Background: Several studies suggested recently, that deep-learning (DL) algorithms may be able to predict colorectal cancer (CRC) patients’ prognosis directly from Haematoxylin/Eosin stained tissue sections. However, due to the mostly ‘black-box’ nature of DL algorithms, widespread implementation into routine diagnostics as well as into research pipelines is still lacking.

Methods: In this study, we investigate the relationship between a recently published DL-based end-to-end CRC patient prognosis prediction score, histomorphological and molecular features of two independent CRC cohorts (DUESSEL-CRC (n=164) and TCGA-CRC (n=207)). Histopathological features included Stroma AReactive Invasion Front Areas (SARIFA), tumor-adipose-feature (TAF), morphometrically measured proportion of tumor (PoT) data, grade of differentiation, lymph node status (pN), and depth of invasion (pT). Molecular features included mismatch repair (MMR) status, expression level of Ki67, BRCA1/2, ATM, consensus molecular subtypes (CMS) and pathway-derived subtypes (PDS).

Results: In DUESSEL-CRC and TCGA-CRC cohorts, presence of SARIFA and of TAF were associated with a higher prediction score, i.e. a higher risk of death according to the DL-based model (all p<0.0001). Higher intratumoral stroma content (PoT-low), immunohistochemical expression level of Ki67, BRCA1/2, ATM, and MMR status were variably related to the DL-based prediction score. Interestingly, the DL-based prediction scores for different transcriptional CMS and PDS groups did not differ significantly. Presence of metastatic lymph nodes or poor grade of tumour differentiation were also associated with a higher DL-based prediction score (all p<0.05).

Discussion: Our current study suggests that linking DL-based prediction scores to known histopathological or molecular features might increase the explainability of DL-based prognostication tools. Further studies need to demonstrate whether DL-based prediction score have added value to already known prognostic factors in CRC patients.
 

Background: Crohn’s Disease (CD) affects up to 1% of the European population and is increasing in incidence. Well-established single nucleotide polymorphisms (SNPs) influence CD risk. CD effects are mostly due to chronic transmural inflammation, sometimes leading to progressive fibrosis of the wall with thickening and partial luminal obstruction.

Purpose: To accurately quantify cells and fibrosis involved in Crohn’s fibrostenosing lesions, within each intestinal wall layer, which has not been previously reported.

Methods: Formalin-fixed, paraffin-embedded terminal ileal resections from 30 controls and 30 CD patients were stained using immunohistochemistry (IHC) to identify specific cell types: CD3, CD4 and CD8 T-cells, CD20 B-cells, CD68 macrophages, CD31 endothelial cells and smooth muscle actin-positive cells. Collagen deposits were visualised using Picro-Sirius Red (PSR) staining. The sections were captured digitally and analysed with QuPath, open-source imaging analysis software, to quantify the fibrosis and cell populations within each ileal layer (mucosa, muscularis mucosae, submucosa, muscularis propria, serosa). Collagen deposits were quantified using a machine-learning classifier, while the DAB-positive IHC-detected cells were counted using the built-in positive-cell detection function.

Results: The quantification of features showed partial loss of the mucosa layer due to widespread ulceration, with a notable expansion of the muscularis mucosae mostly due to increased fibrosis and smooth muscle hypertrophy. The serosa was also expanded by fibrosis. There was increased infiltration of all T and B lymphocytes (forming lymphoid aggregates in all cases), macrophages (forming granulomas in some cases) and endothelial cells, in all layers, except for the mucosa (due to its partial loss from ulceration).

Conclusion: Quantitative analysis showed marked cellular changes in all intestinal wall layers within Crohn’s fibrostensing lesions, especially muscularis mucosae and serosa expansion. QuPath allows accurate quantification of fibrosis and cell infiltrations in all layers of the intestinal wall associated with CD and this may help to guide personalised therapy.
 

Background
Regional lymph node (LN) status is a key prognostic factor in patients with oesophageal cancer (OeC). Tumour derived antigens can lead to immune activation in LNs which might provide clinically useful information of the host’s anti-tumour immune response. It is currently unknown whether the immune response is homogeneous across all tumour-negative LNs (LNneg) within a patient.

Purpose
We hypothesized that all LNneg within an OeC patient have a similar microarchitecture irrespective of their size and analysed the heterogeneity of LNneg morphology to established whether the largest LNneg can be used as a surrogate for the immune response of all LNnegs within a patient.

Methods
82 pN0 patients from the OE02 trial had at least two LNneg. The microarchitectural LN features (germinal centres (GC), lymphocytes outside GCs, histiocytes) were morphometrically quantified. Linear mixed-effects regression models, intraclass correlation coefficients (ICC) and Bland-Altman plots were used to determine systematic bias, reliability/variability and agreement of the LNneg microarchitecture measurements.

Results
Linear mixed-effects models revealed no systematic bias in LNneg morphology within a patient.
The ICC indicated that variability was moderate for lymphocytes (ICC: 0.42; 95%CI: 0.11–0.63, p=0.007)) and GCs (ICC: 0.50; 95%CI: 0.23–0.68, p<0.001), and high for histiocytes (ICC: 0.07 (95%CI: -0.44–0.4, p=0.38). Bland-Altman analyses showed at maximum 8.5 % of values outside of the 95% limits of agreement.

Conclusions
This study is the first to systematically assess the agreement of the microarchitectural features in LNneg within an individual pN0 OeC patient. The absence of systematic bias supports the use of the largest LNneg as surrogate for a patient’s overall immune response in OeC. Further studies are warranted to investigate whether the identified heterogeneity of the LNneg immune reaction within a patient is related to the distance between individual LNnegs and primary tumour or variation in lymphatic drainage routes.
 

Background
Oesophago-gastric cancers (OGC) are common malignancies with high disease-related mortality. Predictive biomarker testing can guide individualised treatment; testing for HER2, MSI/MMR, and PD-L1 is recommended, but implementation varies.

Purpose
To formulate consensus recommendations among pathologists and oncologists for OGC best-practice biomarker testing in the UK.

Methods
A pragmatic literature review was conducted to inform development of statements using a modified Delphi method. UK-practicing pathologists and oncologists were recruited for two rounds of online questionnaires, composed of topics relating to the overall pathway, pre-analytical, analytical, and post-analytical considerations. Statements were rated using a 7-point Likert scale, from 1, ‘strongly disagree’, to 7, ‘strongly agree’, and responses consolidated to quantify levels of agreement. Consensus was pre-determined at 80%. Responses were discussed during a virtual consensus meeting for validation and final recommendations.

Results
Participants from across the UK completed the first (n=26) and second Delphi rounds (n=18), with an even split between pathologists and oncologists. Consensus was reached (both 89%) that clear guidance and standardisation of the pathway across UK laboratories and Trusts is needed, and that coordination throughout the tissue journey is required for best practice. However, while oncologists reached consensus that reflex testing for all biomarkers should be implemented, pathologists did not, raising concerns relating to capacity. Consensus was reached that a minimum of 8 biopsies should be taken (92%), testing metastatic tumour samples was acceptable (100%), biomarker results should be reported in 5 working days (81-85%), and should be visible in the diagnostic pathology report (96%).

Conclusions
Best-practice recommendations for biomarker testing in OGC were formulated using opinions from UK pathologists and oncologists. These results support the need for standardisation and co-ordination throughout the tissue journey. Significant concerns were raised representing barriers to testing, highlighting challenges faced in clinical practice and a need to foster successful working relationships.
 

Background: Colorectal cancer (CRC), the fourth most common cancer in the United Kingdom, differs in prognosis depending on the tumour stage and early (vascular invasion) or late metastatic events (nodal and distant metastases). Genomic analysis from the Cancer Genome Network (1) indicated that the transcription factor Cut Like Homeobox (CUX) 1, may be related to such metastatic events. This was corroborated locally and hence another member of the CUX family, CUX2, was investigated for similar correlations.

Purpose: This study aimed to explore whether expression of the transcription factor CUX2 correlated with clinicopathological variables and the metastatic phenotype of CRC.

Methods: Manual scoring of CRC tissue microarrays (n=380), immuno-stained for expression of CUX2 at the luminal, middle, and the advancing edge of each tumour was performed and analysed using SPSS software against clinicopathological variables to determine statistical significance. Survival analysis was assessed using Kaplan-Meier curves.

Results: Overexpression of CUX2 was significantly (p <0.05) associated with better prognostication, such as lower grade, lower final stage, lower nodal stage, fewer recurrences, and metastases. When looking at MMR proficient and deficient cases it was also determined that there was significant correlation in the MMR proficient tumours with increased CUX2 expression at the advancing edge, correlating with better prognostic outcomes. Though survival advantage was observed for higher CUX2 expression at the advancing edge on univariate analysis, the significance was lost on multivariate analysis.

Conclusion: CUX2 appears to have a protective role in CRC, with loss related to poor prognostic variables. This seems to be a tissue specific effect within CRC as compared to other tumours (2) or may potentially be related to isoforms. Further research, including functional studies, would be necessary to fully delineate the specific role that CUX2 plays in CRC.
 

Background: The incidence of young people <50 years old with colorectal cancer (CRC), termed as early onset colorectal cancer (EOCRC). is nearly 30% of the total CRC patients in Indonesia. Lynch syndrome (LS) is a hereditary type of CRC that is associated with a younger age of onset.
Purpose: Aim of this study was to investigate the frequency of LS in Indonesian CRC patients and its association with oncogenic mutation of KRAS and PIK3CA.
Methods: Archival (2016–2019) formalin-fixed, paraffin-embedded (FFPE) tumour tissues from CRC patients were collected from Sardjito General Hospital Yogyakarta, Indonesia. The previously described high-resolution melting (HRM)-based test called Nottingham Lynch Syndrome Test (N_LyST) was used in this project. The test consisted of five mononucleotide microsatellite markers (BAT25, BAT26, BCAT25, MYB, EWSR1), BRAF V600E mutation and MLH1 region C promoter methylation. Additionally, KRAS (exon 2 and 3), and PIK3CA (exon 9 and 20) were also tested.
Results: There was 50/231 (21.65%) EOCRC cases, with 15/50 (30%) regarded as MSI, as opposed to 29/181 (16.02%) within the older group. In total, 32/231 patients (13.85%) were classified as “Probable Lynch” (MSI, BRAF wildtype and MLH1 promoter unmethylated), which enriched in EOCRC as compared to older patients (24% vs. 11.05%, p=0.035). Nonetheless, 30/50 (76.00%) cases among EOCRC were non-LS (sporadic) and significantly associated with left-sided tumour. Overall survival of both “Probable Lynch” and non-LS (sporadic) groups (n=227) was comparable (p=0.59). KRAS mutation was significantly associated with LS status in 26/32 (81.25%) samples. PIK3CA mutation was present in a higher proportion in LS samples of 19/32 (59.38%), but not statistically significant. KRAS and PIK3CA mutations did not significantly affect overall survival (120 months) in LS and non-LS patients.
Conclusions: There is higher frequency of LS among CRC patients in Indonesia and associated with high frequency of KRAS and PIK3CA mutation.