Epidermal AMBRA1 and Loricrin; a paradigm shift in the prognostication and stratification of AJCC stage I melanomas

1&2Robert Ellis, MB ChB, MD, 1&2 Diana Tang MB BS, MD, 1&3Batoul Nasr MB ChB M Clin Res, 4Alison Greenwood, BSc 1Ashleigh McConnell, PhD, 1Maria Eleni Anagnostou PhD, 1Martina Elias, PhD, 1Stamatina Verykiou MB ChB, PhD,1Dalvir Bajwa MB BS 1, Tom Ewen PhD1 , Nick J Reynolds, MB BS, MD, 3Paul Barrett MB ChB, 5Edward Carling MB ChB, 1&4 Graeme Watson MB BS, PhD, 6Jane Armstrong PhD, 7 A. Joy Allen PhD, 8Stuart Horswell M Math, 1Marie Labus PhD and 1Penny E Lovat ,


Translational relevance
What is already known?
• There is an unmet clinical need for biomarkers of early stage melanoma • AMBRA1 is a pro-autophagy regulatory protein with known roles in cell proliferation and differentiation; and is a known tumour suppressor.
• Loricrin is a marker of epidermal terminal differentiation What does the study add?
• AMBRA1 has a functional role in keratinocyte/epidermal proliferation and differentiation • The combined decrease/loss of peri-tumoural AMBRA1 and Loricrin is associated with a significant increased risk of metastatic spread in AJCC stage I tumours versus melanomas in which peri-tumoural AMBRA1 and Loricrin are maintained, independent of Breslow depth.

What is the translational message?
The adoption of peri-tumoural epidermal AMBRA1/Loricrin biomarker expression into melanoma care guidelines will facilitate more accurate, personalised risk-stratification for patients with AJCC I melanomas, thereby facilitating stratification for appropriate follow-up and informing on post diagnostic investigations including SLNB, ultimately resulting in improved disease outcomes and rationalisation of healthcare costs.

Summary Background
Despite the recent update to the AJCC staging criteria for melanoma, this system is still unable to identify truly high-risk stage I tumour subsets.

Objective
To determine clinical utility of combined epidermal AMBRA1/Loricrin (AMLo) expression as a prognostic biomarker for AJCC stage I cutaneous melanoma.

Methods
Peri-tumoural AMBRA1 expression was evaluated in a retrospective discovery cohort of 76 AJCC I melanomas. Multivariate analysis of AMLo expression was subsequently correlated with clinical outcomes up to 12-years in two independent powered, retrospective validation and qualification cohorts comprising of 379 AJCC I melanomas.

Results
Decreased AMBRA1 expression in the epidermis overlying primary melanomas in a discovery cohort of 76 AJCC I tumours was associated with 81.5% 7-year DFS versus 100% survival with maintained AMBRA1; P<0.081. Following automated IHC protocol development for semi-quantitative analysis of AMLo further analysis was undertaken in validation (n=218) and qualification cohorts (n=161) of AJCC I melanomas. Combined cohort analysis revealed a DFS of 98.3% in the AMLo low-risk group (n=239) versus 85.45% in the AMLo high-risk cohort (n=140, P<0.001). Sub-cohort, multivariate analysis

INTRODUCTION
Melanoma is one of the most devastating skin cancers, with a worldwide incidence that continues to climb (1), (2). Although the introduction of targeted and immune-therapies has revolutionised treatment of metastatic disease, the largest proportion of patients presenting to clinicians however, have thin, early stage melanomas yet to benefit from therapeutic innovation, in part related to a lack of credible biomarkers of disease progression.
Currently, disease staging and risk prediction is based on histological characterisation of the primary tumour; including depth of tumour invasion (Breslow depth) and the presence of epidermal ulceration, forming the basis of the 7 th edition American Joint Committee on Cancer (AJCC) staging criteria (3). The recently updated 8 th edition AJCC guidelines came into effect in January 2018 (4), with removal of mitotic count and reduction of Breslow depth for stage IA melanomas to 0.8mm. However, these criteria are still unable to reliably identify which individuals with seemingly low risk early melanomas are at specific risk of disease progression; occurring in up to 15% of patients with AJCC I melanoma (4,5). An urgent unmet need for credible prognostic biomarkers able to identify patients with high risk early stage melanomas, facilitating appropriate counselling and follow up (including guidance on appropriate need for sentinel lymph node biopsy (SLNB)) or access to clinical trials and potentially adjuvant systemic treatment (6, 7), thus remains.
We have identified two protein markers, AMBRA1 and Loricrin, in the epidermis overlying primary melanomas, whose expression is lost in high-risk AJCC I melanomas, but which are retained over genuinely low-risk tumours. The role of AMBRA1 (a pro-autophagy regulatory protein) in melanoma progression was initially evaluated by immunohistochemistry (IHC) in a retrospective melanoma discovery cohort following previous reports of the role of autophagy in melanomagenesis (8,9); however, unlike previously investigated autophagy biomarkers such as p62 (8), AMBRA1 revealed variations in expression within the epidermis overlying primary melanomas, rather than within the tumour itself, suggesting a specific role for this protein in epidermal differentiation (9). In the present study to investigate peri-tumoural AMBRA1 as a potential prognostic marker, we initially evaluated immunohistochemical semi-quantitative expression in a retrospective discovery cohort of AJCC stage I melanomas. However, although loss of peri-tumoural AMBRA1 expression correlated with disease progression with an assay sensitivity of 100%, the relatively low assay specificity (33%) hindered its clinical utility. Subsequently to improve specificity, peri-tumoural epidermal AMBRA1 expression was assessed in combination with epidermal Loricrin (as a marker of keratinocyte terminal differentiation), with the primary aim of this study to validate combined peri-tumoural AMBRA1 and Loricrin (AMLo) expression as a prognostic biomarker for early stage melanoma.

Patient cohort selection
This study included three, independent, statistically powered retrospective cohorts of AJCC stage I melanomas defined by the AJCC 7 th edition (figure 1). All cohorts were accessed following ethical permission (REC Ref 08/H0906/95+5_Lovat). All steps of biomarker development followed the Cancer Research UK Prognostic Biomarkers Roadmap (10) and reported in line with REMARK guidelines (11).
The initial discovery cohort of 76 patients was derived from Newcastle Hospitals NHS Foundation Trust Semi-quantative analysis of epidermal AMBRA1 expression was undertaken using Leica Digital Image Hub software (Leica Biosystems). Up to ten representative x200 microscope fields were analysed for mean positive pixel intensity of AMBRA1 expression levels and compared to the mean AMBRA1 expression in the epidermis directly above the melanoma allowing a relative percentage expression change to be calculated with the normal epidermis considered as 100% expression.

Validation of AMBRA1 and Loricrin Scoring
Visual AMBRA1 and Loricrin scoring for all cohorts was undertaken by three individuals blinded to outcomes (RAE, EC, GW); consensus agreement was reached for all samples. The "by-eye" analysis was grouped as either "Maintained AMBRA1" or "Decreased/lost AMBRA1". "Maintained AMBRA1" was defined as no discernible difference in AMBRA1 expression between normal and peri-tumoural epidermis. "Decrease/lost AMBRA1" was defined as any decrease or complete loss of AMBRA1 expression between normal and peri-tumoural epidermis.
All loricrin expression analysis was undertaken by eye, and loss of expression defined as any discernible break in the continuity of expression in the stratum corneum.

Statistical Analysis
Survival analyses were conducted using the R function coxph(). Univariate estimates presented are coefficients, with 95% confidence estimates, resulting from a Cox fit with only that covariate as a predictor, while multivariate estimates are the associated coefficient from a full additive model including all reported covariates (in particular, we did not re-fit to exclude non-significant predictors and considered no interaction terms).
Survival curves were generated using the R function survfit() to present Kaplan-Meier curves for a univariate model based on AMBRA1/Loricrin status, with presented P values as the result from a Log Rank (score) test for the associated univariate Cox model. Statistical analysis for disease free survival (DFS) was planned for each cohort independently, as well as in combination for the James Cook University Hospital (JCUH) and the University hospital of North Durham (UHND) cohorts. Further sub-cohort analysis was undertaken in AJCC stage IA and IB patients, as well as patients eligible for sentinel lymph node biopsy (SLNB) under the current UK National Institute for Health and Care Excellence (NICE) i.e with a Breslow depth >1mm.
For the analysis of SLNB data sets, post-test odds are used to calculate the positive and negative predictive values (i.e. clinically relevant measures of diagnostic accuracy). These are calculated by multiplying the pre-test probability of a positive (negative) result (the prevalence or 1 -prevalence) with the diagnostic likelihood ratio of a positive (negative) test (12).

Peri-tumoral AMBRA1 expression and DFS in the NUTH Discovery cohort
To identify an association between epidermal AMBRA1 expression overlying primary tumours (peritumoural AMBRA1) and DFS in patients with AJCC I melanoma, 76 patients within the Newcastle Discovery cohort were stratified as either maintained or decreased AMBRA1 expression by visual analysis of peri-tumoural AMBRA1 expression (figure 2). All samples were also analysed for cytokeratin 5 (CK5; figure2A), a pan-epidermal marker, which revealed no association between the degree of melanoma epidermal invasion and peri-tumoural AMBRA1 expression.
Kaplan-Meier survival analysis revealed reduced 7-year DFS in patients with decreased peri-tumoural epidermal AMBRA1 expression compared to 22 patients with maintained AMBRA1 expression (81.5% vs 100%; P =0.081, figure 2B). The hazard ratio (HR) for disease recurrence among patients with decreased AMBRA1 expression versus maintained expression however, could not be assessed due to a lack of events in the maintained cohort.
Although decreased AMBRA1 expression was associated with a reduced 7-year DFS with a sensitivity for identifying patients at risk of disease progression of 100%, the specificity of the test was only 33.3% (figure 2D), limiting clinical utility.

AMBRA1 as a marker of epidermal differentiation
To further underpin the role of AMBRA1 in epidermal differentiation, western blot analysis for the expression of AMBRA1 and associated epidermal proteins was performed in primary keratinocytes undergoing calcium-induced differentiation in vitro. Results revealed a consistent time-dependent increase in AMBRA1 expression in line with increased differentiation; highlighted by decreased expression of CK14 (a marker of basal keratinocytes) and increased loricrin expression. Conversely, siRNA-mediated knockdown of AMBRA1 in primary keratinocytes resulted in deregulated differentiation as evidenced by down regulation of loricrin at both the mRNA and protein level (supplementary figure 2).

Combined AMBRA1/Loricrin peri-tumoural epidermal expression and DFS in the JCUH and UHND validation cohorts
To compare the method of determining AMBRA1 expression, both visual and semi-quantitative analysis of peri-tumoural AMBRA1 expression were performed in the JCUH cohort (figure 3, supplementary figure 4). Results revealed a significant difference in median percentage loss of AMBRA1 expression from 11.2% in the "no visual loss" group to 84.1% in the "visual loss present" group (figure 3B; Mann-Whitney P<0.001). As such, any decrease or loss of peri-tumoural epidermal AMBRA1 expression when compared "by eye" with the expression of AMBRA1 in normal epidermis was deemed as "high-risk" (supplementary figure 4). Furthermore, these observations confirmed the robustness of visual analysis, which was subsequently used alone for all other cohort analyses, and highlighted the major benefit of having an internal control (normal epidermis) present in each primary tumour excision sample.
Visual assessment of Loricrin also defined two distinct sub-sets ( figure 3B, supplementary figure 5).
To confirm the association of decreased DFS with alterations of the peri-tumoural epidermis, IHC expression of loricrin was also assessed in a small sub-group of the NUTH discovery cohort of AJCC stage I melanomas revealing an association between peri-tumoural loricrin loss and decreased DFS with a high degree of assay specificity but lower sensitivity (supplementary figure 6). Strikingly, however, when epidermal loricrin expression was combined with epidermal AMBRA1 expression, results revealed decreased peri-tumoural expression of both markers was associated with 100% sensitivity and specificity for identifying truly high-risk melanomas in this sub-cohort (supplementary figure 6).
Consequently, the combination of AMBRA1/Loricrin was deemed as "high-risk" if AMBRA1 peri-  figure 5D), suggesting that AMBRA1/Loricrin is a prognostic marker, independent of Breslow depth, which is able to stratify patient risk over-and-above AJCC staging alone.

DISCUSSION
We are currently experiencing what has previously been described as the "golden age" of melanoma therapy (13), with an ever expanding arsenal of systemic medications available for patients with metastatic disease resulting in increased survival periods beyond the expectations of the most optimistic of clinicians even 10 years ago (14,15). The thrust of trials is now aimed at adjuvant initiation of these therapies which has resulted in unprecedented results in patients with surgically resected, AJCC stage III melanoma at high risk of recurrence (16,17). The natural progression to finally tackling melanoma is the initiation of systemic therapy at the earliest possible point at which high-risk individuals can be identified; thus stopping the development of life threatening metastases, or treating them before they have affected the patient's health.
As such, a yet unrealized dream for the management of melanoma is the ability to identify patients at the highest risk of progression as close to the time of diagnosis of a primary melanoma as possible; allowing increased surveillance and earlier therapeutic intervention. Conversely, patients at truly low risk of disease could be more confidently reassured, and follow-up regimes altered. The largest population of patients affected by melanoma have AJCC stage I disease. As such, any improvements in the care of this group of patients will affect the largest number of patients overall, with the associated potential health economic benefits.
AJCC staging of melanoma alone, relying on Breslow depth and the presence of ulceration is not a perfect predictor of outcome in the lowest risk, stage I group; where a small but significant proportion of patients will still die of their disease. Recent, evidence based changes to the AJCC 8 th edition staging criteria highlight a "breakpoint" of 0.8mm Breslow, with non-ulcerated primaries below this depth being classified as stage 1A, with ulcerated tumours <0.8mm, or non-ulcerated tumors from 0.8 -2mm Breslow classified as stage IB. This has increased the number of patients now classified as stage IB, with implications on increased surveillance for stage IB patients (5 years follow-up) versus IA (1 year follow-up).
In the present study we describe the discovery and validation of the combined expression of two protein biomarkers, AMBRA1 and loricrin (AMLo) in the epidermis overlying primary AJCC stage I melanomas as a highly sensitive and specific prognostic biomarker. AMBRA1 is a pro-autophagy regulatory protein and our in vitro data further define a functional role for AMBRA1 in epidermal proliferation, and like loricrin, in epidermal differentiation.
There is ongoing controversy about the clinical role of sentinel lymph node biopsy (SLNB) with large scale trials revealing questions about the prognostic and therapeutic role of SLNB, as well as highlighting the associated morbidity and healthcare costs of patients undergoing the procedure unnecessarily. As with melanoma in general, improved stratification of those patients at the highest risk of disease progression would potentially allow SLNB to further refine individual disease risk.
The MSLT-1 SLNB trial (18) contained a cohort of 765 intermediate thickness primary tumours undergoing SLNB that ranged from 1-3.5mm Breslow depth. Overall, the pre-test probability of a patient in this cohort developing a metastasis was 16%. In patients with a positive SLNB the number developing metastases was higher, giving a post-test probability of developing metastases in this group at 33% (95% CI 29-36%); however, a negative SLNB was still associated with a 13% chance of disease progression (95% CI 11-15%) (figure 6) (18). These data therefore suggest that a positive SLNB is able to further risk stratify patients, yet a negative SLNB adds little to reassure patients about their true risk of metastasis.
Although analysis has been undertaken in separate cohorts, assessment of the outcomes in our SLNB eligible combined cohort (patients with a Breslow depth over 1mm as per UK guidelines) nevertheless revealed a pre-test probability of metastasis of 10% is increased to 18% (95% CI 12 -24%) in the AMBRA1/Loricrin high-risk group. In contrast, low-risk AMBRA1/Loricrin however, was associated with a post-test probability of only 1.4% chance of metastasis (95% CI 0 -5%, figure 6). In this context, these results highlight the potential of AMBRA1/Loricrin as a valuable pre-SLNB test; identifying those patients that would receive no further benefit from SLNB, and increasing the positive predictive value of SLNB through use in only a high risk, AMBRA1/Loricrin refined cohort.
In summary, our study reveals AMBRA1/Loricrin as a novel prognostic biomarker for early stage cutaneous melanoma, over-and-above AJCC staging alone. This simple, IHC-based marker will integrate seamlessly into standard clinical pathways of melanoma diagnostics, and allow a greater degree of certainty around disease outcomes for the treating clinician. Given the current prevailing view that SLNB is a purely prognostic tool, and considering the cost and morbidity associated with SLNB there is potential that the AMBRA1/loricrin biomarker may later replace SLNB. Not only will this benefit the individual patient in terms of reduced psychological burden from greater clarity of disease risk, but it will also allow better healthcare resource utilization internationally. As the golden age of melanoma care continues, the adoption of the AMBRA1/Loricrin biomarker into clinical guidelines thus presents a major paradigm shift in melanoma prognostication and stratified personalized management for the future.      1

Supplementary Methods
For manual detection of AMBRA1 expression tumours were subjected to pre-optimised IHC analysis (1) with antigen retrieval preformed in 10 mM Tris-Hcl (pH 7.6) and primary antibody binding detected with primary AMBRA1 (Abcam) diluted 1:200, visualised using VIP counterstaining (Vector Labs) .
Antigen retrieval conditions and antibody dilutions for the automated IHC detection of AMBRA1