively. From left to right : (E) idasanutlin, (F) venetoclax, (G) doxorubicin, (H) vincristine, and (I) dexamethasone.

The predictive value of the 3D model was validated as the samples were following the response patterns of the established 2D drug response profiling (DRP) platform (Fig. 7C). Percentile ranking was performed to confirm the sensitive/resistant samples of each cohort (Fig. 7D). Remarkably, there was increased viability in 3D upon treatment for the majority of the PDXs (Fig. 7E–I). 14 out of 20 samples (70%) showed decreased sensitivity (>10% difference in normalized cell counts to DMSO) at 100 nM for the 3D co-cultures, suggesting superior ALL protection upon treatment. This phenotype can be attributed to multiple factors, as the presence of both ECM and ECs have been linked to leukemic treatment escape [26, 49, 50]. In addition, a remaining subpopulation can be observed at 10 μM for 4 drugs (exception doxorubicin) indicating the existence of a resistant subpopulation that cannot be detected in 2D (Fig. 7E–I). Finally, a comparative analysis of the steroid response in 2D and 3D was performed for 8 ALL patients with known poor response during the steroid prephase of the AIEOP-BFM-ALL treatment protocol (Clinical information, Supplementary Table 2). Consistent with our observations, significantly increased normalized viability was observed in 3D for poor response ALL patients (Supplementary Fig. 5A). This difference was also evident through normalized integrated viability assessment (integrated normalized cell counts over two concentrations) for both B- and T-ALL samples (Supplementary Fig. 5B). However, further validation of the comparative predictive value of 2D and 3D responses using clinically annotated samples, as well as the inclusion of additional chemotherapeutic agents, is required.

Taken together, this model recapitulates established drug response patterns thus can be used for compound testing applications, while the decreased drug efficacy observed could also potentially lead to the identification of resistance patterns more readily compared to 2D.