The adaptive immune and immune checkpoint landscape of neoadjuvant treated esophageal adenocarcinoma using digital pathology quantitation

Esophageal cancer (EAC) incidence has increased several fold in the last 30 years and is responsible for over four hundred thousand deaths globally per annum (http://​www.​cancerresearchuk​.​org/​health-professional/​oesophageal-cancer-statistics), with adenocarcinoma now the dominant histology reported in the Western world [1]. As well as surgery, the current treatment regimens include cytotoxic chemotherapy and radiation and are associated with significant morbidity. Five year survival rates remain poor at 17% and for resectable disease this figure remains at approximately 45% despite improvements in surgery and neo-adjuvant therapy. The wide variability in patient outcome following resection is indicative of the need for robust biomarkers for prognostication, even following the improved survival rates seen with the introduction of neo-adjuvant chemotherapy [2, 3]. There is an urgent need for targeted therapies and deeper understanding of EAC.

Response to checkpoint inhibition has been shown to be more effective in tumours with a higher mutational load [4]. These tumour types are increasingly being considered as potential targets for immunotherapy. EAC is one example of a tumour type with a high mutational frequency with as many as three hundred thousand mutations per tumour [5].

The immune context of the tumour microenvironment (TME) comprise a range of regulatory proteins which, upon activation, can inhibit effective T-cell response leading to antitumor immunity. This regulation is a vital checkpoint in the inflammatory cascade, which many tumour types are able to evade. However, there have been recent success with checkpoint inhibition in solid tumours where increased expression of checkpoint proteins has been associated with a poor clinical outcome [6, 7]. This has been specifically demonstrated for EAC [8‐10]. There is evidence that PD-L1 expression in EAC tumours is predictive of response in phase 1 trials, NCT01928394, NCT01943461 and NCT01772004 reported in a recent meta-analysis [11]. Nevertheless, increasing data suggest that expression of PD-L1 alone may not be adequate to be predict response due to the lack of response seen in many PD-L1 expressing tumours [12, 13]. This is further compounded by responses reported in PD-L1 negative cases [14]. Consequently, there has been a shift in focus to the immune contexture of the TME to identify cell phenotypes which may serve as prognostic biomarkers [15, 16]. These efforts including identification, quantification and spatial analysis of tumour-infiltrating lymphocytes and other immune and immune checkpoint biomarkers.

We report here on the immunohistochemical expression of several T cell linage markers and immune checkpoint biomarkers and their combinatorial influence on EAC outcome. We highlight a potential therapeutic advantage for a subset of patients and discuss the survival impact of immunological factors.

Here, we described the immunophenotypic landscape of EAC and describe the existence of an ‘immune hot’ subgroup of patients that confers a significant survival advantage. Further delineation of only high CD45RO/ICOS cases was then investigated, informed by multivariate and correlative analysis, improving patient inclusion, stratification and OS advantage. We also describe, by multiplex immunofluorescence, co-expression of CD45RO/ICOS positive cells in these cases, observing a significant difference across the contrasting immune groups. Interrogation of dual labelled cell expression within the tumour and stroma revealed that stromal co-expression was significantly increased in the ‘immune hot’ group.

Efforts towards a better understanding the tumour-immune milieu of EAC are gradually increasing. We believe our collaborative study, inclusive of a validation EAC cohort, is the largest of its kind reported to date. Making these data one of very few comprehensive publications delineating immune biomarkers in EAC.

In some tumour types, NSCLC for example, there is extensive data supporting the assessment of PD-L1 protein expression as a good predictor of patient response to immunotherapy. The evidence is scant for such application in EAC, owing mainly to the wide-ranging expression patterns seen in EAC. Recent results of Keynote 62 study reported that pembrolizumab to first line chemotherapy did not show any benefit [13]. Though membranous epithelial expression has been reported in EAC, which was also observed in our study, PD-L1 expression has been predominantly observed in the surrounding tumour-stroma rather than the membranes of epithelial cells, as reported in KEYNOTE-012 and observationally by others [28‐30]. This is in addition to the subjective and challenging pathological assessment of PD-L1, which may complicate the adoption of the diagnostic test [24]. Recently, negative prognostication of PD-L2 was reported for esophageal cancer. Though this is was mainly in squamous cell carcinoma, the authors found no significant correlation between PD-L2 and PD-L1 expression. However, these data may indicate evaluation of both PD-L2 and PD-L1 expression may be clinically important [31].

Attempts to assess densities of T cells in EAC have been made. The expression of high CD3, CD4, CD8 and FOXp3 cells in EAC was reported in 128 cases, where all markers achieved significance in univariate analysis, closely mirroring our own data. However, only high CD8 cells alone reached positive prognostic significance in multivariate analysis [32]. A similarly powered study demonstrated significance on OS by all markers (CD3, CD8 and FoxP3). Though CD8 positive T cell expression has been shown to be positively prognostic in EAC [33], Thompson et al, recently demonstrated that high CD8 infiltration was correlated with impaired progression free survival and OS [28]. These data directly contrast with our own, where we demonstrate, as with several aforementioned studies that increased CD8 densities correlated with improved OS. We believe disparities are likely due our increased cohort numbers, robust digital pathology quantification methodology and a statistically derived cut-offs in comparison to a lower sample numbers and alternatively defined dichotomisations (median and quartile thresholds).

We acknowledge weaknesses in our study such as the limited number of cases interrogated by multiplex, restricting our assessment of the impact on survival by these CD45RO/ICOS positive dual labelled cells. Nevertheless, in these few exemplar cases we demonstrate a statistically significant difference in expression of these dual labelled cells across ‘immune hot’ and ‘immune cold’ patient groups, which have individually and together shown survival impact by single chromogenic DAB IHC. The limited number of TRG 1 and 2 cases in our cohort has the potential to confound our analysis. To examine this, we undertook an analysis inclusive of only TRG3 and higher. No difference was observed and CD45RO/ICOS positive cases remained significantly positive prognostic by multivariate analysis (Supplementary Table S3). We acknowledge the increased number of gastric case in the validation cohort, which are potentially more inflammatory due to being potentially MSI high or EBV positive. MSI/EBV status is unexplored in this cohort. We also accept the retrospective, post-treatment nature of the cohort and recognise that preoperative treatment may likely to induce a permanent or transient change in the immune microenvironment of EAC [34]. We ideally would wish to compare and contrast the immune profiles of a surgery only cohort, without chemo/chemoradiation treatment, in comparison with our data here. With few patients going straight to surgery without some form of treatment, we are unable to obtain samples from a suitable historical cohort of sufficient size to explore this question. However, this cohort is representative of current clinical practice.

It is rational to assume that PD-L1 inhibitors may have applicability in EAC. However, studies have demonstrated contrasting prognostication of PD-L1 expression in EAC [8‐10, 35]. Interestingly, as with PD-L1, we observe that ICOS, a member of the B7-family of proteins, as is PD-L1, was expressed predominantly in the stroma, conferring a positive prognostic survival advantage. Our data add support to the hypothesis that some of the more novel B7-family members could be regulating CD4 T-cell differentiation toward specific effector cell phenotypes. Indeed, ICOS/CD4 correlation has been reported in colorectal cancer, where expression is associated with improved survival and may be a clinical biomarker for good prognosis [36]. Here, we show the association of CD45RO/ICOS as being greater than that of ICOS/CD4 which maybe highlighting a previously unreported population of ICOS positive memory cells, which encompass CD4. The prognostication of high ICOS/CD4 cells may be being enhanced in our study, by evaluation of all CD45RO positive cells, which include; all memory T cells, B cell subsets, activated monocytes/macrophages and granulocytes, not solely a subpopulation of CD4 positive cells. Our data emphasise the scrutiny required when analysing experimental and trail data to consider 1) The specific cell phenotype of interest. 2) The expression of the target. 3) The expression of associated proteins and ligands. 4) The specificity of the antibody used. Our data may yet be corroborated in clinical trial data with the ICOS agonist GSK3359609 (GlaxoSmithKline, London, UK), currently evaluated in a phase I trial INDUCE-1, NCT02723955. As well as JTX-2011, (Jounce Therapeutics, Cambridge, USA) in phase I/II ICONIC trial, NCT02904226. Future work aims to assess CD45RO/ICOS positive dual labelled cells by multiplexing in a large cohort of cases to assess impact on survival.

We describe the existence of both ‘immune hot’ and ‘immune cold’ patient populations in EAC. Several immune biomarkers examined were shown to be positively prognostic for OS. Upon correlative and multivariate analysis, we identified a high CD45RO/ICOS subgroup of patients, with co-expression of these biomarkers, stratifying patients in to contrasting survival groups. Multiplexing revealed dual labelled CD45RO/ICOS positive cells to be expressed in the stromal compartment of EAC tissue. These dual labelled cells were significantly differentially expressed across the ‘immune hot’ and ‘immune cold’ patient populations. In summary, we have identified a subpopulation of high CD45RO/ICOS EAC patients with significant positive prognostication. Moreover, these data provide evidence for the assessment of CD45RO/ICOS to be employed in the stratification of patients for clinical trials investigating the response to immunotherapy.