Host IDO2 Gene Status Influences Tumor Progression and Radiotherapy Response in KRAS-Driven Sporadic Pancreatic Cancers

Immune-checkpoint therapeutics, such as the CTLA-4–, PD1-, and PD-L1–directed antibodies that have revolutionized the treatment of melanoma and non–small cell lung cancer, have generated enormous interest for their prospects to restore T-cell–mediated immune surveillance in other types of cancer, but several, including pancreatic ductal adenocarcinoma (PDAC), have thus far proven refractory to such agents (1), with the exception of cases with microsatellite instability (MSI-H). The prospect of improving immune-checkpoint therapy responsiveness has intensified the search for other host factors involved in licensing immune escape. The tryptophan catabolic enzyme, indoleamine 2,3-dioxygenase-1 (IDO1), has attracted attention in this regard as a mediator of tumor-promoting inflammation and the locoregional suppression of antitumor T cells (2, 3). IDO1 overexpression occurs in >50% of human cancers where it has been generally associated with a poor prognosis (4). A large body of preclinical work has established that small-molecule IDO inhibitors (IDOi) can reprogram the inflammatory microenvironment and overcome immune barriers erected by tumors against antitumor T cells, including those involving T regulatory cells (Treg), tumor-associated macrophages, and myeloid-derived suppressor cells (MDSC; refs. 2,5). Preclinical studies have argued that IDOi will be particularly effective as part of a combined therapy including DNA-damaging chemotherapy, radiotherapy, and immune-checkpoint therapy (6–8), although the absence of demonstrable benefit in a recent phase III trial indicates that a deeper mechanistic understanding is essential for the effective therapeutic use of these agents. An important genetic paralog, IDO2, has been relatively less investigated, but it has been recently implicated in B cell–mediated autoimmunity (9, 10). Intriguing preclinical studies in mice also suggest genetic and biochemical interactions between IDO1 and IDO2, which may influence IDO1-induced Treg activation (11, 12), a possibility supported by studies of human dendritic cells where expression of both of these distinct IDO genes has been noted (13).

In tumors, overexpression of IDO2 appears to be much less common than IDO1, but PDAC displays the unusual feature of overexpressing both IDO1 and IDO2 (14, 15). Two single-nucleotide polymorphisms (SNP) in the coding region of human IDO2 have been described (rs4503083 and rs10109853), which decrease enzymatic activity through amino acid substitution at the active site (R248W) or by truncation of the enzyme (Y359X), respectively (16). Notably, while both of these SNPs are quite prevalent in human populations, their clinical significance has remained undefined (16). In this study, we have obtained genetic evidence supporting IDO2's relevance to PDAC tumorigenesis using a Kras-driven PDAC mouse model (17) in which Ido2 was genetically targeted for deletion, in conjunction with an analysis of the prevalence of the two IDO2-inactivating SNPs in PDAC patients. On the basis of these findings, we performed retrospective analyses of treatment outcomes for surgically resected PDAC patients based on their IDO2 genotype status. For the subset of patients who had received adjuvant radiotherapy during the course of treatment, these analyses uncovered a significant association between IDO2-deficient status and improved disease-free survival, a finding with potential ramifications for informing future treatment decisions for this intractable disease.

IDO2-nullizygous mice and the genetically defined KC mouse model of KRAS-induced PDAC on a common C57BL/6J background strain have been described (11, 17). PDX-1-cre;LSL-Kras^G12D transgenic mice (KC mice) develop pancreatic intraepithelial neoplasias (PanINs) with complete penetrance along with sporadic focal pancreatic carcinomas with reduced penetrance due to PDX1-cre–mediated activation of the latent oncogenic Kras^G12D allele in pancreatic progenitor cells (17). These KRAS-induced lesions elicit a robust inflammatory response including B-cell contributions (18) where IDO2 may act to influence the tissue microenvironment (10). To investigate this hypothesis in an autochthonous pancreatic tumor setting, we introduced Ido2^−/− (Ido2-nullizygous) alleles (11) into the LSL-Kras^G12D mouse strain. Mice were interbred to generate Ido2^+/+ and Ido2^−/− KC siblings in which KRAS is activated with similar kinetics for longitudinal comparisons of disease initiation and progression for 11 to 13 months duration. Two independent cohorts were generated and analyzed. Histologic analysis of pancreatic lesions was conducted by standard methods as previously described (17, 19).

Single-cell suspensions were prepared from resected pancreata using a gentleMACS Octo Dissociator with the Tumor Dissociation Kit as per the manufacturers' instructions. Levels of the following cell-surface markers were directly measured by flow cytometry on a BD FACSCanto (BD Biosciences) in two separate groups as noted and analyzed using FlowJo Software (Tree Star). Group 1: CD45 (APC; BioLegend), CD11b (PE/Cy7; BioLegend), Gr1 (PerCP; BioLegend), CD11c (PE; BioLegend), F4/80 (Alexa Fluor 488; BioLegend), Fixable Viability Dye (eFluor 780; eBioscience), Group 2: CD45 (APC; BioLegend), Thy1.2 (Alexa Fluor 488; BioLegend), IgMa (PE; BD Pharmingen), CD4 (PE/Cy7; BioLegend), CD8a (PerCP; BD Pharmingen), Fixable Viability Dye (eFluor 780; eBioscience).

The study was conducted in accordance with the ethical guidelines of the Belmont Report with a statement of informed written consent obtained from each subject as appropriate. From the IRB-approved Thomas Jefferson University Hospital data set (TJUH data set, all patients in the cohort have given their informed consent) genomic DNA from surgically resected pancreatic tissue specimens (normal and tumor tissues) was extracted using the DNAeasy Blood and Tissue Kit genomic DNA purification kit (Qiagen Inc.). DNA fragments containing the IDO2-coding region polymorphisms rs4503083 (Exon 11) and rs10109853 (Exon 9) were amplified by PCR as described previously (IDO2 oligonucleotide primers R248W FWD and R248W REV; Y359X FWD and Y359X REV; ref. 14), as detailed in Supplementary Table SI. PCR reactions were performed in 25 μL using 100 ng of gDNA, 0.5 μg/μL of Taq polymerase (Affymetrix), 1 μL of 10 μmol/L oligonucleotide primers (forward and reverse), 2.5 μL of 10X PCR buffer (Affymetrix), and 0.5 μL 10 mmol/L dNTP Mix (Affymetrix). PCR reaction products were purified using a commercial PCR purification kit (Qiagen Inc.). Each PCR reaction was examined by gel electrophoresis on a 0.75% DNA agarose gel before Sanger sequencing by a commercial provider (GenScript Inc.) using the DNA oligonucleotide primers mentioned above (14). Genotyping steps were blinded to clinical data and familial-sporadic patient status. IDO2 genotype in patients was determined by chromatogram (14).

Two pancreatic cancer patient sets were used in this study as described below. Demographic and histologic data are summarized in Table 1.

The cohort used for this data set included 123 patients from The Cancer Genome Atlas (TCGA) research network database (20). All patients included in the analysis were diagnosed with histologically confirmed PDAC. The set included demographic data, operative findings, histologic features including percentages of lymphocyte and neutrophil infiltrates, RNA expression data, complete variant genotyping as well as survival and recurrence data.

The cohort used for this data set included 77 patients who underwent surgical resection with curative intent at the Thomas Jefferson University Hospital, and we had available tissue for DNA analysis (TJUH, IRB Consented). Patient specimens analyzed included PDAC cases that were familial (n = 14, 18%) or sporadic (n = 63, 82%). Familial cases were defined as two first-degree family members with PDAC as established in previous reports (21). Medical history, preoperative laboratory tests, surgical and histologic findings, and oncologic follow-up data were recorded from the patients' medical records. Familial data were extracted from medical records and from the Jefferson Pancreatic Tumor Registry.

TJUH and TCGA data sets were pooled into a single large data set and screened for duplicates. A unified set containing only PDA patients which underwent primary resection was composed (N = 200).

In the TJUH data set, slides were reviewed by Thomas Jefferson University pathologists (MC, TV). For slide quantification, hematoxylin and eosin stained sections of formalin-fixed paraffin-embedded tumor from the selected patients were reviewed for pathologic confirmation and the adequacy of tissue quantity and preservation for NLR determination. NLR was derived from the quotient of the absolute neutrophil count and the absolute lymphocyte count. For each sample, NLR was determined in three areas each measuring 0.785 mm². Final NLRs for each specimen were calculated as the average value of the three areas analyzed. Cases in which no immune cell infiltrates and no tumor cells were found were excluded from the analysis as well as cases with active biliopancreatic sepsis (e.g., cholangitis, acute pancreatitis, peripancreatic abscess, etc.). NLR data of 43 TJUH patients (56%) were available for analysis. In the TCGA data set, hematoxylin and eosin (H&E)-stained sections of snap-frozen OCT embedded tissues were reviewed by TCGA-participating histopathologists for validation as PDAC and analysis of histologic features. Tumor, normal, and stromal components were quantified (in percentages) as well as proportion of immune cells (neutrophils, monocytes, lymphocytes; ref. 22). Slides with no or sparse immune cell infiltrates (lymphocytes% + neutrophils% ≤ 10%) were excluded from the analysis. NLR was calculated for each slide and averaged per patient in cases of multiple patient slides. Cases with no visible neutrophilic infiltration were recorded as 0.01% infiltration to allow subsequent log₁₀ transform and Z-score calculation. NLR data of 56 patients (46%) were available for analysis. Overall, 99 patients were included from the pooled data set.

Categorical data were expressed as percentages and continuous data were expressed as mean ± standard deviation. For normally distributed continuous variables, a Student t test was used. Variables were assessed for normality of distribution with the Kolmogorov–Smirnov test. Comparisons between genotype groups were performed using Mann–Whitney and Jonckheere–Terpstra (J-T) tests for nonparametric distributions and t tests for normal distributions (23). Categorical data were compared with χ² test or Fisher exact test. P values of ≤0.05 were defined as significant.

Genotype distribution was quantified for all sets as well as separately for familial PDAC cases and sporadic PDAC cases. Genotype distribution of each polymorphism was analyzed for Hardy–Weinberg deviation using χ² test and Fisher exact test. A genotype distribution set of Utah residents (CEPH) with northern and western ancestry was available from the 1000 Genomes Project to be used as a control for comparison of the PDAC patient sets. Patients were grouped into two categories dependent on IDO2 genotype: homozygous alleles of either R248W (rs4503083) or the Y359X (rs10109853) polymorphism and patients with a double-heterozygous genotype (i.e., WT:R248W or WT:Y359X) were considered IDO2-deficient genotypes, whereas all other combinations were considered as active or partially active IDO2 genotypes. For correlation studies, a three-tier scale (as detailed in Supplementary Table S2) was used to rate the possible combined R248W/Y359X genotypes in terms of probable IDO2 functionality.

TJUH and TCGA specimens in which histopathologic examination defined a high tumor cellularity (≥50% cellularity) and evidence of inflammatory infiltrate (lymphocyte% + neutrophil% > 10%) were included in the analysis. Bivariate nonparametric Spearman test was used to assess correlations between IDO2 functionality grade and histologic NLR scores. In-between group comparisons were performed for the pooled data set with subanalyses for both data sets. Slides were reviewed by TJUH pathologists (M. Curtis and T. Villatoro).

RNA expression data were collected from the TCGA database using cBioPortal (24, 25). Tissue immune cell counts were imputed from GAPDH-normalized gene-expression levels (based on the Affymetrix 133Plus2 gene-expression set) using the validated MCP-Counts algorithm described previously (26, 27). NLR scores were imputed from the cell count estimates. Estimates and ratios were compared across IDO2 gene functionality grades (fully active genotype, partially active genotype, deficient genotype).

The primary hypothesis was that IDO2-deficient genotypes conferred a favorable prognostic effect. Kaplan–Meier survival analysis stratified by lymph node metastasis was used with log-rank tests to compare survival according to tumor grade, tumor size, perineural invasion, and IDO2 genotype (deficient vs. active/partially active). Survival data from the TCGA-PAAD cohort was limited due to 40% and 63% censorship rates for overall survival (OS) and disease-free survival (DFS), respectively. Survival data from the TJUH cohort revealed censorship rates of 18% and 19% for OS and DFS, which were suitable for subsequent survival analysis. In order to utilize the full extent of the survival data, we performed the survival analyses on the pooled data set as well as in the separate subsets. To prevent possible misclassification biases from patients who were operated upon expecting early-stage disease but actually having advanced disease, analyses of DFS excluded cases in which recurrence of the disease was diagnosed before 2 months had elapsed from surgery (i.e., DFS ≤2 months). Factors with P < 0.2 were subsequently included in a Cox multivariate hazard model and were used to assess the impact of IDO2-deficient genotypes on OS time and DFS time. The model was further optimized by sequential inclusion of statistically relevant factors (P ≥ 0.2) until achievement of a final optimal model fit (P ≤ 0.05). P values ≤0.05 were considered statistically significant. Statistical analysis was performed using the Statistical Package for Social Sciences (IBM SPSS, Ver.20, SPSS Inc.).

On the basis of evidence of frequent overexpression of IDO2 in human PDAC tumors (14), we hypothesized that IDO2 inactivation (i.e., loss-of-function IDO2 alleles) might limit the development of pancreatic cancer. KRAS mutations occur in over 90% of invasive PDAC and are considered to be an early oncogenic event (28). The transgenic mouse model Pdx-1-cre;LSL-Kras^G12D (KC), with an inducible oncogenic Kras allele that is activated in pancreatic progenitor cells, spontaneously develops pancreatic intraepithelial neoplasia (PanIN) with complete penetrance and PDAC with reduced penetrance (17). To investigate our hypothesis that IDO2 contributes to pancreatic tumor development in this autochthonous tumor setting, we introduced Ido2^−/− (Ido2-nullizygous) alleles into the KC strain through interbreeding (11). Cre-mediated expression of the mutant Kras allele resulted in small duct proliferation in the pancreas regardless of the Ido2 status, along with a decreased frequency of macrophages (CD45⁺ CD11b⁺ Gr1⁺ CD11c⁻ F4/80⁺) and an increased frequency of dendritic cells (CD45⁺ CD11b⁺ Gr1⁻ CD11c⁺; Fig. 1A and Supplementary Fig. S1). Increased frequencies of neutrophils/MDSCs (CD45⁺ CD11b⁺ Gr1⁺ CD11c⁻), and helper T cells (CD45⁺ Th1⁺ IgM⁻ CD4⁺ CD8⁻) were also associated with mutant Kras expression but were only significant in the mice that also lacked IDO2. No significant differences associated with either Kras or Ido2 status were observed in the frequency of cytotoxic T cells (CD45⁺ Th1⁺ IgM⁻ CD8⁺ CD4⁻) while the overall number of mature B cells (CD45⁺ Th1⁻ IgM⁺) was consistently too low to produce meaningful comparisons. Ductal adenocarcinomas were identified in Ido2^+/+ KC mice with an overall lifetime incidence of 28% compared with 15% in the Ido2^−/− KC mice (P < 0.08, Fig. 1B and C). Unexpectedly, the impact of Ido2 loss on PDAC development in this model was predominantly associated with females, with no ductal adenocarcinomas identified in the 24 female Ido2^−/− KC mice while the incidence in Ido2^+/+ KC females remained at 30% (P = 0.01, Fig. 1C). These observations provide genetic evidence that IDO2 can contribute to the progression of early-stage pancreatic precursor lesions to malignant carcinoma and suggest that IDO2's involvement in this process may be subject to some degree of sexual dimorphism. Similar findings of dimorphism in KC mice have been reported by Chang, et al. (29) showing that female KC mice on a high fat-high calorie diet were less likely to develop PDAC compared with male KC mice (0% vs. 44% and 33% vs. 50%, at 6 and 9 months, respectively). Similar to these findings in the KC experiment, sequencing of the R248W SNP in the TJHU cohort revealed a significant paucity of R248W homozygous cases (IDO2-deficient status) in females with sporadic PDAC compared with the CEU control population (OR, 0.19; CI 95%, 0.05–0.65; P < 0.01), correlating with the lack of Ido2^−/− murine females with PDAC. A pooled analysis combining the TJUH data set with the TCGA data set (n = 200), looking at all PDAC cases yielded similar results, showing a significant absence of females harboring a homozygous R248W (IDO2-deficient) genotype (OR, 0.35; CI 95%, 0.17–0.74; P < 0.01).

Having obtained genetic evidence implicating IDO2 in pancreatic cancer development, we next examined how IDO2 functional status might influence the responses of PDAC patients to treatment based on the occurrence of the two functionally disruptive SNPs within the coding region of the human IDO2 gene (16). For the purpose of this study, two independent data sets were analyzed, a publicly available data set from The Cancer Genome Atlas Research Network (TCGA-PAAD) and an internally generated data set collected under an IRB-approved study at the Thomas Jefferson University Hospital in Philadelphia (TJUH; refs. 14, 30). Distributions of both IDO2 coding region SNPs rs4503083 (Y359X) and rs10109853 (R248W) in the TCGA-PAAD set were within the normal range of predicted Hardy–Weinberg equilibrium frequencies (χ² test, P = NS), consistent with previously published data (14). In the TJUH patient data set, a statistically significant 2-fold increase in prevalence of the Y359X homozygous genotype was noted compared with the predicted Hardy–Weinberg frequencies (P < 0.05), with additional analysis using Fisher exact comparison supporting some deviation from equilibrium (P < 0.05). However, in a second comparison with the 1000 Genome CEU data set (Utah residents with Northern and Western European ancestry) as a normal control, we observed no significant deviation in distribution (30).

The TCGA data set includes histologic information on NLR for many of the resected tumors, which is categorized as a basic indicator of a protumorigenic inflammatory state. As IDO2 is implicated in immune regulation, the association of IDO2 SNP status with NLR status was interrogated. Pooled analysis of histologic immune data from the TCGA cohort and slides collected from the TJUH cohort (N = 99) showed that IDO2-deficient genetic status significantly correlated with decreased neutrophil infiltration and improved (lower) NLR scores (P = 0.047 and P = 0.034, respectively; Fig. 2). RNA expression data available from the TCGA cohort (N = 123) further corroborate these findings. IDO2 genetic status correlated with decreases in the imputed neutrophil to overall-lymphocytes ratio, neutrophil to T-cell ratio, and neutrophil to B-cell ratio, and an increase in expression of cytotoxic lymphocytes (P = 0.018, P = 0.037, P = 0.082, and P = 0.018, respectively; see Supplementary Fig. S2). These results suggest that the loss of IDO2 function is associated with an improved immune signature.

Given the genetic and histologic indications of IDO2 involvement in PDAC development, we explored whether an association could also be drawn between IDO2 genetic status and patient treatment outcomes (31). Due to the high censorship rate (71%) of the DFS data in the TCGA-PAAD cohort, we pooled these data with our TJUH cohort to perform survival analyses. In the overall PDAC patient pool, Kaplan–Meier analysis suggested a trend toward increased DFS in patients with IDO2-deficient status that did not reach a level of significance (median survival 20.3 ± 3.5 vs. 32.4 ± 9.9 months, Fig. 3A). This observation suggested the possibility of a wider differential among a specific subgroup of patients that is masked in the overall patient pool. Indeed, a subanalysis of microscopic tumor involvement (R1 resection margin cases, N = 61) revealed that IDO2-deficient status was significantly associated with a favorable prognosis (Fig. 3B, P = 0.004) with the IDO2-deficient group not reaching its median survival in 32 months of follow-up. No significant association was seen in related subanalyses based on nodal metastasis, tumor grade or tumor size (≤3 cm or >3 cm) in this cohort (Supplementary Fig. S3), however, a similar favorable trend was observed in IDO2-deficient patients with large tumors (Size >3 cm, P = 0.099). A multivariate Cox regression identified only resection margin status as a statistically significant independent risk factor (HR, 1.81; CI 95%, 1.10–2.97; P = 0.02) for DFS. IDO2-deficient status also showed a strong trend toward positive long-term DFS (HR, 0.64; CI 95%, 0.37–1.10; P = 0.10). Tumor size, nodal involvement, and tumor grade were not found to be significant. These observations encouraged further analysis of whether specific treatment regimens received by patients with positive R1 resection margins might be influenced by IDO2 genetic status.

In considering the possible mechanisms through which IDO2 status modifies survival in patients with positive resection margins, we investigated the use of adjuvant radiotherapy as a potential mediating factor (32). Reinforcing this exploratory logic was the finding of an association between IDO2-deficient status and reduced NLR, the latter of which has been identified by several studies as a positive prognostic factor for patients receiving radiotherapy for various cancers including PDAC (33). Thus, an aggregate evaluation of the patient data suggested the possibility that IDO2 status might be of particular relevance to patients who had received adjuvant radiotherapy during their course of treatment.

A total of 54 PDAC resection cases with DFS ≥2 months and including documentation of administered radiotherapy were available for evaluation in the pooled TCGA-PAAD and TJUH cohorts, together with a corresponding pooled cohort of 77 patients who did not receive radiotherapy. In comparing these two groups, there was no demonstrable improvement in DFS attributable to adjuvant radiotherapy (Fig. 4A), consistent with other studies that have reported a lack of benefit. However, inclusion of the IDO2 status in the survival analysis of the radiotherapy treatment cohort revealed a significant association of the IDO2-deficient genotype with mean DFS almost doubling in duration mean survival (39.0 ± 6.3 vs. 74.1 ± 6.4 months, Fig. 4B, P = 0.023) and over 50% survival during the follow-up time (median survival was not reached). Importantly, IDO2 SNP-based stratification of the cohort that did not receive radiotherapy showed no significant evidence of a DFS benefit associated with IDO2-deficient status (Fig. 4C). Elaborating further on the association of IDO2 host genetic status with radiotherapeutic impact, a Cox multivariate hazard analysis of DFS defined IDO2-deficient genotype to be an independent positive factor (HR, 0.39; 95% CI; 0.18–0.83; P = 0.015; Fig. 4C). In summary, we conclude from this retrospective analysis that IDO2 host genotype appears to be a significant predictor of DFS in PDAC patients who received adjuvant radiotherapy as part of their care.

Our results provide clear evidence linking IDO2 function to PDAC pathophysiology and therapeutic response. Locoregional degradation of tryptophan and accumulation of kynurenine catabolites have been broadly implicated in supporting cancer-promoting inflammation and immune escape (2, 3, 34). Unlike the IDO1 enzyme that has been the primary focus of attention, IDO2 is less widely expressed in human tumors but has been reported in gastric, colon, and renal cancers (35) as well as in PDAC where it appears to be widely overexpressed (14). Our genetic data from mice support the notion that IDO2 can play a contributory role in PDAC tumorigenesis. In contrast, in the case of familial PDAC, we have recently published evidence of increased risk being associated with an IDO2-deficient genotype status although the trend for sporadic PDAC went in the opposite direction (30). A possible explanation might include a double-edged model in which tumor initiation and tumor progression are located on either side of the IDO2 functionality spectrum. However, addressing the observed discrepancies between the mouse KC model, sporadic PDAC, and familial PDAC awaits detailed cellular and molecular interrogation of IDO2's various effects at the tumor cell, the microenvironment, and systematic immune system levels.

As a metabolic modifier of inflammation, IDO2 is likely to exert complex effects. However, it is notable that preclinical studies of the more extensively studied IDO1 enzyme have clearly established its role in mediating tumoral resistance to DNA-damaging chemotherapies and ionizing radiation which are immunogenic in nature (6, 7). Correspondingly, a recent clinical study of non–small cell lung cancer patients by Wang and colleagues reported that activity levels ascribed to IDO1 corresponded with responsiveness to radiotherapy (36). Specifically, their results showed significant correlations between lower kynurenine/tryptophan ratios assessed pre- and postradiotherapy and prolonged OS. Although no direct determination of the specific enzyme responsible for catabolizing tryptophan to kynurenine in these patients was provided in this study, prior evidence suggests that IDO2 is not likely to be directly responsible for this level of activity as it is enzymatically less active than either IDO1 or TDO and its loss has not been found to affect systemic kynurenine/tryptophan levels (37). However, IDO2 has been shown to enable the promotion of regulatory T-cell activation by IDO1 (11). Thus, the possibility that IDO2 may be indirectly affecting the ability of IDO1 to regulate T-cell function highlights the need to consider IDO2 genetic status in situations where IDO1 has been clinically implicated. Additionally, IDO2 has been shown to have a biological role in supporting pathogenic B-cell antibody production in autoimmune settings (9, 10). This finding is particularly noteworthy given that protumorigenic B cells, initially identified with squamous cell carcinomas of the vulva and the head and neck, have also been clinically associated with PDAC (18, 38), and recent preclinical studies have implicated B cells as contributing to PDAC development (18, 38). The striking absence of adenocarcinomas observed in female Ido2^−/− KC mice precluded our ability to evaluate intratumoral B cells in these animals relative to their more susceptible counterparts, while the extremely low levels of B cells detected in the pancreas made comparative analysis of B cells the local microenvironment untenable. However, expression of oncogenic Kras in the pancreas was associated with differences in frequencies of other local immune cell populations, which in several instances appeared to be more pronounced in the animals lacking IDO2. Nevertheless, the trend in these data were not sufficiently robust to rule out the possibility of other functional contributions of IDO2 in this setting, perhaps including nonimmune functions, that are yet to be elucidated. Additional studies will be needed to determine whether any of these associations are functionally relevant to the reduced incidence of PDAC associated with IDO2 loss in female mice. Elucidating the basis for PDAC resistance in this preclinical model may have direct translational relevance given the significant absence of female R248W (IDO2-deficient) SNP representation in our analysis of PDAC patients.

A major limitation of our study is the low number of PDAC samples evaluated. Still, our results may have important ramifications for PDAC treatment, most notably with regard to use of adjuvant radiotherapy where variable efficacy has been reported in patients. In the adjuvant setting, PDAC therapy has remained little changed for over the past two decades in relying mainly on 5FU or gemcitabine as monotherapies (39–41). Newer combinations such as gemcitabine/capecitabine have shown some promise but they produce higher rates of toxicity, limiting completion of treatment protocols for some patients (42). In the search for ways to improve standard of care, the benefits of adjuvant radiotherapy have been highly debated. Studies assessing the impact of radiotherapy on survival have produced results ranging from increased OS (43–45) to no response or even poorer response (46–48). In our pooled analysis of the TCGA-PAAD and TJUH cohorts, we identified a significant association between a functionally ablated host IDO2 genotype and a favorable response to radiotherapy. Given the significant natural variation in the IDO2 coding SNP genotypes among human populations (16), our findings offer a potentially incisive explanation for the large variability in efficacy reported for adjuvant radiotherapy. Thus, IDO2 genotyping may provide a biomarker to personalize this modality for PDAC patients. Further evaluation of the host IDO2 gene as a predictive biomarker is warranted to confirm and extend its utility in additional patient populations.

Finally, our results may also have implications with regard to the potential for developing IDO2 inhibitors for use in combination with radiotherapy to treat those PDAC patients harboring IDO2-active alleles. Some support for this general concept is provided by a recent pilot study in which stratifying brain metastasis patients for the IDO2-active genotype distinguished a positive trend in OS following whole brain radiotherapy together with coadministration of low-dose chloroquine, an indirect inhibitor of IDO2 but not IDO1 activity (31). Although most experimental agents in clinical trials at present are selective for IDO1, and no IDO2-specific enzyme inhibitors are yet available, the IDO pathway inhibitor indoximod (D-1MT) has a different mechanism of action that may encompass IDO2 blockade (2, 9, 16, 37, 49). Indeed, D,L-1MT administration has been reported to improve the efficacy of radiotherapy in mouse tumor models (7, 50). Thus, in theory, the IDO2 genotype has the potential to serve as a decisional biomarker for distinguishing between PDAC patients who are more likely to respond to adjuvant radiotherapy alone versus those patients who may benefit from the coordinated blockade of IDO2.

G.C. Prendergast holds ownership interest (including patents) in Incyte, Kyn Therapeutics, and New Link Genetics and is a consultant/advisory board member for Kyn Therapeutic and Plexxikon. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.J. Muller, J.B. DuHadaway, K. Nagatomo, H. Lavu, L. Mandik-Nayak, G.C. Prendergast, J.R. Brody

Development of methodology: A. Nevler, A.J. Muller, J.B. DuHadaway, H. Lavu, T. Villatoro, G.C. Prendergast

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Nevler, E. Sutanto-Ward, J.B. DuHadaway, K. Nagatomo, K. O'Hayer, J.A. Cozzitorto, T.P. Yeo, T. Villatoro

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Nevler, A.J. Muller, E. Sutanto-Ward, J.B. DuHadaway, K. Nagatomo, E. Londin, J.A. Cozzitorto, B.E. Leiby, L. Mandik-Nayak, J.M. Winter, C.J. Yeo, J.R. Brody

Writing, review, and/or revision of the manuscript: A. Nevler, A.J. Muller, E. Londin, H. Lavu, T.P. Yeo, B.E. Leiby, J.M. Winter, C.J. Yeo, G.C. Prendergast, J.R. Brody

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Sutanto-Ward, K. O'Hayer, H. Lavu, J.M. Winter, C.J. Yeo, J.R. Brody

Study supervision: A.J. Muller, J.R. Brody

Other (histologic analysis, photomicroscopy, composition of pathology related text, review of manuscript): M.T. Curtis

The work was supported by NIH grant R01 CA191191 (J.R. Brody, A.J. Muller, and G.C. Prendergast) with additional support from the Lankenau Medical Center Foundation and Main Line Health (G.C. Prendergast and A.J. Muller) and an American Cancer Society-IRG grant (H. Lavu and J.R. Brody). J.R. Brody is also supported by R01 CA212600 and a RAN grant from the AACR Pancreatic Cancer Action Network. G.C. Prendergast holds the Havens Chair in Biomedical Research at the Lankenau Institute for Medical Research. A. Nevler was supported by a Mary Halinski Fellowship and the P30 CA056036 SKCC Core Grant (TJU).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.