High IDO1 Expression Is Associated with Poor Outcome in Patients with Anal Cancer Treated with Definitive Chemoradiotherapy

Despite increased incidence of anal squamous cell carcinoma (ASCC), treatment recommendations have remained unchanged for the past 35 years. This article profiles the tumor microenvironment of patients with localized ASCC, examining CD8, PD‐1, PD‐L1, IDO1 and HLA class I expression and, specifically, characterizes expression of IDO1 in the context of several key components of the immune microenvironment.


INTRODUCTION
Anal squamous cell carcinoma (ASCC) is a relatively rare malignancy with about 27,000 new cases per year worldwide [1]. The incidence has steadily increased over the last few decades, secondary to exposure to oncogenic subtypes of the human papilloma virus (HPV), the major driver of ASCC in >90% of patients [2][3][4][5]. Despite this increase in incidence, the treatment paradigm for locally advanced ASCC has been essentially unchanged for the past 35 years, when definitive chemoradiation was first found to be an effective alternative to abdominoperineal resection [6]. This standard regimen of 5-fluorouracil (5FU) and mitomycin-C (MMC) with concurrent radiation results in good outcomes for many patients, but despite best practices, about 25% of patients with advanced T-or N-stage cancer will not be cured of their disease [7][8][9].
Identifying novel biomarkers that are associated with poor prognosis will help define the cohort of patients who merit treatment escalation. Several studies have found that HPVnegative tumors have worse outcomes, but the role of other immunohistochemical or genomic markers is less well established [10][11][12][13]. Given the viral etiology of this disease, it is perhaps not surprising that prior studies have shown that the immune microenvironment in ASCC plays an important role in disease prognosis. Tumor-infiltrating lymphocytes (TILs) are the most frequently studied component of the immune microenvironment, with most studies finding higher TIL to be associated with improved outcomes [14][15][16][17]. Other immunologic factors are less well studied in ASCC. Several groups have examined the effect of programmed cell death protein 1 (PD-1)/ programmed death-ligand 1 (PD-L1) expression on prognosis but have come to different conclusions regarding whether higher expression of these immune checkpoints is negatively or positively prognostic [18,19]. However, despite these differing results, there is an emerging consensus that an immunosuppressive tumor microenvironment is likely to be associated with poor outcomes.
A variety of novel immunomodulatory therapies have shown that inhibiting immunosuppressive signaling can improve clinical outcomes. The most clinically developed of these agents are immune checkpoint inhibitors, such as anti-PD-1/ PD-L1 drugs, which have led to exciting new treatment options for a wide variety of metastatic or recurrent malignancies, including HPV-related malignancies such as head and neck cancer, cervical cancer, and ASCC [20][21][22]. Although a subset of patients will have dramatic and often durable responses, fewer than 20% of patients in these studies typically respond to treatment. Predicting which patients will respond to immune checkpoint therapy is challenging. Although PD-L1 expression has been associated with anti-PD-1/PD-L1 response in a variety of settings, several studies have shown a lack of association between PD-L1 expression and response to PD-1 inhibition [23]. A variety of other markers are under investigation.
The current study aims to profile the tumor microenvironment of patients with localized ASCC by examining cluster of differentiation 8 (CD8), PD-1, PD-L1, IDO1, and human leukocyte antigen (HLA) class I expression. This is the first study, to our knowledge, to specifically examine IDO1 in ASCC and demonstrate that patients with tumors expressing high IDO1 levels have significantly worse outcomes. More broadly, this study also characterizes expression of IDO1 in the context of several key components of the immune microenvironment and suggests the existence of several ASCC-microenvironment subtypes, each of which have unique drivers of immune evasion.

Patient Cohort
With Institutional Review Board approval, we reviewed the medical records of all patients with nonmetastatic ASCC treated at Massachusetts General Hospital (MGH) Cancer Center and Brigham & Women's Hospital/Dana-Farber Cancer Institute (BWH/DFCI) who met the following criteria: (a) treatment with definitive chemoradiation between 2005 and 2016, with (b) ≥3 months follow-up from start of radiation (if alive at last follow-up), and (c) accessible formalin-fixed paraffinembedded (FFPE) material with sufficient tumor for further analysis.

Clinical Parameters Collected and Treatment Details
The medical record was retrospectively reviewed for each patient to collect pretreatment clinical characteristics (including age at diagnosis, sex, performance status, smoking history, and immunosuppression), disease characteristics (including stage and grade), and treatment characteristics (including surgery, chemotherapy, and radiation). Patients were followed for clinical complete response, and development of localregional recurrence (defined as recurrence within the pelvis) and distant metastases (defined as recurrence outside the pelvis). Date of death or last follow-up alive was also recorded.
Kaplan-Meier analyses were used to evaluate overall survival (OS) and time to recurrence (any, local, and distant). Patients without an event were censored at the date of last follow-up. To look for associations between immune marker expression and outcomes, we set expression cutoff values to dichotomize the data at the 25th, 50th, 75th, and 90th percentile of each marker's distribution. Each cutoff was tested for association with risk of death, any recurrence, locoregional recurrence (LRR), or distant metastases (DM). Because of multiple testing, a Bonferroni correction was applied such that p < .0125 was considered a statistically significant association with outcome. The cutoff with the lowest p value was selected for further analysis.
The log-rank test was used to compare time from diagnosis to event or censoring. Cox proportional hazards models were used to evaluate associations between outcomes and immune markers, controlling for patient and tumor characteristics. With 63 patients, this study had 80% power to detect hazard ratios for survival ranging from 3.8 (using the 50th percentile as the marker's cutoff) to 6.3 (using the 90th percentile), corresponding to decreases in 3-year survival from 85% (the background rate in this patient population) to levels ranging from 54% to 36%, respectively, based on a two-sided log-rank test at the .05 significance level.

Patient, Disease, and Treatment Characteristics
Sixty-three patients treated for nonmetastatic ASCC between 2005 and 2016 with FFPE tissue available constituted the cohort for analysis. This included 48 patients from MGH and 15 patients from BWH/DFCI. The patient, disease, and treatment characteristics of the cohort are shown in Table 1. The median age of patients in the study was 61 years (range 33-92), with 59% being female and 92% having Eastern Cooperative Oncology Group performance status 0-1. Fiftytwo percent of patients were prior or current smokers, and 30% had a potential source of immunosuppression including human immunodeficiency virus diagnosis or being on immunosuppressive medication (most commonly after organ transplant). The distribution of disease stages was I-11% (7 patients), II-38% (24 patients), IIIA-14% (9 patients), IIIB-33% (21 patients), and unknown-3% (2 patients). HPV status was determined by ISH for 38 patients, with 37 and 1 patients having HPV-positive and -negative ASCC, respectively.
Because of obstructive symptoms, 14% of patients had upfront surgical diversion with colostomy followed by chemoradiation. The remaining patients received definitive chemoradiation. Eighty-three percent of patients were treated as per Radiation Therapy Oncology Group (RTOG) 0529 with dosepainted intensity-modulated radiation therapy to 50.4 Gy/42 Gy for T2N0 disease and 54 Gy/50.4 Gy/45 Gy for T3-4N0-3 disease [39]. Seventeen percent of patients were treated with a 3D-conformal successive cone down technique, but

Disease Outcomes
Median follow-up from initiation of radiation was 35 months (range excluding two patients who died on treatment: 5-145 months). Overall outcomes are shown in supplemental online Table 1. Because 3 patients died prior to follow-up assessment and 1 patient had an upfront abdominoperineal resection (apr), 59 patients were evaluable for clinical response to chemoradiation. Ninety percent of evaluable patients had a complete clinical response. Of the six patients who had residual disease, four had subsequent abdominoperineal resection and the remaining two developed metastatic disease and did not have local surgery. Of 61 patients who were alive at the end of treatment, 16 patients developed disease recurrence, including 7 patients with isolated LRR. However, despite having no documentation of DM, four of these seven patients with isolated LRR died of ASCC (secondary to bleeding or failure to thrive) and three of these patients were salvaged with surgery and alive without disease at last follow-up. Twenty-nine percent of patients died during follow-up, including 2 patients who died on treatment, 5 patients who died during follow-up of causes thought to be unrelated to ASCC, and 11 patients who died of ASCC.

Immune Marker Expression
Examples of immune marker staining in two patients with high IDO1 are shown in Figure 1. Expression of the five immune markers examined varied substantially across the 63 patients in the studied cohort ( Fig. 2A, 2B). Each marker had the following median (med) and interquartile range Fifty-seven percent of patients (n = 36) had at least 1% IDO1+ tumor cells and 10% (n = 6) had a high level of staining, defined as >50% tumor cells being positive. Macrophages were identified by morphology and found to be IDO1+ in 52% of patients (n = 33, median IDO+ macrophages/hpf: 5.8 with IQR 0-21). Forty-one percent of patients had >1% tumor cells that were PD-L1 positive, and one patient had >50% PD-L1-positive tumor cells. Nineteen percent of patients had <50% of tumor cells stain positive for HLA class I, with five patients (8%) having no staining for class I HLA.
Hierarchical clustering of immune marker expression suggested four broad subgroups (Fig. 2C). One subgroup largely consisted of patients with high IDO1 expression and tended to have lower CD8 infiltrate. Correspondingly, the tumors with >50% IDO1 expression were significantly more likely to have the lowest quartile of CD8+ TIL (<40/hpf), with 67% (4/6) of IDO1-high tumors having low CD8+ TIL compared with 21% (12/57) of IDO1-low tumors having low CD8+ TIL (p = .024). A second subgroup consisted of patients with low HLA class I expression. Although CD8+ TIL levels in this subgroup varied, there was significant positive correlation between CD8+ TIL and PD-1+ immune cells (R 2 = 0.40, p = .0083). A third subgroup consisted of tumors with high PD-L1 levels, with a subset of these tumors having high CD8+ TILs. Several patients in this subgroup also had IDO1+ tumors (although <50% positive). The fourth subgroup had a heterogeneous immune microenvironment including a subset with high immune cell PD-1 and a subset with high CD8+ TIL. Overall, patients with higher CD8+ TIL tended to have higher PD-1 (R 2 = 0.47, p = .0006) as well as tumor PD-L1 (R 2 = 0.36, p = .04).

Association Between IHC Features and Clinical Outcomes
As shown in Table 2, among various patient and disease characteristics, only advanced T stage and lymph node positivity were associated with increased risk of death or recurrence on univariate analysis, consistent with prior published results [9]. T3-T4 (vs. T1-T2) disease was specifically associated with a significantly increased risk of any recurrence or LRR (hazard ratio [HR] 3.41, p = .036 and HR 5.56, p = .029, respectively) and showed a trend for association with worse OS (HR 2.58, p = .060). The presence of clinically positive lymph nodes was associated with worse OS (HR 3.35, p = .022) and showed a trend for increased risk of DM (HR 4.39, p = .065). These results are consistent with previously published prospective studies of ASCC. Sex, age, smoking status, and immunosuppression were not significantly associated with outcome.
Of note, no other immune marker was found to be significantly associated with any outcome, including the presence of IDO1+ macrophages. Similarly, when clinical outcomes across the four subgroups defined by hierarchical clustering were compared, the subgroup defined by high IDO1 continued to have significantly worse prognosis without a significant difference among the other subgroups (supplemental online Fig. 1).

DISCUSSION
The current study is the first, to our knowledge, to show that high IDO1 tumor cell expression is associated with poor outcomes for patients with ASCC. IDO1 expression in tumor cells was detectable by IHC in 57% of cases, and IDO1 expression in macrophages was present in 52% of cases; however, only tumor cell IDO1 expression was associated with outcomes. Among the 10% of patients with high (≥50%) tumor IDO1 expression, there was a 4.7-fold increased risk of death and associated decrease in 3-year OS from 88% to 25% (p = .0074). Correspondingly, there was a 6.4-fold increased risk of any recurrence including an 8.7-fold increased risk of LRR and 12.7-fold increased risk of DM. IDO1 is commonly expressed in tumors and has been shown to play a role in resisting immune rejection [24]. More recently, data from The Cancer Genome Atlas have revealed that IDO1 is expressed in a wide variety of malignancies [40]. Like ASCC, cervical cancer and head and neck squamous cell carcinoma are commonly associated with HPV infection and have a subset of cases with high IDO1 expression [40]. Several retrospective studies specifically examining cervical cancer outcomes have confirmed that patients with high tumor IDO1 expression by IHC or high kynurenine levels, indicating a high degree of tryptophan metabolism, are associated with poor outcomes including worse OS in one study and worse disease-specific survival in a second study [32,33].
The mechanism by which IDO1 produces poor clinical outcomes is incompletely understood but may be driven by the depletion of tryptophan and accumulation of kynureninepathway metabolites in the tumor microenvironment. Several groups have shown that tryptophan depletion causes an accumulation of uncharged transfer RNA in T cells, which signals amino acid insufficiency and acts on downstream signaling pathways including activation of the stress response kinase GCN2. This kinase inhibits the translation initiation factor eIF2α and suppresses the mammalian target of rapamycin complex 1/PKCθ growth pathway [41,42]. Ultimately, these signaling pathways prevent T-cell activation and promote de novo regulatory T-cell (Treg) differentiation while also enhancing existing Treg activity. Together, these effects act to create a profoundly immunosuppressive environment.
Several groups have found an association between TIL and outcome in ASCC that was not seen in the current study. Reasons for this discrepancy include differences in the underlying patient population, the methodology used to quantify T cells, and the cutoffs used to define patients with high versus low expression. Specifically, Gilbert et al. quantified lymphocytes on hematoxylin and eosin-stained slides using low-power magnification with a pathologist assessing the degree of infiltrate as low, medium, or high; Hu et al. quantified intratumoral CD8+ cells by IHC examining three high-power fields and setting an average of >10 cells as high; and Grabenbauer et al. performed IHC for CD8 (as well as CD3 and CD4) and used image analysis software to quantify the number of tumor-infiltrating cells per 100 tumor cells with >2.1 intratumoral CD8+ cells per 100 tumor cells considered high. These varying methods of TIL detection, quantification, and dichotomization into highversus low-expression groups differ as well from the approach used in the current study of quantifying intratumoral CD8+ hot spots. It may be that more comprehensive tumor evaluation rather than evaluating the area of maximum signal is a better proxy for total tumor TIL. However, it is not clear whether total tumor TIL versus area of maximal tumor TIL is a better marker of overall tumor-immune recognition.
Several studies have examined the association between immune markers, such as PD-1 and PD-L1, and outcomes in ASCC. One study showed that tumors with high PD-1 (in addition to high CD8+ TIL) had better local control and disease-free survival, with high PD-L1 specifically correlated with better local control [18]. However, a different group has shown presence of PD-L1 on tumor cells to be inversely correlated with CD8+ TIL such that high PD-L1 is negatively prognostic in terms of response to standard therapy [19]. Although our results did not show these markers to be individually prognostic, it is interesting that IDO1 was robustly associated with poor outcome and was associated with the lowest quartile of  CD8+ TIL, which fits in the general paradigm of low TIL being negatively prognostic. Similarly, the current study also showed a statistically significant association between the number of CD8+ TIL and the number of PD-1+ cells as well as the percentage of tumor cells that are PD-L1+, which may suggest that a subset of ASCC tumors in our study represents the exhausted phenotype described by Balermpas et al., which can be activated by chemoradiation to help contribute to good outcomes [18]. Ultimately, in describing the tumor microenvironment, we hoped to gain insight into potential avenues for therapeutic intervention that may improve outcomes for the subset of patients who are not cured by upfront chemoradiation. A phase II study of nivolumab in metastatic ASCC showed a 24% response rate to anti-PD-1-directed therapy, which serves as proof of principle that immunotherapy can have an important role in this disease [22]. Responders were found to have higher CD8, PD-1, and PD-L1 levels.
Given that our study has shown the dismal prognosis of patients with high tumor IDO1 levels, as well as the association of high IDO1 level with the lowest quartile of CD8+ TIL, it seems likely that anti-PD-1 therapy may not be sufficient in treating patients with ASCC with high IDO1. Rather, this study suggests that IDO1-targeted therapies may have an important role in IDO1-high ASCC, either in the metastatic setting or, given the very short time to disease recurrence in patients with IDO1-high primary tumors, in the adjuvant setting after definitive chemoradiation. Future clinical studies will be needed to examine this question with prospective collection of tissue and/or blood before and after standard therapy and any potential immune targeted therapy.

CONCLUSION
ASCC, although near-universally HPV related, is a disease with a diverse immune milieu. Although patients generally do well with standard therapy, those who have tumors with high IDO1 expression have significantly worse survival and increased risk of disease recurrence. Tumors with high IDO1 expression also tend to have the lowest quartile of CD8+ TIL, suggesting that high IDO1 expression can drive an immunosuppressive microenvironment, which may be responsible for the resulting poor prognosis. Collectively, this study suggests that IDO1 may serve as a prognostic indicator while also helping identify a patient population who could benefit from IDO-targeted therapies.