Three-dimensional tumor cell growth stimulates autophagic flux and recapitulates chemotherapy resistance

Current preclinical models in tumor biology are limited in their ability to recapitulate relevant (patho-) physiological processes, including autophagy. Three-dimensional (3D) growth cultures have frequently been proposed to overcome the lack of correlation between two-dimensional (2D) monolayer cell cultures and human tumors in preclinical drug testing. Besides 3D growth, it is also advantageous to simulate shear stress, compound flux and removal of metabolites, e.g., via bioreactor systems, through which culture medium is constantly pumped at a flow rate reflecting physiological conditions. Here we show that both static 3D growth and 3D growth within a bioreactor system modulate key hallmarks of cancer cells, including proliferation and cell death as well as macroautophagy, a recycling pathway often activated by highly proliferative tumors to cope with metabolic stress. The autophagy-related gene expression profiles of 2D-grown cells are substantially different from those of 3D-grown cells and tumor tissue. Autophagy-controlling transcription factors, such as TFEB and FOXO3, are upregulated in tumors, and 3D-grown cells have increased expression compared with cells grown in 2D conditions. Three-dimensional cultures depleted of the autophagy mediators BECN1, ATG5 or ATG7 or the transcription factor FOXO3, are more sensitive to cytotoxic treatment. Accordingly, combining cytotoxic treatment with compounds affecting late autophagic flux, such as chloroquine, renders the 3D-grown cells more susceptible to therapy. Altogether, 3D cultures are a valuable tool to study drug response of tumor cells, as these models more closely mimic tumor (patho-)physiology, including the upregulation of tumor relevant pathways, such as autophagy.

The success rates for investigational cancer drugs in clinical development are poor. The clinical approval rate of compounds for the treatment of solid tumors is 10% or less. 1,2 Improving basic research models is critical for achieving clinical success. Conventionally, preclinical assessment of chemotherapeutic effectiveness starts in two-dimensional (2D) cultures, where cell-cell contacts, cell shape and morphology significantly differ from tumor cells in a physiological setting. All of these features strongly influence cellular growth, behavior and metabolism. 3 Three-dimensional (3D) growth cultures have been introduced for preclinical drug screening to improve the correlation between cell cultures and tumors. 4 Three-dimensional cell growth is associated with a spherical shape, affecting gene and protein expression, survival, proliferation, differentiation, and metabolism. 5,6 Furthermore, 3D-grown tumor cells display enhanced resistance to radioand chemotherapy. 7,8 Additional important characteristics of the physiological setting include the flow of extracellular fluids, leading to shear stress, compound flux and removal of metabolites. Small 3D bioreactor systems mimic these properties by pumping medium at a physiologically representative flow rate. 6,9 Neuroblastoma, a common pediatric tumor of the sympathetic nervous system, is characterized by a wide range of clinical courses. 10 Despite intensification of treatment, high-risk neuroblastoma patients have a very poor prognosis due to chemotherapy resistance. 10,11 We and others have previously reported that macroautophagy (hereafter autophagy) supports chemotherapy resistance in neuroblastoma cells. 12,13 Thus, neuroblastoma is a good model to investigate autophagy-related drug resistance.
Autophagy is an evolutionarily conserved process, involving sequestration of cytoplasmic components within a doublemembrane structure (autophagosome) and subsequent delivery to lysosomes for degradation. 14 Metabolic or therapeutic stress, e.g. DNA-damaging drugs, may induce autophagy, 15 which is regulated by autophagy-related (ATG) genes. 14 Whereas autophagy plays a tumor-suppressive role in early stages of tumorigenesis, it can support growth at later stages, allowing tumor cells to survive with limited oxygen and nutrients, as well as under cytotoxic treatment conditions. 16 Histone deacetylase (HDAC) inhibitors interfere with key tumor-relevant pathways, including proliferation, apoptosis, differentiation and autophagy in several cancer entities in vitro and in vivo. [17][18][19] The human HDAC family comprises four classes: class I, II, subdivided into classes IIa and IIb, III and IV. [20][21][22] Class IIb family members (HDACs 6 and 10) are linked to cellular stress, protein degradation and autophagy. 12,[23][24][25][26] We have previously identified HDAC10 as a mediator of autophagic flux in neuroblastoma and inhibition as well as depletion of HDAC10 sensitized monolayer neuroblastoma cells to cytotoxic drugs. 12 It remains unclear how different cell culture settings influence autophagy, particularly in the context of drug development and therapy resistance. Here we show that 3D culture models are a suitable tool for screening drug responses of tumors. Additionally, these models are useful for mechanistically studying cancer-relevant processes, as they more accurately resemble transcriptional profiles of tumors and recapitulate the regulation of physiologically relevant pathways, including autophagy. Our data suggest that 3D tumor models are advantageous when studying autophagytargeting treatment and resistance mechanisms.

Results
Expression profiles from MYCN-amplified neuroblastoma cell lines and tissues are highly discrepant. Tumor cells tend to lose many hallmarks of cancer over time when cultured in vitro under standard conditions. We explored differences between cultured tumor cells and primary tumor tissues by comparing the gene expression profiles of MYCN amplified neuroblastoma cell lines with tissue samples from a publically available data set (Mixed Neuroblastoma -Versteeg; R2 database). A principal component analysis (PCA) performed with all genes revealed two distinct clusters (Figure 1a, Supplementary Figure S1a), indicating that the gene expression profile after two-dimensional (2D) growth shifts away from tumor tissue. We hypothesized that threedimensional (3D) growth would better recapitulate neuroblastoma physiology. Thus, we seeded neuroblastoma cells in a collagen type I-coated, ridged scaffold model, which yielded reproducible one-size 3D structures. The polymeric scaffolds contain 187 microcavities per chip, promoting 3D growth of multicellular spheroids approximately 200 μm in diameter (Figure 1b). Corroborating previous findings, 27 the collagen substratum in our system was a suitable extracellular matrix protein, affecting neither population doubling time nor response to cytotoxic treatment (Supplementary Table 1). To account for additional physiological factors, we also inserted the chips into a bioreactor system, 9 which simulates the flow of extracellular fluids (Supplementary Figures S1b and c).
A typical histomorphological feature of undifferentiated and aggressive neuroblastoma tumors is its small round blue cell phenotype (Supplementary Figure S1d), which is lost upon growth under classical 2D culture conditions. Two-dimensionally grown cells exhibit flat morphology with much larger nuclei and fewer cell-cell contacts. In contrast, 3D-grown cells more closely resembled small round blue cells, characteristic for undifferentiated neuroblastoma tissue (Figure 1c, Supplementary Figure S1e). Solid tumors are also metabolically heterogeneous, with areas of maximal growth, slow-growing starved regions or even necrotic parts. 28 Similar structures are observable within multicellular spheroid 3D models. 29 We investigated whether growing neuroblastoma cells in the 3D-chip model affected overall cell proliferation and death rate. Three-dimensionally grown cells exhibited an increased population doubling time, but still grew exponentially, with slightly decreased overall viability (Supplementary Table 2). To investigate whether 3D growth affects gene transcription, we analyzed the gene expression patterns of BE(2)-C cells grown in 2D, 3D and bioreactor-3D cultures and compared them to tumor tissue samples from three different neuroblastoma patients (German Neuroblastoma Trial cohort 30 ) resembling the characteristics of BE(2)-C cells (stage 4, amplified MYCN, undifferentiated and 1p deletion). A PCA revealed that 3D-cultured (static and bioreactor) neuroblastoma transcriptomes were closer to the neuroblastoma tissue profile than the 2D samples ( Figure 1d, Supplementary Figure S1f). These results support our hypothesis that 3D growth of cells not only resembles more phenotypic histology, but also reduces the discrepancy between transcriptional programs in cultured cells and primary tumor material.
Resistance to cytotoxic treatment is enhanced in 3Dgrown neuroblastoma cells. To compare drug response, we treated 2D-and 3D-grown neuroblastoma cells with the cytotoxic compounds doxorubicin and vincristine. Threedimensional growth strongly hampered the drug-induced reduction in BE(2)-C cell number compared to 2D growth, with the IC50 values for doxorubicin and vincristine increasing approximately 7-and 5-fold, respectively (Figures 2a  and b). IMR-32 cells were similarly affected (Supplementary Figures S2a and b). Neuroblastoma cells grown threedimensionally in the bioreactor system were completely resistant to vincristine (Figure 2c). To examine long-term effects of drug treatment, we treated 2D-and 3D-cultured BE (2)-C as well as IMR-32 cells for six days with doxorubicin and confirmed prolonged resistance in tumor cells cultured under 3D conditions (Figure 2d). Using the 3D-BE(2)-C model we tested other clinically-relevant drugs, such as cisplatin, etoposide and the MET/ALK-inhibitor crizotinib. In all cases, 3D-cultured cells exhibited a weaker response than 2D-grown cells (Figure 2e). Moreover, six out of a panel of seven neuroblastoma cell lines had more cells survive doxorubicin treatment under 3D compared with 2D conditions ( Figure 2f). Only NB-1 cells were highly sensitive to doxorubicin, irrespective of culture conditions. Both MYCN amplified and MYCN-depleted (shMYCN) IMR5/75 cells respond less to doxorubicin when cultured in 3D. However, the difference in drug sensitivity between both conditions was greater among cells expressing MYCN (2-fold) compared to MYCN-depleted cells (1.4-fold), suggesting MYCN involvement in mediating resistance. Our findings support the notion that the lower sensitivity to cytotoxic drugs observed in 3D cultures more accurately recapitulates chemotherapy resistance than 2D-grown cells and that more representative culture conditions could avoid a misleading prediction of drug sensitivity.
Drug resistance of three-dimensionally grown tumor cells translates into a suppressed apoptotic response. As 3D growth yields more treatment-surviving neuroblastoma cells, we investigated whether programmed cell death was affected. The number of dead cells (BE(2)-C, IMR-32) in the subG1-area of the cell cycle profile following doxorubicin or vincristine treatment clearly differed between 2D and 3D conditions ( Figure 3a). Correspondingly, significantly reduced activation of effector caspases in response to drug treatment under 3D culture conditions was observed in both cell lines (Figure 3b). The difference in caspase activity in 3D versus 2D-grown cells was also reflected by different levels of PARP cleavage, exemplarily shown for IMR-32 cells (Figure 3c). Following treatment, the ratio of cleaved to full-length PARP increased to a significantly greater extent in 2D-grown compared with 3D-grown cells (8-fold versus 2-fold and 10fold versus 5-fold) for doxorubicin and vincristine, respectively  (c) One day after seeding, chips containing 3D-cultured BE(2)-C cells were transferred into the bioreactor system (3D-BR) and medium or medium containing vincristine (10 ng/ml) was pumped through the system for 48 h. (d) One day after seeding, 2D-and 3D-cultured BE(2)-C or IMR-32 cells were treated for 6 d with doxorubicin, where indicated. Doxorubicin concentrations: BE(2)-C: 0.5 μg/ml and IMR-32: 0.01 μg/ml. (e) BE(2)-C cells grown in 2D or 3D were treated 24 h post-seeding with doxorubicin (0.5 μg/ml), vincristine (0.01 μg/ml), cisplatin (15 μg/ml), etoposide (7.5 μg/ml) or crizotinib (0.8 μM). (a-f) Number of viable cells was assessed by automated cell counting and trypan blue exclusion and was normalized to untreated control of each condition. Significant differences between groups were tested using an unpaired, two-tailed t-test. n.s., not significant; * Po0.05; **Po0.01; ***Po0.001 3D models mimic pathophysiological autophagic flux C Bingel et al ( Figure 3d). In summary, our data suggest a decreased apoptotic response in 3D-grown tumor cells.
Three-dimensional growth affects expression of autophagy-related genes in neuroblastoma cells. Growth of tumor cells under spheroidal 3D conditions has been described to affect drug resistance via cellular pathways including proliferation, hypoxia and autophagy. [31][32][33] As we have previously shown the importance of the autophagic degradation pathway in neuroblastoma cell cultures, we focused on genes associated with the Gene Ontology (GO) term 'autophagy' (GO:006914) and used the globaltest R package 34 to check for differential regulation between 2D and 3D cultures. Differentially regulated genes were ranked according to the component P-value (Supplementary Table  3), and the ten genes with the lowest P-values (TOP10) were  (Table 1). We additionally included HDAC10 in the validation assays as this HDAC is also involved in the regulation of autophagic flux in neuroblastoma cells 12 and is closely related to the TOP10 hit HDAC6. Nine of the TOP10 genes showed differences on RNA level concordant with the results of the gene expression analysis. Additionally, we re-evaluated gene expression data from the MYCN amplified neuroblastoma cell lines and tumors contained in the 'Mixed Neuroblastoma' data set (Versteeg; R2 database), performing a PCA using all genes associated with the GO term 'macroautophagy'. Tissue samples and cell lines were clearly separated by the PCA (Figure 4a, Supplementary Figure S3a). Using the same data set, we performed a clustering analysis with the hit list of genes obtained from our own microarray analysis. It revealed an almost complete separation of cell lines from tumor samples ( Figure 4b). A clustering analysis of our own microarray data using autophagy genes with a P-value less than or equal to 0.005 revealed two main clusters: 2D versus 3D/tumor tissue ( Figure 4c). Of the nine PCR-validated genes, eight were significantly upregulated in tumor samples compared to 2D-cultured cell lines not only in our own microarray data, but also in the publically available R2 data set (Table 1, Figure 4d). A direct comparison of the TOP10 (plus HDAC10) gene expression among 2D-and 3D-grown cells with tumor tissue samples using our SDHA-normalized real-time PCR data as well as the SDHA-normalized Versteeg microarray data revealed that the gene expression of 3D-grown cells approaches that of tissue material ( Figure 4e). To determine whether protein levels reflect our RNA expression profiles, we performed western blot analyses of ULK1, HDAC6, MAP1LC3A, ATG16L2 and HDAC10. Additional validation experiments were performed for HDAC10, as this candidate was not initially identified within the TOP10 list (Supplementary Figures S3b and c). Consistent with the upregulation of their transcripts, greater quantities of all five proteins were found in treatment-resistant 3D-compared to 2D-cultured BE(2)-C cells (Figure 4f).
Three-dimensional cell culture increases autophagic flux. It has been proposed that tumor cells adopt autophagy as a survival mechanism to cope with metabolic and cytotoxic stress. 15,35,36 Similarly, spheroid cultures might use this survival strategy, especially in the inner starved zones. 33 We examined autophagic flux in our standardized 3D-chip culture using stably transfected BE(2)-C cells expressing the tandem construct mCherry-EGFP-LC3B. Appearance of red dots indicates activated autophagic flux, characterized by successful fusion of autophagosomes with lysosomes, forming low-pH autolysosomes where acid-sensitive green fluorescence is lost. Three-dimensionally grown BE(2)-C cells clearly showed characteristic LC3 punctate staining with red mCherry, but not green EGFP fluorescence (Figure 5a). In contrast, EGFP fluorescence was detected upon treating 3D-grown cells with the flux-inhibiting lysosomotropic agent chloroquine (CQ), resulting in yellow punctae in the merged image ( Figure 5a). Inhibition of autophagic flux by CQ also increased the amount of propidium iodide-positive dead cells inside the 3D structure (Figure 5b), further hinting at a pro-survival role of autophagy in spheroids. As an indicator for metabolic stress-induced autophagy and autophagosome formation, we analyzed mTOR activity and determined WIPI1 mRNA expression levels 37 Figure S3g). Finally, 3D-grown BE(2)-C cells exhibited depletion of the autophagy substrate sequestosome 1 (p62/SQSTM1), which, together with SQSTM1 enrichment upon autophagy inhibition with CQ, implies more active autophagosomes (Figure 5g).
These results indicate that 3D culture conditions enhance preexisting autophagic flux in autophagy-competent neuroblastoma cells.
Transcriptional regulation of autophagy. As elevated expression of autophagy genes points towards transcriptional upregulation, we investigated six transcription factors known to regulate the expression of autophagy-related genes. [39][40][41][42] All transcription factors, except NRF2 (NFE2L2), showed elevated expression in tumors compared to 2D cell lines. Our 3D cultures expressed four of the transcription factors at a level between that of 2D and tumor samples, showing that 3D growth alters expression in the same direction seen in tumors ( Figure 6a). Additionally, FOXO3 expression significantly correlated with the expression of seven TOP10 genes plus HDAC10 (Supplementary Table 4). Three-dimensionally cultured BE(2)-C, but not NB-1 cells, displayed increased protein levels of FOXO3 (Figure 6b). Depletion of FOXO3 in 3D-cultured BE(2)-C cells altered expression of seven genes plus HDAC10 as well as GABARAPL1, a well-known FOXO3 target gene 43 (Figure 6c; exemplarily shown on protein level for HDAC6, ATG16L2 and HDAC10; Figure 6d). We employed an siRNA-mediated knock-down approach to further validate the role of autophagy regulators and HDACs in mediating therapeutic resistance of 3D cultures. Threedimensionally grown BE(2)-C cells depleted of BECN1, ATG5, ATG7, HDAC6, HDAC10 and FOXO3 exhibited impaired autophagic flux, demonstrated by SQSTM1 accumulation ( Figure 6e) and increased chemotherapeutic sensitivity close to the level seen in 2D cultures, potentiating the effects of vincristine (Figures 6f and g), though HDAC6 depletion achieved comparatively weak effects. The same approach also sensitized 2D-grown cells, however the combination exhibited no potentiation effect (except HDAC10; Supplementary Figure S4a). As 2D cell viability was not decreased by siRNA knockdown alone (except HDAC10), the more pronounced effects on viable cell number 6 d after transfection can be explained by the higher proliferation rate of 2D-grown cells (Supplementary Tables 2 and 5). Thus, transcriptional activation of autophagy supports enhanced expression of autophagy-related genes in 3D models and cancer tissues. Furthermore, the data suggest that interference with autophagy is a promising approach to break therapy resistance in neuroblastoma cells.
Pharmacological inhibition of autophagic flux sensitizes 3D-grown neuroblastoma cells to cytotoxic treatment. Based on our findings that 3D-grown cells are sensitized to cytotoxic treatment when genetic interference disrupts autophagic flux, we explored whether a pharmacological block would yield similar results. To pharmacologically interfere with late-stage autophagy, we used chloroquine (CQ) and bufexamac, a class IIb HDAC inhibitor (Supplementary Figure S4b). Bufexamac impaired autophagic flux, demonstrated by an increased quantity of yellow mCherry-EGFP-LC3B dots, elevated SQSTM1 and LAMP-2 levels (Supplementary Figures S4c and d).
The IC50 values for vincristine (100 ng/ml) and doxorubicin (540 ng/ml) were decreased 15-to 100-fold upon combination with CQ or bufexamac (Figures 7e and f). These results demonstrate that co-treatment is able to shift effective concentrations into a range that is clinically achievable in patient plasma (doxorubicin: approximately 60 ng/ml; vincristine: approximately 7 ng/ml). 44,45 Co-treatment of 3D cells   transferred into the bioreactor system with bufexamac also significantly decreased the quantity of viable cells (Figure 7g) and increased cell death (Figure 7h). Decreased intracellular doxorubicin, due to the 3D culture itself, the thereby increased autophagic flux, or a combination of both factors, could be one possible mechanism involved in resistance to cytotoxic treatment. Hence, we exploited the inherent fluorescence of doxorubicin to quantify intracellular doxorubicin by FACS in 2D and 3D cultures. Doxorubicin fluorescence was significantly decreased in 3D cultures (Figure 7i), while co-treatment with CQ and bufexamac reversed this effect (Figure 7j), yielding comparable levels achievable in 2D cultures (Figure 7i). Likewise, intracellular doxorubicin concentration was enriched when cells were depleted of FOXO3, ATG5 or HDAC10 (but not HDAC6) (Supplementary Figure S4h). These results demonstrate the substantial impact of autophagy on drug response at two levels: i. cell survival and ii. decreased intracellular drug concentration (as shown for doxorubicin), suggesting that pharmacological inhibition of autophagic flux effectively sensitizes 3D-grown cells to cytotoxic treatment.

Discussion
Drug screens in 2D systems have successfully identified compounds that were subsequently translated into clinical use. Notable examples are inhibitors targeting tyrosine kinases, such as ALK (crizotinib), Bcr-Abl (imatinib) and HER2 (trastuzumab), which are aberrantly active due to genetic alterations. [46][47][48][49] Nevertheless, there is a poor correlation between effectiveness of emerging anti-cancer agents in cell culture and ultimate treatment success rates in patients, which are low and need improvement. 50 Culture conditions impact the regulation of cell fate and responsiveness to external stimuli. Two-dimensional cultures are favored due to high-throughput testing capability, but may lose phenotypic and functional characteristics, 5 which can be overcome by culturing cells under more physiological 3D conditions. A remaining practical challenge is large-scale testing of compound libraries, under reproducible 3D conditions. The chipbased system used here yields 3D cultures with a well-defined and consistent geometry, stable and comparable culture conditions, and compatibility with a bioreactor system, allowing for an even more representative environment.
Several DNA microarray studies demonstrated significant, tumor-relevant molecular changes induced by 3D growth. 3 Our gene expression studies corroborate these results and show that 3D growth affects the expression of ATG genes and autophagy-controlling transcription factors, such as TFEB and FOXO3. FOXO transcription factors are regulated by a wide range of external stimuli, including nutrients and oxidative stress, 51 and can support stress resistance through induction of antioxidant as well as ATG proteins. 52 Our results indicate that autophagy and the expression of ATG genes may be induced as an adaptive survival strategy due to stress conditions inside the 3D structure, which is also reflected in strongly reduced mTOR activity. Several publications indicate that there is cross-talk between the activity of the mTOR complexes and the activity of FOXO3a via AKT, which can be activated by mTORC2. 53 AKT, in turn, negatively regulates FOXO3a by phosphorylation, preventing its nuclear translocation. 54 Thus, it is conceivable that reduced activity of the mTORC/AKT axis supports FOXO3 transcriptional activity in our model system. Overall, autophagy induction protects the cells, as co-treatment with autophagic flux blockers increases sensitivity of 3D cultures to chemotherapy. This is in line with several studies demonstrating that inhibition of autophagy sensitizes a diversity of tumor entities towards chemotherapeutic drugs in vitro and in vivo. [55][56][57][58] Accordingly, several early phase clinical trials are currently under way to evaluate autophagy inhibition using (hydroxy-)chloroquine in combination with chemotherapy or targeted agents. 58,59 Unraveling the exact mechanism as to how autophagy is protecting which type of tumor cell is the subject of current research, which also indicates a particular role for lysosomes as mediators of drug resistance. 60 Although technically different 3D culture systems are similar in terms of more tissue-like cell morphology, 3 the role of autophagy might depend on tumor entity, tumor stage and 3D culture context. For example, in contrast to our 3D model, the 3D laminin-rich ECM model resulted in increased sensitivity to doxorubicin and compromised autophagy. 61 Accurate determination of susceptibility to treatment is crucial for preclinical assessment of a compound's efficacy. In our setting, 2D-grown BE(2)-C cells are sensitive to doxorubicin and vincristine in clinically achievable concentrations, while the 3D model predicts resistance. 44,45 Consistent with the latter, the BE(2)-C cell line was isolated from a relapsed neuroblastoma patient following multi-agent chemotherapy, including doxorubicin. 62 Cytotoxic drugs combined with autophagy inhibitors shifted the effective chemotherapy concentrations back to clinically achievable levels. We observed, at least in the case of doxorubicin, that sensitization is accompanied by an increased uptake (or decreased efflux) of the cytotoxic drug into the 3D structure and that inhibition of autophagy plays an important role in this effect. However, we cannot exclude that additional, potentially drug-specific mechanisms account for the beneficial effects of combination treatment. Overall, 3D models decrease the discrepancy between cell culture and tumor tissue and are useful for the identification of points of vulnerability for cancer treatment. Our work with 3D cultures underscores two critical aspects, which are (at least partly) linked together: (i) the upregulation of cell-protective autophagy and (ii) decreased intracellular drug concentration.
Altogether, our data favor 3D cell cultures as tissue-like models, recapitulating solid tumors with respect to cell morphology, gene expression and cellular functions. The inclusion of 3D models is highly advantageous, particularly for mechanistic studies and investigations of drugs targeting the autophagic pathway.
Materials and Methods 3D and 2D cell cultures and patient samples. Human neuroblastoma cell lines BE(2)-C (ECACC), IMR-32 (DSMZ) and the tetracycline-inducible shMYCN IMR5/75 63 (generously provided by the laboratory of F. Westermann) were cultured under standard conditions (DMEM with L-glutamine and 4.5 g/l glucose containing 10% FCS (Sigma) and 1% non-essential amino acids (NEAA; Invitrogen, Darmstadt, Germany)). Human neuroblastoma cell lines SK-N-BE, NB-1 (#RCB1953, RIKEN cell bank, Japan) and Kelly (DSMZ) were cultured in RPMI with L-glutamine containing 10% FCS (Sigma, Taufkirchen, Germany) and 1% nonessential amino acids (NEAA; Invitrogen). Wildtype and HDAC10-knockout HAP1 cells (Horizon Discovery Group, Cambridge, UK) were cultured in Iscove's Modified Dulbecco's Medium (IMDM) with 10% FCS (Sigma). All cell lines were regularly checked for contamination (Multiplexion, Heidelberg, Germany) and verified using DNA fingerprinting authentication by the DSMZ, Germany. The 3D cultures were prepared on ridged scaffolds (MatriGrid structures), with a total active area of 5 × 5 mm 2 , as described previously. 9,64 Briefly, the microcavity-containing polymeric scaffolds, each with a diameter of 300 μm and a depth of 207 μm, 64 were coated with collagen type I (10 μg/cm 2 , Sigma, #C3867) as substratum, allowing cells to adhere to the scaffold. After 2 h of drying, cell solution was added to the prepared chip. For the static 3D model, cells were cultured on 3D scaffolds for 3-7 days, as indicated in the respective figure legend. For the bioreactor system, cells were cultured on 3D scaffolds for 2 days, then the cell-containing chips were inserted into the bioreactor cube 9,64 and cultured for another 2-5 days, as indicated in the respective figure legend. Complete medium containing test compound or solvent control was pumped through the bioreactor system (Ismatec, Wertheim, Germany) with a flow rate of 25-35 μl/min. If necessary, the spheroids were recovered from the cavities by trypsinization (3-4 min at 37°C) and one to maximum two very careful washing steps with fresh medium.
Immunohistochemistry and nuclear staining: Immunohistochemistry and nuclear staining were performed on BE(2)-C cells grown under 2D or 3D conditions, and a neuroblastoma tissue sample. Briefly, cells were fixed with ice-cold methanol for 20 min and stained with hematoxylin and eosin (HE) or diamidino-2-phenylindole (DAPI). Before staining, spheroids were freed from the polymeric chip by trypsinization (4 min at 37°C), followed by careful transfer onto glass slides after addition of medium. The neuroblastoma tissue sample (#14T1OL) was provided by the tissue bank of the National Center for Tumor Diseases (NCT, Heidelberg, Germany) in accordance with the regulations of the tissue bank and with the approval of the ethics committee of Heidelberg University. Sections were deparaffinized, incubated with Tris-buffer saline containing 0.5% Triton X-100 (TBST) for 20 min at room temperature. Following staining with HE or Hoechst dye (30 min at room temperature), the slides were washed and mounted. Fluorescent images were acquired on a Zeiss LSM 710 confocal microscope using the 40 × objective.
Quantification of CYTO-ID via flow cytometry: Three days after siRNA transfection, BE(2)-C cells were seeded in 3D scaffolds and grown for an additional 72 h. Where indicated, cells were treated with 5 μM CQ the night before staining with CYTO-ID (Enzo Life Sciences). For staining (6 d post-siRNA transfection), chips were washed once with medium w/o phenol red, transferred to a fresh 6-well plate and covered with 500 μl CYTO-ID staining solution (1 × assay buffer, 5% FCS, 1 μl CYTO-ID) for 30 min at 37°C. Cells were washed with medium w/o phenol red, trypsinized (4 min. 37°C), centrifuged and washed again in cold medium w/o phenol-red. CYTO-ID fluorescence was quantified on a BD FACSCanto II platform using an Alexa488 filter setting Quantification of doxorubicin uptake: BE(2)-C cells were seeded in 3D scaffolds 24 h prior to treatment. Spheroids were treated for 48 h with solvent control (DMSO), doxorubicin (500 ng/ml), combinations of doxorubicin with bufexamac (30 μM), chloroquine (5 μM) or both. For the knockdown approach, BE(2)-C cells were seeded in 3D scaffolds three days after transfection with ATG5, FOXO3a, HDAC6, HDAC10 or control siRNAs, grown for an additional 48 h and treated with 500 ng/ml doxorubicin 24 h before staining. For FACS analysis, chips were washed with RPMI w/o phenol red (10% FCS), transferred to a fresh 6-well plate, trypsinized (4 min, 37°C), centrifuged and washed again in cold medium w/o phenol-red (10% FCS). For the analysis of 2D conditions, BE(2)-C cells were seeded in monolayers in six-well plates 24 h prior to treatment. Thereafter, the same protocol as for the 3D scaffolds was followed. Doxorubicin fluorescence was quantified on a BD FACSCanto II platform using a Phycoerythrin (PE) filter setting.
Cell counting, cell viability and cell death assays. For cell death detection, adherent as well as floating cells were collected.
Viability assay: Trypsinized cells were pooled with corresponding supernatant, centrifuged and resuspended in 1 ml cell culture media. Cell viability was measured by automated trypan blue staining with the Vi-Cell XR Cell Viability Analyzer from Beckman Coulter (Krefeld, Germany).