Between 10% and 40% of patients with acute myeloid leukaemia (AML) are resistant to induction therapy with cytarabine (AraC) and anthracycline.1 While patient outcome could be improved by selecting the right induction therapy,2 reliable prediction of treatment efficacy based on cytogenetics, molecular profile or gene expression signatures is still a challenge. The impact of the bone marrow microenvironment (BMM) is increasingly recognised as a reason for this. The aim of the present study was to develop a functional drug test that incorporates the impact of the BMM to predict the clinical response to induction chemotherapy [AraC + daunorubicin (Dnr)]. While it is routine clinical practice to isolate mononuclear cells from marrow and there are sophisticated methods to culture AML blasts,3-6 the usability of these systems to predict the clinical response of a patient is not known. We have established a co-culture model where primary AML blasts were cultured in direct contact on a layer of bone marrow stromal cells [BMSCs; healthy donor-derived BMSCs immortalised with human papillomavirus type 16 [HPV-16] E6/E7 [HS-5], or with human telomerase [immortalised mesenchymal stromal cells, iMSC]), providing extracellular matrix (ECM), cell–cell contact and paracrine trophic signals, replicating multiple features of the BMM. Immortalised BMSCs retained key characteristics of primary BMSCs, as they deposited an ECM with BM-typical ECM components (fibronectin, collagen I, IV and VI) and established a proteo-cellular network comparable to that of primary BMSCs (Figure S1), supported AML blast viability (Figure S2A), secreted key BM-characteristic cytokines (Figure S2B,C, Table S1) and could closely replicate the effect of the patients’ own primary BMSCs in mediating drug resistance against BCL-2 homology domain 3 (BH3)-mimetics (ABT-737, ABT-199) and chemotherapeutics (AraC, Dnr) (Fig 1A, Figure S3A,B). BMCS-mediated drug resistance was associated with a profound change in kinase activation in the AML cells co-cultured with BMSCs, where out of the 109 kinases tested, 43 kinases were activated and eight inhibited (Fig 1B, Figure S4A). Ingenuity signal transduction network analysis (Ingenuity Pathway Analysis; Qiagen, Germantown, MD, USA) showed that this kinase activation drove several converging, BM-typical pro-survival signalling pathways, demonstrating the extensive effect BMSCs have on AML cells equipping them against cytotoxic drugs (Fig 1B,C, Figure S4B). To determine how faithfully the BMSC-based co-culture can replicate BMM-mediated drug resistance and the consequent clinical response, samples from patients receiving 3 + 10 AraC + Dnr therapy were screened (patient characteristics summarised in Table S2). AML blasts were plated either as a single culture or kept on an iMSC layer for 24 h followed by treatment with AraC + Dnr for 48 h; and induction of cell death in the bulk AML population and the frequency of the CD34+/CD38−, putative leukaemic stem cells pool in the surviving population were quantified. By generating receiver operating characteristic (ROC) curves and calculating the area under the ROC curve (AUROC) the accuracy of predicting the clinical response (complete remission or refractory disease) of the single culture was compared to that of the co-culture. The co-culture showed a much stronger accuracy (AUROC of 0·919 for co-culture vs. 0·781 for single culture; Fig 2A). Also, using Pearson’s correlation, we found there was only a moderate positive correlation in the percentage of surviving blasts between single- and co-cultures (r = 0·615, P = 0·0039; Figure S5A), with a noticeably weakening correlation with increasing drug resistance, confirming the role of BMSCs in modulating drug sensitivity. To better capture the interpatient variability in the kinetics of drug sensitivity, drug response was quantified by calculating drug-efficacy scores based on the area under the curve (AUC) from the cell death graphs. Using the data from 20 patients, the samples were segregated into drug-resistant and drug-responsive patients by applying a cut-off of 35% (Fig 2B), where the score predicted the clinical response with very high accuracy (AUC = 0·94, P < 0·0001, Fig 2B,C, Table S2). The drug-efficacy score cut-off value was validated using an additional 12 samples, where the test correctly predicted the clinical response in all 12 cases, underscoring its predictive potential and high accuracy (Fig 2D, Table S2). The clinical follow-up of the patients showed that 52·1% (12/23) of the complete remission (CR) patients relapsed. Relapse has been linked to the drug-resistant leukaemia-initiating cell population,7, 8 broadly identified as the CD34+/CD38− population. We found that the CD34+/CD38− population had a much higher resistance to AraC + Dnr than the bulk blast population (Fig 2E); however, there was no correlation between either pre-treatment CD34+/CD38− cell frequency or their accumulation rate after treatment with relapse (AUROC = 0·667, P = 0·175, Figure S5B). Layered co-cultures to model normal haematopoiesis and to culture AML blasts are widely used since Dexter et al.9 first described it in 1977. It is also well established that BMSCs can provide survival signals to AML cells and protect them from cytotoxic drugs.10, 11 However, to our knowledge, this is the first report to show that an ex vivo co-culture can predict the clinical response. The current best models to predict induction treatment-failure incorporate aspects of cytogenetics, mutational profile, physical status and gene expression signatures.2, 12 The accuracy of these models is between AUROC = 0·68–0·78,13-15 which is well-below that of a robust clinical diagnostic tool (AUROC >0·9). The genetic, epigenetic and biological heterogeneity of AML indicates that a functional drug testing may predict treatment response with higher efficacy. Our present study highlights that a relatively simple, layered co-culture consisting of healthy donor-derived BMSCs and AML blasts can replicate the in vivo drug response, and thus can be used as a theranostic test to predict a patient’s response to AraC + Dnr induction therapy. Ability to predict drug response within days would allow clinicians to consider alternative treatment options for refractory patients, thus preventing exposure of the patient to a treatment that is not effective and potentially weakening the patient so that they become unfit for alternative chemotherapy. The present study focussed on AraC + Dnr treatment, as most patients receive this therapy. However, the results warrant further studies to assess its ability to predict the efficacy of other drugs and drug combinations, such as Fms-like tyrosine kinase-3 (FLT3) inhibitors, or BH3-mimetics. Method descriptions are in the Data S1 section. The project was supported by funding to Eva Szegezdi by the Science Foundation Ireland (SFI) and the Irish Cancer Society (BCNI,14/ICS/B3042), SFI to Eva Szegezdi (12/TIDA/B2388), The College of Science Scholarship, NUI Galway and Thomas Crawford Hayes Research Award to Sukhraj Pal S. Dhami. We are thankful to Diana Gaspar for helping with the ECM studies. The authors would also like to acknowledge the BCBI staff of University Hospital Galway, Galway (Tatiana Cichocka, Amjad Hayat, Margaret Murray, Sorcha NiLoingsigh, Michael O’Dwyer); Beaumont Hospital, Dublin (Tara Kenny, Lorna Mulvihill) and Cork University Hospital (Oonagh Gilligan, Vitaliy Mykitiv, Derville O’Shea, Juliet Barry), Cork, Ireland for providing patient samples. John Quinn has honoraria from Janssen. All the other authors declare no conflict of interest. Eva Szegezdi designed the study. Sukhraj Pal S. Dhami, Andrea Tirincsi, Denis Baev performed the wet-laboratory experiments. Dimitrios Zeugolis contributed to the ECM studies. Eva Szegezdi and Sukhraj Pal S. Dhami carried out the statistical analyses. Janusz Krawczyk, Mary R. Cahill and John Quinn provided clinical information. Sukhraj Pal S. Dhami and Eva Szegezdi wrote the manuscript. Sukhraj Pal S. Dhami, Andrea Tirincsi, Denis Baev, Janusz Krawczyk, Mary R. Cahill, John Quinn, Dimitrios Zeugolis and Eva Szegezdi contributed to data interpretation, edited and approved the manuscript. Fig S1. Deposition of extracellular matrix proteins is comparable between immortalised and primary bone marrow mesenchymal stromal cells. Fig S2. Bone marrow stromal cells support AML viability and secrete bone marrow-specific cytokines and chemokines. Fig S3. Comparison of drug resistance mediated by different BMSC types. Fig S4. Phospho-proteome analysis of FLT3-ITD AML cells in contact with BMSC Fig S5. Correlation and receiving operating characteristics (ROC) curve analysis of drug response determined in single AML blast culture versus co-culture and CD34+/CD38− population and clinical response. Fig S6. Flow cytometry gating strategy. Data S1. Supplementary materials and methods. Table S1. Cytokines/chemokines commonly secreted by pBMSC, HS-5 cells and iMSCs Table S2. Clinical data of patient samples, treatment and response. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Introduction- The mainstream therapy for AML is cytarabine (AraC) and anthracycline-based intensive chemotherapy. Between 10-40% of patients however have a refractory disease. It is becoming increasingly accepted that the risk for refractory disease could be reduced by finding better ways of selecting the induction therapy. While multiple in vitro and in vivo experimental models exist for pre-clinical drug efficacy testing, they are either labour intensive, too time consuming or generate variable results, making them unsuitable for clinical application. Furthermore, very few, if any of these methods have been shown to replicate the clinical response of patients. The aim of this study was to develop a functional drug testing system that can rapidly and faithfully predict the clinical response of the patient and therefore be utilized to help the identification of patients suitable for AraC+daunorubicin (DnR) therapy.

Cancer immune surveillance is essential for the inhibition of carcinogenesis. Malignantly transformed cells can be recognized by both the innate and adaptive immune systems through different mechanisms. Immune effector cells induce extrinsic cell death in the identified tumor cells by expressing death ligand cytokines of the tumor necrosis factor ligand family. However, some tumor cells can escape immune elimination and progress. Acquisition of resistance to the death ligand-induced apoptotic pathway can be obtained through cleavage of effector cell expressed death ligands into a poorly active form, mutations or silencing of the death receptors, or overexpression of decoy receptors and pro-survival proteins. Although the immune system is highly effective in the elimination of malignantly transformed cells, abnormal/dysfunctional death ligand signaling curbs its cytotoxicity. Moreover, DRs can also transmit pro-survival and pro-migratory signals. Consequently, dysfunctional death receptor-mediated apoptosis/necroptosis signaling does not only give a passive resistance against cell death but actively drives tumor cell motility, invasion, and contributes to consequent metastasis. This dual contribution of the death receptor signaling in both the early, elimination phase, and then in the late, escape phase of the tumor immunoediting process is discussed in this review. Death receptor agonists still hold potential for cancer therapy since they can execute the tumor-eliminating immune effector function even in the absence of activation of the immune system against the tumor. The opportunities and challenges of developing death receptor agonists into effective cancer therapeutics are also discussed.