Prognostic Significance of POLE Proofreading Mutations in Endometrial Cancer

Background: Current risk stratification in endometrial cancer (EC) results in frequent over- and underuse of adjuvant therapy, and may be improved by novel biomarkers. We examined whether POLE proofreading mutations, recently reported in about 7% of ECs, predict prognosis. Methods: We performed targeted POLE sequencing in ECs from the PORTEC-1 and -2 trials (n = 788), and analyzed clinical outcome according to POLE status. We combined these results with those from three additional series (n = 628) by meta-analysis to generate multivariable-adjusted, pooled hazard ratios (HRs) for recurrence-free survival (RFS) and cancer-specific survival (CSS) of POLE-mutant ECs. All statistical tests were two-sided. Results: POLE mutations were detected in 48 of 788 (6.1%) ECs from PORTEC-1 and-2 and were associated with high tumor grade (P < .001). Women with POLE-mutant ECs had fewer recurrences (6.2% vs 14.1%) and EC deaths (2.3% vs 9.7%), though, in the total PORTEC cohort, differences in RFS and CSS were not statistically significant (multivariable-adjusted HR = 0.43, 95% CI = 0.13 to 1.37, P = .15; HR = 0.19, 95% CI = 0.03 to 1.44, P = .11 respectively). However, of 109 grade 3 tumors, 0 of 15 POLE-mutant ECs recurred, compared with 29 of 94 (30.9%) POLE wild-type cancers; reflected in statistically significantly greater RFS (multivariable-adjusted HR = 0.11, 95% CI = 0.001 to 0.84, P = .03). In the additional series, there were no EC-related events in any of 33 POLE-mutant ECs, resulting in a multivariable-adjusted, pooled HR of 0.33 for RFS (95% CI = 0.12 to 0.91, P = .03) and 0.26 for CSS (95% CI = 0.06 to 1.08, P = .06). Conclusion: POLE proofreading mutations predict favorable EC prognosis, independently of other clinicopathological variables, with the greatest effect seen in high-grade tumors. This novel biomarker may help to reduce overtreatment in EC.

article Endometrial cancer (EC) is the commonest gynecological malignancy in the Western world, and is rising in incidence because of increasing obesity and ageing of the population (1). Most cases (80%) are detected at an early stage (FIGO stage I) (2) because of early symptoms. The standard management of EC consists of hysterectomy and bilateral salpingo-oophorectomy with or without postoperative vaginal brachytherapy (VBT) or pelvic external beam radiotherapy (EBRT) depending on recurrence risk (3,4). Women with grade 3 cancers or advanced disease are increasingly treated with adjuvant chemotherapy (5,6), the role of which will be further defined by ongoing studies. Current risk stratification is based on both clinical (age) and pathologic factors (FIGO stage, tumor type, grade, and lymphovascular space invasion [LVSI]). Despite refinement in the use of postoperative treatment in EC over the last two decades, over-and underusage of adjuvant therapy remains a clinical problem. Approximately seven patients with stage I EC with risk factors need to receive VBT to prevent one recurrence, while 8% to 10% patients develop distant metastases that may have been prevented with adjuvant chemotherapy (3)(4)(5)(6). Consequently, the identification of molecular markers predictive of recurrence risk or treatment benefit beyond current clinicopathological factors would represent a major advance (7). While studies have investigated the prognostic significance of several molecular alterations involved in endometrial carcinogenesis, including microsatellite instability (MSI), PIK3CA and TP53 mutation (8,9), to date none have been incorporated into routine clinical practice.
We recently showed that germline variants in the exonuclease domain of the DNA polymerases POLE and POLD1 predispose to cancer, including EC, by impairing polymerase proofreading and greatly increasing the rate of base substitution mutations (10). We subsequently demonstrated that somatic POLE proofreading mutations are found in about 7% of sporadic ECs, where they strongly associate with high tumor grade (11). Similar findings have been reported in parallel by The Cancer Genome Atlas (TCGA), which also demonstrated that while POLE mutations were only found in endometrioid tumors, they were not inversely associated with TP53 mutation (12). These pathogenic POLE proofreading mutations, about 90% of which cluster in exons 9 and 13, localize to amino acids within, or close to, conserved motifs essential for proofreading function (11,12). In keeping with this, POLE proofreading-mutant ECs are ultramutated, with a base substitution mutation frequency among the highest in human tumors (13).
Different forms of genomic instability in cancers are known to be associated with clinicopathological features, including prognosis (14)(15)(16). Favorable outcome of women with POLEmutant ECs has been suggested (12), but only reached statistical significance when limited to analysis of grade 3 tumors in a recent report (17), and current evidence is insufficient to inform practice (18). In this study, we have analyzed associations between POLE proofreading mutations, recurrence-free and cancer-specific survival in two large, randomized controlled trials (PORTEC-1 and -2) (3,4) of early-stage (FIGO stage I), (high-) intermediate risk EC, with central pathology review and mature follow-up data, and in three additional smaller EC series (9,12,19).

PORTEC Study Details
Details of the PORTEC-1 and PORTEC-2 studies have been published previously (see the Supplementary Materials, available online) (3,4). PORTEC-1 compared pelvic EBRT with no additional treatment (NAT) in 715 women with intermediate  risk, stage I EC recruited between June 1990 and December 1997  (3). PORTEC-2 randomly assigned 427 women with high-intermediate risk stage I/IIA EC between May 2002 and September  2006 to either EBRT or VBT following surgery (4). The median (range) duration of follow-up was 159.6 (33.6-222) months in PORTEC-1 and 89 (18-122) months in PORTEC-2. The PORTEC study protocols were approved by the Dutch Cancer Society and by the medical ethics committees at participating centers. All patients provided written informed consent to study participation and treatment.

Additional EC Series
The Leuven (n = 187) (9), Zurich/Basel (n = 267) (19), and TCGA (n = 373) (12) series have also been reported previously (Supplementary Materials, available online). The Leuven and TCGA cohorts were collected prospectively, while Zurich/Basel cases were identified retrospectively. These sets included both endometrioid and nonendometrioid ECs (EECs and NEECs), and also included patients with stage III/IV disease (21.8-31.2%). Central pathology review was mandated in both TCGA and the Zurich/Basel cohorts, while the Leuven cases were reviewed by a single academic pathologist. Patients in the Leuven and Zurich/Basel cohorts were managed according to standard protocols, while treatment in the TCGA series was at the discretion of the attending physician: median (range) follow-up in each was 29 (1-184) months, 46 (1-173) months, and 28.7 (0.6-185.6) months, respectively. Follow-up data varied between series. RFS data were available for the Leuven and TCGA series, but not the Zurich/Basel set, while CSS data were available for the Leuven and the Zurich/Basel series, but absent from the TCGA study. Collection and analysis of the Leuven and Zurich/Basel series were approved by the scientific ethics committee from all centers (UZ Leuven Medical ethics committee and KEK-ZH-NR: 2010-0358, respectively). Ethical approval for anonymized tumor molecular analysis was granted by Oxfordshire Research Ethics Committee B (Approval No. 05\Q1605\66).

Demographic and Clinicopathological Variables
Baseline demographic and clinicopathological variables were treated as either categorical (eg, grade, stage, EEC vs NEEC) or continuous (age) as appropriate. All analyses were based on data from central pathology review.

Molecular Analysis
Tumor DNA was extracted from formalin-fixed, paraffin-embedded (FFPE) blocks from 434 (60.1%) and 398 (93.2%) ECs from the PORTEC-1 and -2 studies, respectively. There were no statistically significant differences between the biomarker and the total study populations in demographic/clinicopathological characteristics, treatment or survival. DNA was extracted from 187 fresh-frozen ECs in the Leuven set, 373 fresh-frozen ECs in the TCGA set, and 260 FFPE ECs in the Zurich/Basel set (Supplementary Methods, available online). The Leuven, Zurich/Basel, and TCGA biomarker cohorts were similar to the total series population in each case (Supplementary Methods, available online).

Statistical Analysis
Analyses performed and reported in this biomarker study are listed in Supplementary Table 2 (available online) in accordance with published guidelines (22,23). For analysis of the association of POLE mutation with outcome, our primary endpoint was recurrence-free survival (RFS), with secondary endpoints of cancerspecific survival (CSS), and overall survival (OS). In the PORTEC studies, RFS was defined as the time from random assignment to relapse, with censoring at last contact or death in case of no recurrence. CSS was measured as the time from random assignment to death from EC, with censoring at date of last contact or noncancer death. OS was measured as the time from random assignment to death from any cause, with censoring at date of last contact in patients still alive. The same criteria were used in the additional series, with the exception that survival measurements were from time of diagnosis. Survival curves were plotted using the Kaplan-Meier method and compared by the log-rank test. We used Cox proportional hazards models to calculate hazard ratios (HRs) for RFS, CSS, and OS of POLE-mutant ECs relative to POLE wild-type tumors by univariable analyses and, following adjustment for baseline characteristics and prognostic factors (age, tumor type, grade, LVSI, depth of myometrial invasion, and treatment), by multivariable analyses (see Supplementary Methods  and Supplementary Tables 3-6, available online). Proportionality of hazards in Cox models was confirmed by visual inspection of complementary log plots or by interaction terms of covariables and (log)time. In view of the similar patient populations, the limited number of cancer-related events in both PORTEC trials, and the modest frequency of POLE mutations, the PORTEC studies were combined for most analyses. For Cox regression analysis of the PORTEC grade 3 subset and the Leuven, Zurich/Basel, and TCGA series, we applied Firth's correction (24), owing to the absence of events in the POLE-mutant groups. For multivariable analysis of the additional series, we included disease stage as a covariable; although myometrial invasion, LVSI, and treatment were not included because of lack of data, we confirmed that omission of these variables from the PORTEC multivariable analyses did not alter estimates of RFS or CSS with POLE mutation (P = .93 and P = .87 respectively) (see Supplementary Methods, available online). Statistical analyses were performed using SPSS (SPSS Inc., Chicago, IL), Stata (StataCorp, College Station, TX) and R (http://www.r-project.org/). All P values were two-sided. A P value under .05 was considered statistically significant.

Patient Characteristics and POLE Proofreading Mutations in PORTEC Studies
Demographic and clinicopathological characteristics of the PORTEC study participants in whom tumor POLE sequencing was successful are shown in Table 1. The majority (about 98%) of cancers in both studies were EECs. POLE exon 9 and 13 proofreading mutations were detected in 48 of 788 (6.1%) tumors from the combined PORTEC studies (Table 1), with similar distribution across study arms (11,12). With the exception of two tumors harboring a germline polymorphism of uncertain pathogenicity (rs150032060; c.1282G>A, p.Ala428Thr), detected in tumorfree myometrium in both cases and excluded from subsequent analyses, all mutations were recurrent substitutions at somatic mutational hotspot codons (Supplementary Table 7, available online) known to cause ultramutation (11,12). Apart from one neuroendocrine tumor, all POLE mutations occurred in endometrioid ECs. Analysis of 48 available preoperative curettings identified POLE mutations in all five cases in which they were detected in the subsequent hysterectomy, resulting in 100% concordance. Compared with POLE wild-type ECs, POLE-mutant tumors occurred in younger women (median age 63.5 vs 68.5 years, P < .001, t test), and were more commonly grade 3 (31.3% vs 12.7%, P < .001, χ 2 test), though LVSI and deep (>50%) myometrial invasion were less frequent (0% vs 9.5%, P = .03, and 58.2% vs 71.9%, P = .045, respectively, χ 2 test) ( Table 1).

Clinical Outcome by POLE Proofreading Mutation in PORTEC Studies
We first examined the association of POLE proofreading mutation with EC recurrence. Three of 48 (6.2%) women with POLEmutant tumors developed local or distant recurrence during study follow-up, compared with 104 of the other 740 (14.1%) patients ( Figure 1A, Table 2). All three recurrences in the POLEmutant cohort were distant metastases without locoregional relapse and occurred in women with grade 1 EECs and deep myometrial invasion, managed by observation, EBRT, and VBT, respectively. The univariable HR for recurrence-free survival (RFS) with tumor POLE proofreading mutation was 0.41 (95% CI = 0.13 to 1.28, P = .13), with little change following adjustment for known prognostic variables by Cox regression (HR = 0.43, 95% CI = 0.13 to 1.37, P = .15) ( Figure 1A, Table 2; Supplementary  Tables 3 and 5, available online). As many local EC recurrences are salvageable with therapy, we next examined whether cancerspecific survival (CSS) varied according to tumor POLE mutation. In women with POLE-mutant EC, there was one (2.3%) EC death during follow-up, compared with 72 (9.7%) in the rest of the study population. The unadjusted HR for CSS was 0.20 (95% CI = 0.03 to 1.46, P = .11), with minimal alteration following multivariable analysis (HR = 0.19, 95% CI = 0.03 to 1.44, P = .11) ( Figure 1B, Table 2; Supplementary Tables 4 and 6, available online). Overall survival of women with POLE-mutant ECs was not statistically significantly greater than that of other patients by univariable or multivariable analysis (10-year OS = 76.2% vs 70.4%) ( Table 2).
The strong association of POLE mutations with high tumor grade (11) caused us to hypothesize that their apparent prognostic effect would be most evident in this group, who are commonly considered for treatment intensification. Of 109 patients with grade 3 tumors, there were no recurrences or cancer deaths in the 15 (13.7%) POLE-mutant case, compared with 29 (30.9%) relapses and 25 (26.6%) EC deaths in the remaining 94 women (Figure 2, A  and B), reflected in statistically significantly improved RFS in univariable analysis (HR = 0.09, 95% CI = 0.001 to 0.66, P = .01) and, following adjustment for other prognostic variables in multivariable analysis (HR = 0.11, 95% CI = 0.001 to 0.84, P = .03) ( Table 3;  Supplementary Tables 8 and 9, available online). These results were essentially unchanged after limiting RFS analysis to the 97 grade 3 endometrioid ECs (univariable HR = 0.11, 95% CI = 0.001 to 0.78, P = .02; multivariable-adjusted HR = 0.12, 95% CI = 0.001 .03  Table 8, available online). Notably, POLE proofreading mutation was a stronger predictor of recurrence and EC death than all other prognostic variables examined in these analyses (Supplementary Tables 8 and 9, available online).

Pooled Analysis With Additional EC Series
While the PORTEC analyses demonstrated a tendency for POLE proofreading-mutant ECs of all grades to improved outcome, the generally good prognosis of patients limited our ability to confirm this beyond the grade 3 subgroup. We therefore sought to support our results by analysis of two EC series from Leuven and Zurich/Basel, together with the published TCGA set, collectively comprising an additional 628 patients. POLE proofreading mutations were detected in 33 (5.3%) case patients in the additional series (Supplementary Table 7, available online). As in the PORTEC studies, POLE-mutant tumors were more frequently high-grade (54.5% vs 32.4%, P = .013, χ 2    Figures 1B and 3, Supplementary Tables 13 and 15, available online). As anticipated given their sizes, no individual series demonstrated statistically significant differences in the RFS/CSS of POLE-mutant ECs by either univariable or multivariable analyses (Supplementary Tables 13-15, available online). We therefore combined multivariable-adjusted HRs from each series with those from the PORTEC studies by meta-analysis to generate pooled estimates for RFS and CSS according to POLE proofreading mutation (Figure 3, A and B). POLE-mutant ECs were associated with statistically significantly greater RFS compared with other tumors (HR = 0.33, 95% CI = 0.12 to 0.91, P = .03), with no evidence of heterogeneity among studies (P = .66), though the difference in CSS was not statistically significant using a two-tailed test (HR = 0.26, 95% CI = 0.06 to 1.08, P = .06). Interestingly, the only POLE-mutant tumors that recurred were early-stage, grade I ECs from the PORTEC cohorts (Supplementary Table 16, available online). To date, most POLE proofreading mutations have been detected in EECs (11,12,17), although POLE-mutant NEECs have been reported (11,25,26). Given the difficulty in histotyping high-grade ECs, we considered the possibility that the four POLE-mutant tumors reported as NEECs in the Leuven series may have been misclassified, thus biasing our results because of the poor prognosis of NEEC. We therefore confirmed that the estimate of RFS was essentially unchanged by repeating the meta-analysis after excluding POLE wild-type NEECs from all cohorts (HR = 0.34, 95% CI = 0.12 to 0.93, P = .04) (Supplementary Tables 5, 6, 13, and 14, available online).

Discussion
By analyzing tumors from nearly 800 women in two large, randomly assigned controlled trials and pooling data with three independent series comprising over 600 additional patients, we have demonstrated that POLE proofreading-mutant ECs of all grades display excellent prognosis, independent of other known prognostic factors. POLE-mutant tumors have a risk of recurrence approximately one third that of other ECs, and a relative risk of cancer death that appears even lower, though the latter finding was not statistically significant in our analysis. This is despite a strong association of POLE proofreading mutations with high tumor grade-a characteristic that predicts a high risk of metastases in early EC (27). Indeed, in the PORTEC studies, no POLE-mutant grade 3 EC recurred, compared with 30.9% of grade 3 tumors in the rest of this subgroup, consistent with a recent report, which combined high-grade ECs from the TCGA series with an additional, small retrospective series (17). The role of adjuvant chemotherapy in patients with high-risk grade 3 tumors is currently under investigation, and as systemic therapy was not used in PORTEC-1 or -2 these data suggest that patients with POLE proofreading-mutant EC may be unlikely to benefit from such treatment. Similarly, the absence of locoregional recurrence of POLE-mutant ECs was observed across all arms of the PORTEC studies, including the NAT arm of PORTEC-1. While further studies are required before firm conclusions can be drawn on the implications of POLE proofreading mutations for postoperative treatment in EC, our results suggest that minimization of adjuvant therapy for POLE-mutant ECs localized to the uterus may be worthy of investigation. They also suggest POLE proofreading mutation should be considered for inclusion in cancer gene panels used in EC, as it may improve prognostication, particularly for grade 3 tumors.
While highlighting the strengths of using high-quality clinical trial sample banks for biomarker research (18), our study has limitations. The challenge of confirming even a strong effect of a marker with modest (<10%) frequency in a population with relatively favorable prognosis meant that we used additional EC series to confirm the improved RFS of POLE-mutant ECs. While acknowledging the limitations of sets containing a mixture of histological subtypes and stages, with limited follow-up and lacking comprehensive treatment data (18) the similar results from each following multivariable analysis suggests that our results are unlikely to be because of chance or confounding by inclusion of NEECs in our analyses. However, as highlighted above, it will be important to confirm the favorable prognosis of POLE-mutant ECs in further independent series, and particularly in tumors of advanced stages. It should also be noted that article as a single biomarker survey of POLE hotspot exons in ECs of predominantly endometrioid histology, we are presently unable to determine the effect of the about 10% of pathogenic variants outside exons 9 and 13, (11,12) or whether the effect of POLE mutation varies according to tumor molecular subtypes. Both questions are likely to be addressed by future studies. POLE proofreading-mutant cancers are a molecularly distinct group of tumors with a striking mutation burden and distinctive mutation signature (11)(12)(13). Whether these characteristics contribute to their favorable prognosis awaits confirmation. Study of mutator polymerases in yeast has confirmed the existence of a mutational threshold, which, if exceeded, results in decreased viability because of lethal mutations in essential genes prior to cell division (28). It will be of interest to determine whether the dramatic increase in mutation rate in S. cerevisiae caused by the corresponding substitution to human POLE p.Pro286Arg (29) approaches this error threshold. Similarly, analysis of the burden of deleterious mutations accumulated in POLE proofreading-mutant cancers may provide insights into their behavior (30).
Over the last two years, we and others have shown that somatic mutations in the proofreading domain of POLE occur in several human tumors (10)(11)(12)20,31). We now demonstrate that, despite a strong association with high grade, POLE proofreadingmutant ECs have a favorable prognosis. While the frequency of POLE mutation in EC is modest, it is worth noting that it is broadly similar to that of many novel molecular aberrations recently discovered by TCGA and other sequencing efforts. As most common cancer variants are currently not actionable, similar analyses of these modest-frequency (5% to 10%) molecular subgroups are likely to be essential if we are to realize the ambition of personalized cancer medicine during the next decade.

Notes
The funders had no role in the design, analysis, interpretation, or manuscript preparation for this biomarker study.
The authors would like to thank the many individuals who participated in the PORTEC studies and the numerous institutions and their staff who have supported recruitment. We also thank Michelle Osse, Enno Dreef, Martina Storz, Susanne Dettwiler, Annette Bohnert, Chrissie Fieger, Peggy Tzscheetzsch, and Sonja Brun-Schmid for excellent technical assistance, the Tissue Biobank of the University Hospital Zurich for providing samples and Holger Moch and Rosmarie  article Caduff for constructing the tissue microarrays for the Zurich/ Basel series, and the patients who contributed samples. We thank the research staff of the Vesalius Research Center, FWO-Flanders where Frédéric Amant is a senior scientist, and the patients who provided samples for the Leuven series. We are also grateful to Victoria Strauss (Centre for Statistics in Medicine, Oxford) for statistical advice. The results published here include data generated by The Cancer Genome Atlas project funded by the National Cancer Institute and the National Institutes of Health (cancergenome.nih.gov); we would like to thank the many patients and researchers who contributed to this study. The authors have no conflicts of interest to declare.