Clinicopathological and molecular characterisation of ‘multiple‐classifier’ endometrial carcinomas

Abstract Endometrial carcinoma (EC) molecular classification based on four molecular subclasses identified in The Cancer Genome Atlas (TCGA) has gained relevance in recent years due to its prognostic utility and potential to predict benefit from adjuvant treatment. While most ECs can be classified based on a single classifier (POLE exonuclease domain mutations – POLEmut, MMR deficiency – MMRd, p53 abnormal – p53abn), a small but clinically relevant group of tumours harbour more than one molecular classifying feature and are referred to as ‘multiple‐classifier’ ECs. We aimed to describe the clinicopathological and molecular features of multiple‐classifier ECs with abnormal p53 (p53abn). Within a cohort of 3518 molecularly profiled ECs, 107 (3%) tumours displayed p53abn in addition to another classifier(s), including 64 with MMRd (MMRd–p53abn), 31 with POLEmut (POLEmut–p53abn), and 12 with all three aberrations (MMRd–POLEmut–p53abn). MMRd–p53abn ECs and POLEmut–p53abn ECs were mostly grade 3 endometrioid ECs, early stage, and frequently showed morphological features characteristic of MMRd or POLEmut ECs. 18/28 (60%) MMRd–p53abn ECs and 7/15 (46.7%) POLEmut–p53abn ECs showed subclonal p53 overexpression, suggesting that TP53 mutation was a secondary event acquired during tumour progression. Hierarchical clustering of TCGA ECs by single nucleotide variant (SNV) type and somatic copy number alterations (SCNAs) revealed that MMRd–p53abn tumours mostly clustered with single‐classifier MMRd tumours (20/23) rather than single‐classifier p53abn tumours (3/23), while POLEmut–p53abn tumours mostly clustered with single‐classifier POLEmut tumours (12/13) and seldom with single‐classifier p53abn tumours (1/13) (both p ≤ 0.001, chi‐squared test). Finally, the clinical outcome of patients with MMRd–p53abn and POLEmut–p53abn ECs [stage I 5‐year recurrence‐free survival (RFS) of 92.2% and 94.1%, respectively] was significantly different from single‐classifier p53abn EC (stage I RFS 70.8%, p = 0.024 and p = 0.050, respectively). Our results support the classification of MMRd–p53abn EC as MMRd and POLEmut–p53abn EC as POLEmut. © 2019 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.


Introduction
Of the many advances in the field of endometrial cancer (EC) during the last decade, perhaps the one with most impact is the molecular classification proposed by The Cancer Genome Atlas (TCGA) [1], which has gained prominence in recent years [2]. This classifies ECs into four molecular subtypes -POLE/ultramutated (POLE), microsatellite instability-high/hypermutated (MSI), somatic copy-number alteration high/serous-like (SCNA-high), and somatic copy-number alteration low (SCNA-low) -with significantly different prognoses [3], and thus of potential clinical relevance. For example, the favourable outcome of POLE (exonuclease domain) mutant EC, independent of adjuvant treatment [4], has led to proposals to de-escalate post-operative therapy in this subgroup [4]. In contrast, the consistently poor clinical outcome of patients with SCNA-high (serous-like) ECs suggests that intensification of adjuvant treatment may be worthwhile. The reported response of MSI tumours to immune checkpoint inhibitors [5,6], as well as the possibility of a better response to adjuvant radiotherapy [7], also opens new treatment opportunities for this group of patients. A randomised controlled clinical trial (PORTEC4a [8]) is currently testing the added value of integrating this molecular approach into risk assessment, to individualise adjuvant treatment and reduce over-and under-treatment in patients [9].
One important contributor to the impact and rapidly increasing adoption of the TCGA classifier is the relative ease in which subgroups analogous to those originally described can be identified by techniques in routine clinical practice [9][10][11][12][13][14]. ECs analogous to the SCNA-high subclass can be identified by p53 immunohistochemistry (p53abn EC) [15], the MSI subclass by immunohistochemistry for MMR proteins (MMRd EC), and the POLE subclass by targeted sequencing of the POLE exonuclease domain (POLEmut EC) (the latter being the most difficult to implement in routine practice). Tumours lacking these three prior features are classified as p53 wild-type (p53wt EC) [11] or no specific molecular profile subtype (NSMP EC) [10], analogous to the SCNA-low subclass.
Using this surrogate marker approach, most ECs can be classified into a single molecular class (henceforward referred to as 'single-classifier' ECs). However, 3-6% of tumours have more than one molecular classifying feature (hereafter referred to as 'multiple-classifier' ECs) [9][10][11][12], and include those with combined POLE exonuclease domain mutation (EDM) and abnormal p53 (POLEmut-p53abn), combined DNA mismatch repair deficiency (MMRd) and abnormal p53 (MMRd-p53abn), combined MMRd and POLE EDM (MMRd-POLEmut), and all three defects (MMRd-POLEmut-p53abn) [9][10][11][12]. There is currently no consensus on how these tumours should be classified or treated; some studies have excluded them from further analysis [10], while others have allocated them to one of the four subtypes [11] without detailing their clinicopathological or molecular features. Consequently, the biology and prognostic significance of multiple-classifier ECs are unclear, and how they should be managed is unanswered. This creates an important problem for tumours that carry opposite features, as one may favour treatment de-escalation (POLEmut) and the other intensified treatment (p53abn).
In this study, we aimed to perform a comprehensive analysis of the clinical, morphological, and molecular characteristics of EC with abnormal p53 immunostaining and/or TP53 mutation in combination with POLEmut and/or MMRd in order to inform clinical management of these multiple-classifier ECs.

Patient and tissue selection
Patient identity was protected by study-specific patient numbers. Informed consent and ethical approval were obtained according to the local protocol in each participating centre.
We obtained patient data and tumour tissue from ECs with more than one classifying feature (a pathogenic POLE exonuclease domain variant, MMR protein loss of expression or p53 abnormal expression) from previously published datasets [1,10,12,13,16,17]. A total of 2988 ECs had been molecularly classified in previous studies [10,12,13,16,17], in which MMR and p53 status were determined by immunohistochemistry (IHC) and POLE variants by Sanger sequencing or NGS of the complete exonuclease domain (exons [9][10][11][12][13][14] or targeted sequencing of exons 9, 13, and 14. For this current study, all POLE variants were reviewed and tumours were excluded when the POLE variant was not considered pathogenic following the recommended approach presented by León-Castillo et al [18]. Additionally, following the criteria mentioned previously, four tumours reported as having all three molecular features (POLEmut-MMRd-p53abn) but with a non-pathogenic POLE EDM based on review were classified as MMRd-p53abn ECs. ECs with MMR loss and a pathogenic POLE variant (MMRd-POLEmut ECs) were excluded from this study, as they are described by León-Castillo et al [18]. Clinical follow-up and, when available, slides were centrally collected for further analyses. Follow-up data were provided by each centre in an anonymised dataset and datasets were combined into a final password-protected database.
Our study cohort was further extended by 530 TCGA ECs in which a combination of a pathogenic POLE EDM, a TP53 variant (excluding TP53 variants classified as benign by SIFT or neutral by VEP), and/or MSI-H (based on Bethesda protocol classification [19] was obtained from the Genome Data Analysis Center (GDAC) (available at http://www.broadinstitute.org/ cancer/cga). TP53 mutations were assessed using the public databases COSMIC [20], ClinVar [21], and IARC TP53 mutations database [22], as well as the in silico tools SIFT [23] and PolyPhen [24]; only mutations classified as (likely) pathogenic or variants of unknown significance (VUS) were included in the study. This resulted in a final study cohort of 3518 molecularly profiled ECs.

Scoring MMR and p53 IHC
As multiple-classifier ECs are identified by IHC, we re-evaluated or restained tumours with available unstained slides to exclude potential misinterpretation in the original study. MMRd was defined as loss of MMR nuclear staining, for at least one MMR protein, with positive internal control. Subclonal loss of MMR expression [25] was defined as abrupt and complete regional loss of expression of an MMR protein with intervening stromal positivity serving as an internal control in the regions of absent tumour cell staining. Cases with heterogeneous or non-abrupt patchy staining thought to be a result of sub-optimal pre-analytic handling or those with absent internal control staining were not considered MMRd. An abnormal/mutant p53 IHC stain was defined as strong nuclear expression in over 75% of the tumour cells (overexpression), complete loss of nuclear stain in the presence of internal control staining (complete absence), or cytoplasmic staining (cytoplasmic), following scoring described by Singh et al [15]. Subclonal abnormal p53 IHC staining was defined as abrupt and complete regional abnormal p53 expression, in which the subclonal region was at least 10% of the total tumour volume. If the p53 IHC staining pattern could not be classified into one of the previously mentioned patterns (inconclusive p53 stain), TP53 mutational status was used for molecular profiling through Sanger sequencing of exons 5-8.

Histopathological review
Tumours with at least one haematoxylin-eosin (H&E) slide available were selected for histological review, blinded for molecular classification, by one pathologist (AL, TB, BG or RS). This review consisted of scoring for each case predefined histological features: hobnailing, slit-like spaces, papillary growth, squamous metaplasia, tumour intraepithelial lymphocytes (TILs), peritumoural lymphocytes, solid growth greater than 50%, and the presence of tumour giant cells. For the purpose of this study, we did not reassess tumour histotype. Type of myometrial invasion and presence of lymphovascular space invasion (LVSI) were also annotated.

Evaluation of somatic nucleotide and copy number variation in TCGA ECs
Somatic mutations were classified into the 96 categories described by Alexandrov et al [26]. Distance metrics for the difference between pairs of samples were generated using the mutational changes (1 -cosine similarity/2) or the proportion of genome that had a copy number change (both of these range between 0 and 1). These distances were combined into a single metric using Euclidean distance. Individual and combined metrics were used to cluster the samples and plot a heatmap using the function heatmap.2 from the package 'gplots' to visualise similarity between samples.

Statistical analysis
We assessed 5-year recurrence-free survival (RFS) and 5-year overall survival (OS) comparing Kaplan-Meier curves with the log-rank test. Follow-up time was calculated with the reverse Kaplan-Meier method. We used Mann-Whitney tests to compare non-parametric continuous variables; categorical variables were assessed with Fisher's exact test or the chi-squared test. Two-sided P value less than 0.05 was considered significant.

Identification and clinicopathological characteristics of multiple-classifier ECs
Our initial cohort comprised 3518 tumours, 3353 being classified to a single molecular subtype. Of the remaining 167 tumours, 30 were classified as MMRd-POLEmut ECs and will be reported separately (León-Castillo et al [18]), leaving 138 tumours (3.9%) that were assigned a provisional status of multiple-classifier with abnormal p53 (mutant p53 expression by IHC or a mutation in TP53). Stringent quality control including central pathological review resulted in the exclusion of 35 of these ECs for the following reasons: (1) reassignment of p53 immunostaining from mutant to wild-type pattern (n = 27, 21 being initially evaluated using tissue microarrays); (2) non-pathogenic POLE variant (n = 3); and (3) lack of evidence of MMRd on IHC review (n = 1) ( Figure 1). The remaining 107 tumours (3% of total) met the inclusion criteria and were used for subsequent analyses.
p53 IHC was available for central review in 28 of the 64 MMRd-p53abn tumours. Nine (30%) showed p53 overexpression, while one tumour had an inconclusive staining pattern in combination with a confirmed pathogenic TP53 mutation. Interestingly, the remaining 18 tumours (60%) showed abrupt strong p53 nuclear overexpression (> 75% of all tumour nuclei) in a well-defined area (at least 10% of tumour volume), a pattern defined as subclonal p53abn staining ( Figure 2).
Hierarchical clustering of ECs by SNV and SCNA proportions revealed that 12 of 13 POLEmut-p53abn ECs clustered with single-classifier POLE-mutant ECs, while a single case clustered with the single-classifier p53abn tumours (p ≤ 0.001, chi-squared test) (Figure 3). Separate analysis of SNV and SCNAs rendered similar results (supplementary material, Figure S2).
We further analysed the seven MMRd-POLEmut-p53abn ECs available in the TCGA. Five tumours had a TP53 (likely) pathogenic mutation and two had a VUS. Hierarchical clustering based on SNV and SCNA proportions (supplementary material, Figure S3) showed five tumours clustering with single-classifier POLEmut ECs, two with single-classifier MMRd ECs, and none with single-classifier p53abn ECs. Of note, the two ECs that clustered with single-classifier MMRd cancers had a V411L and L424I POLE variant, respectively, with low C>A substitutions and high indels [18] and pathogenic TP53 mutations.
The limited number of MMRd-POLEmut-p53abn ECs with available clinical data (n = 9) did not allow for survival analysis.

Clinical outcome of multiple-classifier ECs versus single-classifier ECs
Finally, we investigated the outcome of MMRd-p53abn ECs and POLEmut-p53abn ECs in the pooled study population (supplementary material, Table S2). Fourty-four MMRd-p53abn ECs were available for survival analysis. Patients had a 5-year RFS and OS of 83.4% and 82%, respectively (supplementary material, Figure S4). RFS and OS could be analysed for 23 patients with POLEmut-p53abn ECs. The survival analysis revealed that these ECs were associated with a favourable outcome with a 90.9% 5-year RFS and 5-year OS of 95.2% (supplementary material, Figure S4).
We further analysed the clinical outcome of multiple-classifier ECs by comparing their 5-year diagnostic pathology. Together with stage, the molecular subtype of EC provides powerful prognostic information; unlike stage, this information can be very accurately determined at the time of initial biopsy [33]. When using the surrogate approach to determine the four molecular subtypes (POLEmut, MMRd, p53abn, and NSMP EC), a small but clinically relevant number of ECs (3-5%) have been unclassifiable due to the presence of more than one molecular classifying feature. These so-called 'multiple-classifier' ECs create a hurdle in the application of the molecular EC classification in practice, as it is not immediately apparent how these should be classified/treated. This issue is particularly relevant for the co-occurrence of abnormal p53 IHC in the context of MMRd or POLEmut, as these features, when present singly, are associated with opposing clinical outcomes. The present study is the first to provide an extensive characterisation of these uncommon multiple-classifier ECs and by doing so, provides guidance on how these should be interpreted.
The molecular landscape of single-classifier POLEmut EC, p53abn EC or MMRd EC cases is considered to be shaped by their driver alterations, being POLE EDM, TP53 mutation or MMR deficiency. Interestingly, we observed similar mutational changes and SCNA between MMRd-p53abn EC and POLEmut-p53abn EC with single-classifier MMRd and POLEmut EC, respectively. This strongly suggests that TP53 variants occurring in the context of an MMRd or POLEmut EC are likely passenger events, not affecting the molecular landscape of the tumour. This is further supported by the phenotype of the multiple-classifier ECs, as pathology review revealed an enrichment for features associated with MMRd or POLEmut EC (TILs, peritumoural lymphocytes, squamous metaplasia) and not serous-like (p53abn) features. Another important observation in support of this interpretation is the unusual high frequency (47-60%) of subclonal abnormal p53 staining in the multiple-classifier EC. Of note, cases designated as single-classifier p53 EC from the previously published PORTEC-1/-2 cohorts (n = 74) did not show subclonal expression [10]. This same observation was made independently by Singh et al [15]. Together, these data strongly support the interpretation that the TP53 mutation is a later event during tumour progression in multiple-classifier EC, without affecting the molecular landscape or the phenotype.
Multiple-classifier EC can be identified on the basis of sequencing approaches (e.g. presence of POLE and TP53 mutations by sequencing) or by the use of surrogate markers (e.g. abnormal p53 IHC). By sequencing, more TP53 mutations will be identified, resulting in the higher number of multiple-classifiers present in TCGA (TP53 mutational status-based) compared with the cases identified using the surrogate marker approach (p53 IHC-based) [34]. This is likely the result of the high number of 'non-hotspot' TP53 mutations in POLEmut-p53abn EC compared with single-classifier TP53mut EC, as well as the higher sensitivity of sequencing approaches to identify subclonal events (low allele frequency). These non-hotspot TP53 mutations appear to not always impact the expression and function of p53. This is in line with Singh et al [15], where a high proportion of cases with such a TP53 variant and p53 wild-type staining pattern were identified in multiple-classifier ECs.
We also describe a limited number of MMRd-POLEmut-p53abn ECs ('triple-classifier'). The data available are insufficient to suggest how to treat patients with these cancers. The presence of p53abn subclonal staining in these cases suggests that, as in MMRd-p53abn and POLEmut-p53abn ECs, the TP53 mutations occur as a secondary event. However, it is difficult to assess, with the present data, whether POLEmut or MMRd are the driving events in these cancers. A larger number of these triple-classifier cases will be required to study the biological behaviour of these ECs. Until then, considering the results reported by León-Castillo et al [18] on MMRd-POLEmut ECs and the present study, we suggest that these triple-classifiers be classified as POLEmut ECs if they have a pathogenic POLE EDM based on whole-exome sequencing (WES) data or, in the case of absence of WES data, if the POLE EDM corresponds to one of the 11 pathogenic mutations described by León-Castillo et al [18]. In tumours in which the triple-classifier ECs carry a POLE EDM that does not comply with the previous criteria, we recommend that they be classified as MMRd ECs, as discussed in depth by León-Castillo et al [18].
Finally, we addressed the question of whether the biological behaviour of POLEmut and MMRd ECs is impacted by the presence of abnormal (subclonal) p53 expression. Although the number of cases is limited, the clinical outcomes of MMRd-p53abn ECs and POLEmut-p53abn ECs are strikingly different from what would be expected in single-classifier p53abn (SCNA-high/serous-like) ECs [1,[9][10][11][12][13][14]. This supports the concept that passenger events do not affect biological behaviour. For clinical management, this means that the presence of TP53 mutations in the context of MMRd EC or POLEmut EC should not prompt intensified treatment.
In addition to multiple-classifier ECs with mutant p53 expression/TP53 mutation, a fourth group of ECs with multiple classifying features can be encountered: MMRd-POLEmut ECs. The genomic architecture and clinical outcome of MMRd-POLEmut ECs differ depending on the pathogenicity of the POLE exonuclease domain variant and the ultramutated phenotype it confers to the tumour [18]. Thus, the base change and indel proportion of ECs with pathogenic POLE EDM (as defined by the POLE score) and MSI are similar to MSS ECs with pathogenic POLE EDM [18]. Additionally, when examining a cohort of MMRd ECs with pathogenic POLE EDM, although the number of patients was limited (n = 14), a good clinical outcome was observed (5-year RFS 92.3%), in line with the prognosis described previously for single-classifier POLEmut ECs (supplementary material, Figure S5) [9][10][11][12][13][14]18,35,36]. These findings support the classification of tumours with a pathogenic POLE EDM and MMRd as single-classifier POLEmut ECs.
In conclusion, this study is the first to describe evidence in support of categorising multiple-classifier POLEmut-p53abn EC as single-classifier POLEmut, and MMRd-p53abn EC as single-classifier MMRd. Although rare (∼3%), correct designation of multiple-classifier ECs facilitates the implementation of the molecular EC classification, enabling them to be included in future studies and, more importantly, providing valuable information for clinicians and patients to guide management.  Table S1. Morphological features of MMRd-p53abn, POLEmut-p53abn, and MMRd-POLEmut-p53abn ECs Table S2. Clinicopathological features of MMRd-p53abn and POLEmut-p53abn ECs with available survival data