JAK-STAT inhibitor as a potential therapeutic opportunity in AML patients resistant to cytarabine and epigenetic therapy

ABSTRACT

The prognosis of AML is generally poor, with 5-year survival rate of 25%. There has been substantial progress in identification of new therapeutic targets, along with approval of at least three targeted therapies for AML in recent years. Nevertheless, treatment has largely remained unchanged over couple of decades, with ~40% patients not achieving remission. AML is a highly heterogenous disease and there is a need for a preclinical platform to understand the heterogeneity and tumor microenvironment that can guide therapy selection. In this study, we employed an ex vivo tumor explant model to study tumor microenvironment and to select a treatment course for AML patients. Our data reveal dysregulation of DNA methyltransferase (DNMT) and histone deacetylase (HDAC) in a subset of AML patients. Based on this observation, epigenetic modulators azacitidine and panobinostat alone and in combination, were evaluated as treatment regimens in cytarabine refractory tumors. More than 50% of the treated samples showed response to the combination therapy. In order to explore alternate treatment modalities for tumors refractory to these epigenetic modulators, TCGA data analysis was done which revealed increased expression and hypomethylation of IFNGR1/2, suggesting activation of JAK/STAT pathway in AML. This was further interrogated ex vivo, with p-STAT3 expression in patients’ samples. Fedratinib, a JAK/STAT inhibitor was evaluated and 78% tumor efficacy response was achieved. Taken together, our data indicate that ex vivo platform derived from patient samples is capable in guiding optimal therapy selection for various classes of drugs including identification of novel targeted therapies.

KEYWORDS:

• AML
• ex vivo explant
• fedratinib
• epigenetic modulators

Introduction

The genetic plasticity of cancer cells results in the inherent ability of tumor to constantly evolve and adapt under pressure. This leads to a highly heterogenous and dynamic landscape of diverse clonal subpopulations. Further, this landscape shifts with disease progression and/or during a treatment. This spatio-temporal heterogeneity of cancer often leads to failure of predictive and prognostic biomarkers. Hence, a targeted therapy that is being directed toward an actionable mutation often falls short in efficacy. This is predominantly due to molecular and phenotypic heterogeneity among cancer cells, particularly if mutation(s) is/are present in a small subclone¹. Further, a cytotoxic chemotherapy itself can induce mutations in cells or can bias the selection toward a drug-resistant subclone by eliminating sensitive clones.² This aspect of cancer cells, represents a significant challenge in therapy selection.

Acute myeloid leukemia (AML) has been established as a highly heterogenous hematological malignancy. Cytogenetic analysis in adults with de novo AML has shown that about 52% of patients have nonrandom chromosomal abnormalities (e.g., deletions, translocations). Some of these abnormalities have been associated with longer remission and survival, while some have been associated with poor therapy outcome and shorter survival.³ Even within the cytogenetically normal AML subgroup, a substantial genetic and epigenetic diversity exists with complex interplays,^4,5 and not all patients within this subgroup show a similar response to treatment. Thus, the risk pattern in AML and the clinical outcomes are determined based on cytogenetic abnormalities, mutations leading to over- or underexpression of one or many proteins, and/or epigenetic changes. This is further complicated by the fact that the disease evolves over time and multiple competing clones coexist at any given point of time.

Several new targeted agents have been recently approved for clinical use. In 2017, FDA approved 2 targeted therapies for AML, midostaurin (Rydapt®) for patients with mutation in the FLT3 gene⁶ and enasidenib (Idhifa®) for patients with IDH2 mutation.⁷ In 2018, ivosidenib (Tibsovo®) was approved by FDA for the treatment of patients with IDH1 mutation. Despite the progress in understanding of AML biology and identification of new therapy targets, treatment of AML has largely remained unchanged over the last couple of decades.⁸ This is probably because clinical implications of the biological intricacies of AML are still largely unclear. Cytarabine has historically been one of the major drugs used in the treatment of AML. It is used both as remission-induction therapy and postremission consolidation therapy.^9,10 However, with the current treatment protocols, about 20%–40% of patients still do not achieve remission with standard chemotherapy, and 50%–70% of first complete remission (CR) patients often relapse over a time period of 3 years. Especially in elderly patients over 60 years, despite achievement of reasonable complete response rates (CR rates 35%–55%), intensive chemotherapy is associated with high incidence of 4-week mortality and with 3- to 5-year survival rates of  < 10%.^11,12 Further, prognosis for patients with AML refractory to first-line treatment or in relapsed cases is generally poor.¹⁰ Recent studies have raised the possibility that chemotherapy itself contributes to relapse of AML by generating new mutations in the founding clone or one of the subclones, which then undergoes selection and clonal expansion.¹³ Therefore, rational use of treatment options is critical, and it is important that physicians make informed decision toward the course of treatment. Attempts have been made to develop risk stratification panels based on cytogenetics and mutational patterns that broadly classifies AML patients into various risk groups, with an aim to predict therapy outcomes.^12,14,15 However, there is still a pressing need for a platform that will provide near real-time information on treatment outcome for individual patients in a personalized setting.

In this study, we employed an ex-vivo tumor explant model (CANscript™) to select a treatment course for AML patients. CANscript™ platform offers a comprehensive system that mimics patient tumor microenvironment by preserving heterogeneity, tumor/stroma ratios and integrity of pathway networks in an individualized ex vivo functional system. The platform is capable of guiding optimal treatment selection for various classes of agents including epigenetic modulators and at early stage of clinical evaluation in many indications including AML.¹⁶

Results:

Ex vivo, functional platform predicts response to cytarabine therapy

Eight patient samples, which were confirmed cases of AML were recruited in the first round of this study. Tumor samples were analyzed on a personalized tumor explant culture system (CANscript^TM) to predict response to cytarabine, which is one of the standard of care options in treatment of AML. Treatment efficacy was evaluated by assessing changes in tumor cell viability (CCK8), morphology (H&E), proliferation (Ki-67) and induction of apoptosis (cleaved caspase-3) after treating tumor explants with cytarabine. Inputs from these parameters were integrated into a mathematical algorithm (SVMpAUC) to generate a single score (S-Score).¹⁶ Our study demonstrated that S-Score can segregate patients into populations of responders (S-score >19.1) and nonresponders (S-score ≤ 19.1) to a given therapy with high positive predictive value. Analysis of S-scores revealed that of the eight samples treated with cytarabine, three were predicted responders and 5 were predicted nonresponders to the drug (Table 1). Amongst these eight patients, two patients (P4 and P8) were treated with low dose cytarabine monotherapy (30 mg/m²) in clinic and there was no remission of the disease showing concordance with CANscript prediction (Figure 1(a), Supplementary Table S1). The other six samples were treated in clinic with a different treatment regimen than what was tested using our ex vivo platform.

Table 1. Cytarabine response pattern in first round of recruited patient samples

Patient ID	Cytarabine
P1	Responder
P2	Responder
P3	Responder
P4	Nonresponder
P5	Nonresponder
P6	Nonresponder
P7	Nonresponder
P8	Nonresponder

S-Score > 19.1 is classified as responders whereas S-Score ≤ 19.1 is considered nonresponders.

Figure 1. (a) Flow diagram showing AML patient recruitment, CANscript^TM prediction to cytarabine monotherapy and their clinical response. (b) Representative IHC images showing expression levels of: DNMT1, DNMT3A, DNMT3B, HDAC1, and HDAC6. The images were taken from sections of a tumor sample responder to cytarabine and a tumor sample nonresponder to cytarabine. The responder and nonresponders were determined from S-score, with S-Score > 19.1 being classified as responders and S-Score ≤ 19.1 being considered nonresponders. The IHC score (%) have been shown with each image. (c) IHC scores (%) are shown for the biomarkers DNMT1, DNMT3A, DNMT3B, HDAC1, and HDAC6, in two responder and four nonresponder samples. IHC scores were based on percentage of tumor cells expressing the marker

Figure 1. (Continued)

Epigenetic modulation in tumors nonresponsive to cytarabine

Epigenetic abnormalities have been implicated in drug response pattern in AML. Epigenomic profiling have revealed hypermethylation in AML.^11,17 In AML, frequent upregulation of DNMT1, DNMT3A, and DNMT3B have been reported.¹⁸ Histone deacetylases, HDAC1 and HDAC6, have been reported to be frequently overexpressed in AML.^19,20 Studies reveal that down-regulation of both HDACs 1 and 6 is critical in enhancing cytarabine-induced apoptosis in pediatric AML.²¹

We evaluated the expression of DNMT1, DNMT3A, DNMT3B, HDAC1, and HDAC6 using IHC in the cytarabine nonresponder samples. In one tumor sample (P4), the markers could not be evaluated owing to the paucity of sample. All four predicted nonresponders to cytarabine revealed high scoring pattern for DNMT1, DNMT3B (≥ 30%) compared to the responders (≤ 20%) (Figure 1(b,c)). Previous studies have correlated expression levels of DNMT1 and DNMT3B with aberrant hypermethylation of p15^INAK4B tumor suppressor gene in AML.¹⁸ Further, analysis of HDAC expression revealed that all nonresponders had high expression scores of both HDAC1 and HDAC6 (>50%) (Figure 1(b,c)). In contrast, in the two cytarabine responder samples, the expressions of all epigenetic markers evaluated were much lower than that of the nonresponders (≤ 20%) (Figure 1(b,c)). Due to limitation of sample availability, we were unable to test the DNA methylation status in these samples using methylation specific PCR. Based on the elevated expression of the epigenetic markers in the nonresponder group, we attempted to evaluate if epigenetic modulators can be selected as alternative treatment regimens for patients who do not respond to cytarabine treatment. Azacitidine is a DNA methyltransferase (DNMT)-inhibiting cytosine nucleoside analogue, approved for the treatment of MDS and AML in the United States and European Union. Azacitidine reduces genomic DNA methylation by irreversibly binding to DNA methyltransferases (DNMT) after incorporation into newly synthesized DNA.²² In addition, the covalent protein-DNA adduct formation triggers DNA damage and results in eventual cell death.²² Panobinostat is a hydroxamic acid-based pan-HDAC inhibitor approved under FDA’s accelerated approval program for the treatment of multiple myeloma. This prompted us to undertake next round of assessment that involved evaluating azacitidine and panobinostat alone and in combination for the treatment of AML.

In the next round, seven patient samples were recruited. Each of the seven samples were treated on the ex vivo platform with four treatment arms, namely (i) cytarabine monotherapy; (ii) azacitidine monotherapy; (iii) panobinostat monotherapy; and (iv) azacitidine + panobinostat combination therapy. Drug responsiveness was determined using multiparametric phenotypic assays. All 7 tumor samples were nonresponders to cytarabine, one (P28) responded to azacitidine monotherapy, and one sample (P11) responded to panobinostat monotherapy. Five samples (including the two samples that responded to azacitidine monotherapy (P28) and panabinostat monotherapy (P11)) responded to azacitidine + panobinostat combination therapy (Figure 2(a), Table 2). Two samples were nonresponders to all four treatment arms, including azacitidine + panobinostat combination therapy (P9 and P13). Supplementary Figure. S1A and S1B show representative images for H&E, Ki-67, and Caspase-3 C in cytarabine refractory samples which were responders and nonresponders respectively to epigenetic modulators. In the clinic, four patients (P9, P11, P13, and P14), were treated with cytarabine monotherapy, one patient (P28) was treated with azacitidine monotherapy, and two patients (P10 and P12) were treated with decitabine, another DNMT-inhibiting cytosine nucleoside analogue. Incidentally, in an earlier report, azacitidine and decitabine were shown to have similar clinical outcome in terms of objective response rates, overall survival, and event-free survival.²³ Clinical outcome revealed that of the four patients treated with low dose cytarabine monotherapy none showed remission of disease, thereby matching the CANscript prediction. The other three patients were not treated with cytarabine monotherapy, hence it was not feasible to match their clinical prognosis with CANScript response prediction. In addition, one patient (P28) who was predicted as a responder to azacitidine monotherapy, also showed response in clinic when treated with azacitidine. Two patients (P10 and P12), who were treated with decitabine did not show remission, hence matching the CANscript predicted outcome to azacitidine (Figure 2(a), Supplementary Table S1). None of the recruited patient samples were treated in clinic with panobinostat or with the combination of azacitidine and panobinostat.

Table 2. Response pattern to the following drug arms (i) cytarabine, (ii) azacitidine, (iii) panobinostat, and (iv) combination of azacitidine and panobinostat in the second round of recruitment

Patient ID	Cytarabine	Azacitidine	Panabinostat	Azacitidine + Panabinostat
P9	Nonresponder	Nonresponder	Nonresponder	Nonresponder
P10	Nonresponder	Nonresponder	Nonresponder	Responder
P11	Nonresponder	Nonresponder	Responder	Responder
P12	Nonresponder	Nonresponder	Nonresponder	Responder
P13	Nonresponder	Nonresponder	Nonresponder	Nonresponder
P14	Nonresponder	Nonresponder	Nonresponder	Responder
P28	Nonresponder	Responder	Nonresponder	Responder

S-Score > 19.1 is classified as responders whereas S-Score ≤ 19.1 is considered nonresponders.

Figure 2. (a) Flow diagram showing AML patient recruitment, CANscript^TM prediction to the indicated treatment arms and their clinical response. (b) Representative IHC images showing baseline expression levels of p-STAT3, in a azacitidine and panobinostat (Aza + Pano) responder and a Aza + Pano nonresponder. The responder and nonresponders were determined from S-score. (c) IHC scores (%) are shown for basal expression in two nonresponder and three responder samples. IHC scores were based on percentage of tumor cells expressing the marker

Figure 2. (Continued)

Activation of JAK/STAT pathway in tumors refractory to epigenetic modulators

Dysregulation of JAK-STAT pathway has been documented in many hematologic malignancies and members of the JAK/STAT pathway have been proposed as targets in chemotherapy resistant AML.²⁴ In order to elucidate if JAK/STAT pathway is indeed activated in AML, we studied the expression and methylation status of genes in AML patients using The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/projects/TCGA-LAML). TCGA data indicated the highest variation in DNMT3A in AML and its association with poor survival. To assess the expression level of DNMTs, RNAseq (transcriptomic data) and methylation data (n = 100) was derived from TCGA database for AML patients. Data pre-processing and normalization steps were performed using in-house perl scripts. The resulting matrix contained mRNA expression information for ~17 K genes across all analyzed samples. Genes with FPKM >1 were considered for further analysis. Transcriptomic data indicates increased expression of DNMT (DNMT1, DNMT3A, and DNMT3B) family of genes in AML patients (Supplementary Table S2A). These findings were in line with the experimental data, which showed high expression of DNMTs in the cytarabine nonresponders and suggested the possibility of using a DNMT inhibitor along with a HDAC inhibitor in cytarabine refractory tumors. In addition, the transcriptomic data also revealed increased expression of IFNGR1, IFNGR2, and STAT3 (Supplementary Table S2A). Next, methylation data preprocessing and normalization steps were performed using in-house perl scripts. Genes with beta value ≥ 0.8 were considered as hypermethylated and beta value ≤ 0.2 were considered as hypomethylated. We observed hypomethylation of IFNGR1, IFNGR2, and STAT3 genes (Supplementary Table S2C, S2E). The TCGA data showing increased expression along with hypomethylation of IFNGR1, IFNGR2 suggested activation of the JAK/STAT pathway.²⁵ TCGA data indicated hypomethylation of several gene components of JAK-STAT pathway, which was correlated with an increased expression of these genes in RNAseq data (Supplementary Figure S2, Supplementary Table S2). Taken together this data indicates aberrant expressions for DNMTs/HDACs and activation in JAK/STAT pathway in AML patients. This prompted us to test the expression of activated STAT-3 and BCl2L10 (a Bcl-2 family protein and an independent prognostic factor of azacitidine outcome in AML)²⁶ in patients who are predicted nonresponders to azacitidine and panobinostat.

We studied the cellular expression of p-STAT3 and BCL2L10 in tumor samples, which were nonresponders to azacitidine and panobinostat therapy. Data revealed that that the two nonresponders (to all the three treatment arms, P9 and P13) had high basal expression of both p-STAT3 (Figure 2(b,c)) and BCL2L10 (Supplementary Figure S3). In contrast, the 3 responders to azacitidine and panobinostat combination therapy (P10, P11, and P14) demonstrated lower p-STAT3 scores (Figure 2(b,c)). We next attempted to evaluate efficacy of fedratinib, a JAK inhibitor shown to suppress phosphorylation of STAT3.²⁷

In the next round, 16 patient samples were recruited. Ten of the 16 samples were each treated on the ex vivo CANScript platform with 5 treatment arms, (i) cytarabine; (ii) azacitidine monotherapy; (iii) panobinostat monotherapy; (iv) azacitidine + panobinostat combination therapy; and (v) fedratinib. Four samples were treated with only cytarabine owing to insufficient size of tissue. One sample was treated with two treatment arms, cytarabine and fedratinib, and one sample was treated with cytarabine and all three arms of epigenetic modulators. Drug response patterns were determined using phenotypic assays. Interestingly, of the 11 samples treated with fedratinib, 8 tumors showed response, of which 5 were nonresponders to all other treatment arms (Table 3, Figure 3(a,b,c)). For this cohort, none of the patients received treatment regime as was tested using CANScript. We studied the cellular expression of p-STAT3 in tumor samples, which were responder and nonresponder to fedratinib therapy. It was observed that nonresponder to fedratinib had high basal level of p-STAT3 (38%) and it remained unchanged (36%) after treatment with fedratinib. In contrast, in the responder sample, there was a decrease in p-STAT3 level post fedratinib treatment (15% and 5% respectively) (Figure 3(d)).

Table 3. The response pattern to the following drug arms (i) cytarabine, (ii) azacitidine, (iii) panobinostat, (iv) combination of azacitidine and panobinostat, and (v) Fedratinib in the final round of recruitment

Patient ID	Cytarabine	Azacitidine	Panabinostat	Azacitidine + Panabinostat	Fedratinib
P15	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Responder
P16	Responder	NP	NP	NP	NP
P17	Nonresponder	Nonresponder	Nonresponder	Responder	Responder
P18	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Responder
P19	Responder	NP	NP	NP	NP
P20	Responder	NP	NP	NP	NP
P21	Responder	NP	NP	NP	NP
P22	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Responder
P23	Responder	NP	NP	NP	Nonresponder
P24	Responder	Nonresponder	Nonresponder	Responder	Responder
P25	Responder	Nonresponder	Responder	Responder	Responder
P26	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Nonresponder
P27	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Responder
P29	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Nonresponder
P30	Nonresponder	Nonresponder	Nonresponder	Responder	NP
P31	Nonresponder	Nonresponder	Nonresponder	Nonresponder	Responder

S-Score > 19.1 is classified as responders whereas S-Score ≤ 19.1 is considered nonresponders. NP – indicates CANScript^TM assay not performed due to paucity of sample.

Figure 3. (a) Flow diagram showing AML patient recruitment, CANscript^TM prediction to the indicated treatment arms. (b) CANscript^TM identifies responders to fedratinib test arms of tumors refractory to all treatment arms of epigenetic modulators. Representative images taken from sections stained with H&E, Ki-67, and active Caspase-3. (c) CANscript^TM identifies nonresponders to fedratinib test arms of tumors refractory to all treatment arms of epigenetic modulators. Representative images taken from sections stained with H&E, Ki-67, and active Caspase-3. (d) Representative IHC images showing expression levels of p-STAT3 at basal and posttreatment with fedratinib, in a fedratinib responder and a nonresponder

Figure 3. (Continued)

Using this platform, we demonstrated that overall ~32% (10/31) of AML patient samples responded to cytarabine. Based on an overall trend toward a higher expression of epigenetic markers in the nonresponder group, we postulated that epigenetic modulators azacitidine and panobinostat, could be tested as an alternate treatment regimen for cytarabine refractory AML. We observed that 50% (9/18) of patient samples responded to azacitidine+panobinostat combination therapy, and 11% (2/18) and 5.5% (1/18) responded to panobinostat and azacitidine monotherapies respectively. Next, based on high expression of p-STAT3 among nonresponders to treatment by epigenetic modulators, we selected fedratinib as a treatment arm. Interestingly ~73% (8/11) of the treated patient samples responded to fedratinib (Supplementary Figure S4). When we compared the expression levels of epigenetic markers DNMT1, DNMT3A, DNMT3B, HDAC1, and HDAC6, we observed a general trend toward a higher expression of the markers in cytarabine nonresponders and a lower expression of the markers in cytarabine responders (Supplementary Figure S5A and S5B). Similarly, expression of p-STAT3, both under basal condition and posttreatment with azacitidine and panobinostat combination, showed a higher expression in the nonresponder group of azacitidine and panobinostat combination therapy compared to the responder groups (Supplementary Figure S6A and S6B). However, there were outliers within both groups. We concluded that while biomarkers used in this study provide important insight into the choice of therapy, the biomarkers alone were not sufficient as predictive/prognostic factors in selecting treatment at an individual level for AML patients.

Our study demonstrates that CANscript^TM as a suitable platform to guide personalized treatment selection in hematological malignancies like AML. Furthermore, data from our CANscript platform suggests that the JAK/STAT pathway can be further investigated as a novel therapeutic target in AML.

Discussion

In the present study, we have used an ex vivo tumor microenvironment platform (CANscript^TM) that effectively recreates intratumoral heterogeneity close to that of the patient. The platform provides functional and phenotypic assessment of drug response on a patient’s tumor conserved in a native tumor microenvironment. Previous studies described ex vivo and in vivo preclinical models for AML to investigate therapeutic response or for understanding mechanistic underpinnings.^28–33 CANscript^TM is a unique platform that has been developed as a clinically relevant predictive model, in which a tumor ecosystem is reconstructed ex vivo with grade-matched tumor matrix support in the presence of matched autologous serum. The ex vivo tumor ecosystem, together with a machine learning algorithm form a powerful personalized system for predicting the therapeutic efficacy of anticancer drugs in hematological malignancies based on purely phenotypic readouts.

Using the ex vivo, functional platform, we demonstrated that only 32% of AML patient samples responded to the standard of care cytarabine. Studies have shown complete cytarabine response rates between 40 and 80% of patients stratified into poor, intermediate, and favorable genetic risk groups respectively³⁴ defined according to recommendations published earlier.¹⁴ Chemoresistance in AML has been attributed to hypermethylation of gene promoters and associated silencing of tumor suppressor genes. Methylation of mammalian genomic DNA is catalyzed by DNA-methyltransferase enzymes (DNMTs). Drugs that inhibit DNA methylation have been used in AML and shown to enhance cytarabine sensitivity.³⁵ Posttranslation modifications in histone, including acetylation, deacetylation, and phosphorylation are other important epigenetic changes in humans. Their dysregulation is brought about by changes in histone acetylation (HAT) and deacetylation (HDAC) and has been implicated in several malignancies including AML.³⁶ HDAC enzymes have been shown to be overexpressed in several leukemias and are evaluated as drug targets. Several HDAC inhibitors (HDI) are in clinical trials and four are approved in the clinic.³⁷ We evaluated the epigenetic markers both in an attempt to explain cytarabine nonresponsiveness and also to guide our decision toward the selection of epigenetic modulators as one of the treatment options for patients refractory to cytarabine. Based on the expression pattern of the epigenetic markers, we next included three additional treatments arms, namely, azacitidine monotherapy, panobinostat monotherapy, and azacitidine and panobinostat combination therapy. Our data indicated that 5% and 11% of patient samples in the overall study cohort responded to azacitidine and panobinostat monotherapies respectively as per response prediction done on an ex vivo, functional platform developed in our labs. Previous reports indicate that when panobinostat was used alone in the clinics for hematologic malignancies, responses were observed in 4/83 (5%) patients with AML and 1/11 (9%) patients with MDS.³⁸ In contrast, panobinostat used in combination with azacitidine resulted in an overall clinical response of 35% (AML 31%, MDS 50%).³⁹ In addition, in a preclinical study, panobinostat was shown to act synergistically with decitabine, another demethylating agent, causing greater reduction in AML cell viability compared to the single agents alone.⁴⁰ Contrasting results were obtained from another recent clinical trial data, where panobinostat did not enhance the therapeutic benefits of azacitidine in AML, even though improved responses were seen in CMML and MDS with the combination therapy.⁴¹ Using our ex vivo platform, we observed that ~50% of patient samples responded to azacitidine and panobinostat combination therapy. Previous studies have used whole genome approaches to determine the genetic causes for cytarabine sensitivity.^42,43 While these studies are important in predicting response of AML patients to cytarabine treatment, very few studies have actually provided an actionable strategy for rationally selecting a treatment course in patients who do not respond to cytarabine. Using our personalized approach, we have shown that azacitidine and panobinostat combination can be an alternative treatment option for patients who are refractory to cytarabine. Notably, of the 15 patients’ samples recruited in the first two rounds, 9 patients had received a treatment matching the drug arms tested on CANscript^TM. A full concordance was observed in this small cohort of patients, between clinical response and CANscript^TM predicted response. Next, we attempted to rationalize a therapy option for tumors that did not respond to any of the epigenetic modulators.

We observed activation of the JAK/STAT pathway using TCGA data for AML patients. STAT signaling contributes to cellular transformation through modulation of target genes. Following phosphorylation by various kinases, STATs translocate to the nucleus, bind to a consensus site in the promoter of target genes, and activate or repress transcription. Targets of STATs include genes involved in promoting proliferation, survival, angiogenesis and invasion, including the Bcl-2 family of genes.⁴⁴ Aberrant activation of STAT3 has been observed in AML and been associated with poor prognosis and chemoresistance in AML.⁴⁵ Activation of STAT3 has been correlated with the overexpression of Bcl-2 family of proteins.^46,47 High levels of phosphorylated STAT3 have also been implicated in lack of clinical response to vorinostat, another hydroxamic acid-based pan-HDAC inhibitor.⁴⁸ In other studies, BCL2L10, an antiapoptotic member of the Bcl-2 family, has been correlated with azacitidine resistance.²⁶ Majority of the nonresponders to azacitidine and panobinostat combination therapy showed both high basal levels of phosphorylated-STAT3 compared to the responders. We evaluated fedratinib, a JAK2 inhibitor shown to reduce STAT3 phosphorylation, as a potential treatment regimen for AML.²⁷ In a previously reported clinical trial, ruxolitinib, the only FDA approved JAK2 inhibitor was evaluated in patients with refractory AML and showed modest antileukemic activity.⁴⁹ In a more recent study, fedratinib was shown to reduce leukemia in mouse xenograft models in 59% of cases, while ruxolitinib was ineffective.⁵⁰ Using our ex vivo platform we showed a predicted response rate of ~73% (8/11) to fedratinib treatment. We do not rule out the activation of other pathways in AML cases which are nonresponder to cytarabine monotherapy or epigenetic modulators. Indeed, we do note that amongst the 16 patients recruited in third round (Figure 3), 3 patients did not respond to fedratinib as per our ex vivo platform prediction. Further experiments are required to understand the cause of this nonresponse, which is beyond the scope of this manuscript.

In this study, we have developed an ex vivo, phenotypic platform that maintains the native tumor contexture and thereby captures the inherent complexity of this key hematological malignancy, AML in a personalized setting. Using the platform, we have demonstrated rational selection of treatment modalities for AML. We have depicted fedratinib as a novel therapeutic agent in AML patients who are nonresponders to cytarabine or azacitidine. Recent advances in diagnosis and management of AML based on genetic testing has been recommended by European Leukemia Net (ELN), which also includes the dose recommendations of cytarabine.^15,51 Taken together our current data establish CANscript^TM as a personalized ex vivo platform to guide treatment course in AML. We have also shown a good correlation with results from previously reported clinical studies. Further, we have shown concordance with clinical outcome in patients who received matching treatment arms as those tested on the CANscript^TM platform. Our study is limited by the small cohort of patient samples and further by the fact that matching clinical outcome was available for an even smaller subset. Further clinical validation of this using larger sets of patient samples is required to understand the clinical outcome of fedratinib in AML.

Materials and methods

Patients

Bone marrow biopsies (n = 31) from patients with AML (Supporting Information Table 1) were collected in Lactate Ringer’s transport buffer and transported in BD Vacutainer tubes (BD Biosciences, USA) to the laboratory. Informed consent was received prior to collection of clinical tissue specimens from the patients. Blood samples were also collected from the same patients and maintained under refrigerated condition for further use. The experimental study protocol was approved by the Institutional Review Board (IRB) of the respective hospitals and carried out in accordance with regional guidelines and standards. All experiments were commenced within 24 hours of the biopsies. A summary of the demographic is listed in Supplementary Table S3.

Drugs

Azacitidine (Natco Pharma, India; Catalog no.: 500152), Panobinostat (Selleckchem, TX, USA; Catalog no: S1030), Cytarabine (Sigma-Aldrich, MO, USA; Catalog no: C1768), Fedratinib (Selleckchem, MO, USA; Catalog no: S2736) were procured. The explants were treated with the dugs at the following concentrations based on their C_max values. The C_max values were 638 ng/ml for Azacitidine, 21.6 ng/ml for Panobinostat, 26.3 ng/ml for Cytarabine, and 525 ng/ml for Fedratinib.

CANscript explant culture

Bone marrow biopsies were manually dissected to the explant size of 2–3 mm³. These explants were randomized in the culture to ensure tumor heterogeneity. The explants were cultured either in triplicate or quadruplicate in 48 well plate coated with indication specific tumor matrix proteins as described.^16,52 The plates were observed under inverted phase contrast microscope every 24 h to ascertain tissue integrity.

Viability assay

Tumor cell viability was assessed by Cell Counting Kit-8 (CCK-8) (Dojindo). Briefly, one-tenth of the volume of the CCK-8 solution was added to each well of the plate and incubated at 37°C for 3 h in a 5% CO₂ incubator under humidified condition. Absorbance was measured at 450 nm using a multimode microplate reader (EnSpire®, Perkin Elmer, MA, USA). Baseline samples (T0) were used as calibrators (1×) to normalize intersample variation in absorbance readings. The results were expressed as a percentage of tumor cell viability or inhibition relative to vehicle controls.

Histopathological evaluation

Histological characterization of AML explants was conducted before and after culture. A small part of the bone marrow biopsies were fixed in 10% formalin immediately after the biopsy and embedded in paraffin (T0 baseline). Following decalcification (5% HCl for 2–3 days), the explants were sectioned at 4 µm and stained with hematoxylin and eosin. The explants receiving different treatments were graded and scored according to standard clinical criteria.⁵³

Immunohistochemical analysis

Immunohistochemistry (IHC) of the tissue sections were performed using tissue sections, to study the expression profiles of biomarkers Ki-67, active Caspase-3, DNMT1, DNMT3A, DNMT3B, histone deacetylase 1 (HDAC1), and histone deacetylase 6 (HDAC6). Antigen retrieval was done by using Antigen Unmasking Solution (Vector Laboratories, CA, USA). For each IHC staining, appropriate positive and negative controls were included. Microphotographs were taken at 200× using Leica DM-2500 Microscope (Leica Microsystems, Germany). Changes in the levels of cell proliferation or cell death after drug treatment (compared to vehicle control) were evaluated by specific proliferation/apoptosis markers using rabbit polyclonal Ki-67 at 1:600 dilution (Vector Laboratories, CA, USA), and rabbit polyclonal caspase-3 (cleaved) antibodies at 1:600 dilution (Cell Signaling Technology, MA, USA). For detecting levels of the epigenetic markers, the following antibodies were used, rabbit polyclonal anti-DNMT1 at dilution of 1: 500 (Catalog number: ab19905; Abcam, USA), rabbit polyclonal anti-DNMT3A at of dilution 1:50 (Catalog number: HPA026588; Atlas Antibodies, Sweden), rabbit polyclonal anti-DNMT3B at dilution of 1:200 (Catalog number: HPA001595; Atlas Antibodies, Sweden), rabbit polyclonal anti-HDAC1 at dilution of 1:100 (Catalog number: ab19845; Abcam, USA), and rabbit polyclonal anti-HDAC6 at dilution of 1:200 (Catalog number: ab117516; Abcam, USA). Rabbit polyclonal anti-BCL2L10 (Catalog number: ab151419; Abcam, USA) was used at dilution 1:100 and rabbit monoclonal antibody Phosphorylated-STAT3 (Tyr705) (Catalog number: 9145; Cell Signaling Technology, MA, USA) was used at dilution 1:50. The IHC scores were given as percentage of tumor cells expressing the marker. For DNMT1, DNMT3A, DNMT3B, HDAC1, HDAC6, BCL2L10, and Phosphorylated-STAT3 (p-STAT3), the results of IHC staining were also categorized using the following cutoff percentages, 0 (0%), 1 (≤5%), 2 (6–25%), 3 (26–50%), 4 (51–75%), or 5 (>76%), respectively.⁵⁴

S-SCOre

Values from histology (morphology), tumor cell proliferation, cell death and viability were used as inputs in a machine learning algorithm as described earlier.^16,52 The algorithm finally generates a single score (S-Score) which has the potential of predicting clinical outcome to therapy. A value greater than 19.1 may clinically correspond with response to the drug(s) or combinations tested in CANscript^TM, while a value lower or equal to 19.1 may indicate nonresponse.^16,52

Transcriptome and methylation analysis using TCGA data

Txt files of 100 patients were downloaded from TCGA portal (https://portal.gdc.cancer.gov/) for AML patients. For RNAseq analysis, FPKM txt files were downloaded. Matrix of FPKM (fragments per kilo-base of transcripts per million mapped reads) values was generated using in-house perl script. Data was normalized and median value of FPKM was calculated for each gene. Genes with FPKM >1 were considered for further analysis. Similarly, for methylation analysis txt files were downloaded and a matrix of (Probe_ID, Beta_value, Gene_symbol, and Position_to_TSS) was generated using in-house perl scripts. Data normalization and median beta value for each gene was calculated. The coding genes were only considered for further analysis. Genes with a beta value ≥ 0.8 were considered as hypermethylated and a beta value ≤ 0.2 were considered as hypomethylated.

Authors’ contributions

Conception and design: Govind Babu and Saravanan Thiyagarajan

Development of methodology: Manoj Rajappa and Saravanan Thiyagarajan

Acquisition of data (CANscript data; Pathology Images/interpretation; Patient Handling and information.): Manoj Rajappa, Manjusha Biswas, Biplab Tewary, Bipinesh Sansar, Chethan Rajegowda, and Govind Babu

Computational analysis: Jayshree Advani, Aneesha Radhakrishnan, Padmaparna Chaudhuri, and Aditi Chatterjee

Writing, review, and/or revision of the manuscript: Saravanan Thiyagarajan, Ram Shankar Upadhayaya, Govind Babu, Padmaparna Chaudhuri, and Aditi Chatterjee

Study supervision: Padhma Radhakrishnan and Pradip K Majumder

Disclosure of Potential Conflicts of Interest:

Except Bipinesh Sansar, Chethan Rajegowda, Govind Babu and Jayshree Advani all authors are either employees of Mitra Biotech and/or hold equity of Mitra Biotech. The remaining authors declare no competing financial interests.

Acknowledgments

We are grateful to Dr. Biswanath Majunder for critically reading the manuscript and scientific inputs. We thank Arun Prasath for assisting in S-score generation and Dr. Debapriya Mehrotra for pathological insights.

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website.