A novel immunohistochemical score to predict early mortality in acute myeloid leukemia patients based on indoleamine 2,3 dioxygenase expression

Indoleamine 2,3 dioxygenase-1 (IDO-1) is an enzyme in the kynurenine pathway which augments tumor-induced immune tolerance. Previous studies in childhood acute myeloid leukemia (AML) have shown a negative correlation of IDO-1 mRNA expression with outcomes. The aim of our study was to develop a practical and objective immunohistochemical technique to quantify IDO-1 expression on diagnostic bone marrow biopsies of AML patients in order to facilitate its use in routine clinical practice. IDO-1 mRNA was extracted from diagnostic bone marrow specimens from 29 AML patients. IDO-1 protein expression was assessed in 40 cases via immunohistochemistry and quantified by a novel ‘composite IDO-1 score’. In a univariate analysis, higher age (p = 0.0018), male gender (p = 0.019), high risk cytogenetics (p = 0.002), higher IDO-1 mRNA (p = 0.005), higher composite IDO-1 score (p < 0.0001) and not undergoing allogeneic stem cell transplant (SCT, p = 0.0005) predicted poor overall survival. In a multivariate model that included the aforementioned variables, higher composite IDO-1 score (p = 0.007) and not undergoing allogeneic SCT (p = 0.007) was found to significantly predict poor outcomes. Further, patients who failed induction had higher composite IDO-1 score (p = 0.01). In conclusion, ‘composite IDO-1 score’ is a prognostic tool that can help identify a certain subset of AML patients with ‘early mortality’. This unique subset of patients can potentially benefit from specific IDO-1 inhibitor therapy, currently in clinical trials.

carcinoma 12,13 , endometrial carcinoma 14 , serous ovarian tumors, melanoma 15 , hepatocellular carcinoma 16 and colonic adenocarcinoma 17,18 . In tumors, inhibition of the IDO pathway is theorized to help ameliorate a state of immune privilege created by tumor cells enhancing endogenous T-cell mediated response against the tumor 17 . In the case of Acute Myeloid Leukemia (AML), preclinical studies in both adults and children have found a positive correlation of increased expression of IDO1 mRNA or functional activity in leukemic blasts correlated with worse overall survival (OS) [19][20][21] . This has prompted initiation of a clinical trial in which the IDO-pathway inhibitor indoximod will be combined with standard idarubicin/cytarabine chemotherapy in newly-diagnosed adult AML (NCT02835729). In adults, the subset of patients with by far the worst prognosis, fails to enter remission with induction chemotherapy. These patients often have a relentlessly downhill course despite best available therapy, and in certain high-risk populations such as elderly patients, over half will be dead within 6 months of diagnosis. This "early-mortality" subset represents a population that is in particularly urgent need of improved treatment. We hypothesized that the early-mortality population might represent patients with the highest IDO-1 expression, and thus the candidates most in need of an IDO-inhibitor drug as a component of their treatment regimen. At present, however, there are only a few clinical scores to predict in advance which patients will fail induction 22 and it would be useful to have other novel biomarkers such as IDO-1 to predict induction success or failure. We hypothesized that immunohistochemical staining of initial diagnostic bone-marrows biopsies for a combination of extent and intensity of IDO-1 staining might be used to generate an objective pathologic score of IDO-1 expression; and that this would allow accurate prospective identification of those patients at highest risk of induction-failure and early mortality.
IDO-1 mRNA has a negative correlation with OS. IDO-1 mRNA expression (n = 29) was compared across the 4 different survival groups defined in Methods (Fig. 1A), with the highest levels in <6 month group ("early-mortality" group). Values were measured as per the fold change, with 18S as the housekeeping gene. 1.39, 0.99, 0.36 and 0.31 were the respective fold change values in groups 1, 2, 3 and 4 respectively.

IDO-1 protein expression has a negative correlation with OS.
Analysis of IDO-1 protein expression by immunohistochemistry (IHC) (n = 40) also showed a similar statistical correlation. After a receiver operating characteristic (ROC) analysis, a cut-off point of composite IDO-1 score was arrived at 0.45 and samples were divided into two groups: high (≥0.45) and low (<0.45). Kaplan-Meir survival analysis highlights direct correlation of poor survival with higher composite IDO-1 score (p = 0.0005, Fig. 1B,C). When patients were stratified by conventional risk categories, the highest composite IDO-1 score was seen in the poor risk category (median 1.5, n = 9), followed by intermediate (median 0.6, n = 27) and good risk categories (median 0.13, n = 4, p = 0.005, Fig. 1D). Further, IDO-1 mRNA and protein expression had a positive correlation as assessed by one-way analysis of variance (ANOVA, p = 0.024).

Multivariate analysis proves composite IDO-1 score to be an independent prognostic factor.
In a univariate model that included age, sex, race, body-mass index (BMI), allogeneic SCT, induction treatment strategy (standard cytarabine and anthracycline induction versus hypomethylating agents), remission achieved versus not, diagnostic peripheral and bone marrow blast percentage, only higher age (p = 0.0018), male gender (p = 0.019), high risk cytogenetics (p = 0.002), higher IDO-1 mRNA levels (p = 0.005), higher IDO-1 protein expression (as measure by composite IDO-1 score, p < 0.0001) and not undergoing allogeneic stem cell transplant (SCT, p = 0.0005) all correlated with a poor OS. However in a multivariate model which included age, sex, allogeneic stem cell transplant, IDO-1 mRNA levels, composite IDO-1 score and risk category, only higher composite IDO-1 score (p = 0.003) and not undergoing allogeneic SCT (p = 0.007) remained significantly associated with poor OS (Table 2). Multivariate analysis was performed using the Cox proportional hazards model. Induction failure was inversely correlated with composite IDO-1 score. One-way analysis of variance (ANOVA) of composite IDO-1 score was compared with remission achieved with either one or multiple rounds of induction therapy (either with standard cytarabine and anthracycline induction versus hypomethylating agents) and we found that patients who did not achieve remission had higher composite IDO-1 score (p = 0.01, Fig. 1E).
The Cancer Genome Atlas (TCGA) data confirms the prognostic role of IDO-1. The importance of IDO1 as a prospective risk classifier was validated from the external TCGA AML database. High β median value of IDO-1 promoter methylation status correlated with worse OS (p = 6e-04). Survival analysis of the TCGA AML patients' cohort based on IDO1 mRNA expression is shown in Fig. 2A and promoter methylation is shown in Fig. 2B.
Scientific RepoRts | 7: 12892 | DOI:10.1038/s41598-017-12940-0 Discussion IDO-1 expression has been shown to affect survival in multiple tumors, including AML. Our study shows that increased IDO-1 mRNA and protein expression correlates with a lower OS. Importantly, increased IDO-1 expression in leukemic blasts at diagnosis, defined by a composite IDO-1 score performed on readily-available, archival diagnostic biopsies, correctly identified the subset of patients with highest risk of early mortality. In a multivariate model, high composite IDO-1 score at diagnosis was found to be an independent predictor of poor outcomes in AML.
In AML, tumor microenvironment is inherently hostile to immune effector cells 19 . IDO-1 is hypothesized to be a critical contributor to this phenomenon. Similar to our results, Chamuleau ME et al. showed that high IDO-1 mRNA in blasts was shown to correlate with poor outcomes 20 . Also, IDO-1 activity was shown to correlate with poor outcomes in children by Folgiero et al. 19 . Constitutive IDO-1 expression in AML blasts has been shown in about 50% of cases of AML, with the remaining half showing upregulation of IDO-1 mRNA after IFN-y stimulation 19  versus leukemia immunity play a critical role in clearance of leukemic blasts. Therefore, it seems reasonable that inhibition of a potent immunosuppressive mechanism such as IDO-1 might further enhance this immunotherapeutic effect and improve outcomes. The current study shows that it is possible to prospectively identify a subset of AML patients with the highest IDO-1 expression at the highest risk for early mortality. This may help as a companion diagnostic approach to identify those patients most likely to benefit from the addition of IDO-inhibitor drugs to front-line therapy for AML. Despite the fact that we used rigorous statistical modeling and an external cohort to validate our results, there are certain limitations of our analysis. Due to small sample size, it is possible that our multivariate analysis is underpowered. To address this limitation, we used an external cohort (TCGA) to confirm our results. Due to fewer numbers and the fact that all patients with allogeneic SCT were performed at CR1, survival data for this variable should be interpreted with caution. Mutational analysis was not used as whole or targeted exome  sequencing analysis was not performed on our cohort. It would be interesting to assess how IDO-1 expression correlates with different mutations.

Materials and Methods
Patient samples. Fresh-frozen paraffin embedded (FFPE) diagnostic bone marrow biopsy specimens were retrospectively collected from forty AML patients after obtaining institutional review board (IRB) approval. Detailed clinical, laboratory and outcomes data was recorded. Risk categories were defined as per the 2010 European Leukemia Net (ELN) classification system (due to limited numbers, intermediate-1 and 2 sub-categories were regrouped under one intermediate group) 24 . Messenger RNA (mRNA) was extracted from 29 specimens using a commercial kit (Qiagen) according to manufacturer's protocol, following which, IDO-1 gene expression was checked via real time quantitative PCR. IDO-1 protein expression was checked via immunohistochemistry in 40 specimens. All methods were performed in accordance with the relevant guidelines and regulations of Augusta University and were approved by its ethical committee and IRB. Prior bone marrow biopsy specimens were used and do not require informed consent.

RNA isolation and cDNA synthesis.
Composite IDO-1 Score. IDO-1 protein expression was quantified through IHC. Two slides were prepared for each of the 40 specimens, one for hematoxylin/eosin staining and the other for staining with IDO-1 antibody. IHC experiments were carried out on 4% FFPE tissues. 4 µm in thickness sections were deparaffinized in xylene and hydrated in water. Endogenous peroxidase was blocked with 0.3% hydrogen peroxide in distilled water for 10 minute in room temperature (RT). Slides were rinsed with distilled water 3 times and then with Phosphate buffered saline with Tween 20 (PBST) for 5 minutes. Proteinase K pretreatment for 12 minutes at RT was done for antigen retrieval. Slides then washed one times with PBST and were blocked with 0.5% nonfat dry milk in PBST for 30 minutes at RT. For detection of IDO-1 protein expression, specimens were incubated overnight at +4° C with antimouse IDO-1 antibody in a 1:100 dilution. Visualization of bound antibodies was carried out using a streptavidin-biotin-peroxidase complex and 3-amino-9-ethylcarbazole as chromogen. Normal breast FFPE sections were used as a negative control. Percentage of mononuclear cells stained was noted and intensity of staining was graded as weak (1), moderate (2) or strong (3) by two independent pathologists (Fig. 3). Composite IDO-1 score was calculated by multiplying the grade of staining intensity by the percentage of stained mononuclear cells.
External cohort analysis. TCGA database of AML patients was downloaded from the University of California, Santa Cruz (UCSC) website. The RNA-seq and Illumina 450 K methylation array datasets are downloaded from the UCSC Cancer Genome Browser (https://genome-cancer.ucsc.edu). The downloaded data are already processed and normalized by the UCSC Cancer Genome Browser team. The RNA-seq data are log2 transformed FPKM values. We did not perform any further processing on the downloaded RNA-seq data. The methylation data downloaded from the UCSC cancer genome browser are β-values subtracted by 0.5 and the values are between (−0.5 and 0.5). We restored the data to β-value by adding 0.5 to each data point. No other alteration was done to the methylation data set. Statistical analysis. Data was analyzed through JMP Pro (Version 7) statistical software. Unpaired t tests were used as appropriate. One-way ANOVA was used to assess correlation between variables. Multivariate analysis was performed using the Cox proportional hazards model. A p-value of 0.05 was considered significant and log-rank statistical test was used to derive the results. Data is available for verification.