Automated Prostate Cancer Identification Facilitates Prognosis Marker Assessment in 11,845 Prostate Cancers Using Artificial Intelligence and BLEACH&STAIN Multiplex Fluorescence Immunohistochemistry

Introduction/Objective

Although most prostate cancers behave in an indolent manner, a small proportion is highly aggressive. To evaluate the patient’s risk, several prognosis parameters, that can be accompanied by a high interobserver variability has been established. A reproducible prognostic evaluation is lacking.

Methods/Case Report

To enable automated prognosis marker quantification, we have developed and validated a framework for automated prostate cancer detection that comprises three different artificial intelligence analysis steps and an algorithm for cell-distance analysis of BLEACH&STAIN multiplex fluorescence immunohistochemistry (mfIHC). We have used the analysis framework to measure PSA, PSMA, INSM1, AR, Ki-67, CD56, Chromogranin A, Synaptophysin, CD8 in a cohort of 11,845 prostate cancers.

Results (if a Case Study enter NA)

The Ki-67 labeling index provided the strongest prognostic information among all analyzed prognosis marker in 11,845 successfully analyzed prostate cancers (p<0.001 each). The combined analysis of the Ki67-LI and Gleason grades obtained on identical tissue spots showed that the Ki67-LI added significant additional prognostic information in case of classical ISUP grades (AUC:0.82 [p=0.002]) and quantitative Gleason grades (AUC:0.83 [p=0.018]). Several combinations of these 8 prognosis markers were combined to prognosis scores and used for unsupervised clustering to identify a proportion of prostate cancers with a particularly poor prognosis (p<0.001 each).

Conclusion

Automated prostate cancer identification enables fully automated prognosis marker assessment in routine clinical practice using deep learning and BLEACH&STAIN mfIHC.

Automated prognosis marker assessment in 2,004 breast cancers using an artificial intelligence-based framework for BLEACH&STAIN multiplex fluorescence immunohistochemistry

Introduction/Objective

Introduction: Prognostic markers in routine clinical practice of breast cancer are currently assessed using multi-gene panels. However, the fluctuating tumor purity can reduce the predictive value of such tests. Immunohistochemistry holds the potential for a better risk assessment.

Methods/Case Report

Methods: To enable automated prognosis marker detection (i.e. HER2, GATA3, progesterone-[PR], estrogen- [ER], and androgen receptor [AR], TOP2A, Ki-67, TROP2), we have developed and validated a framework for automated breast cancer identification, which comprises three different artificial intelligence analysis steps and an algorithm for cell-distance analysis of 11 + 1 marker BLEACH&STAIN multiplex fluorescence immunohistochemistry (mfIHC) staining in 2,004 breast cancers.

Results (if a Case Study enter NA)

Results: The optimal distance between Myosin+ basal cells and benign panCK+ cells was identified as 25 µm and used to exclude benign glands from the analysis combined with several deep learning-based algorithms. Our framework discriminated normal glands from malignant glands with an AUC of 0.96. The accuracy of the approach was also validated by well-characterized biological findings, such as the identification of 13% HER2+, 73% PR+/ER+, and 14 triple negative cases. Furthermore, the automated assessment of GATA3, PR, ER, TOP2A-LI, Ki-67-LI and TROP2 was significantly liked to the tumor grade (p<0.001each). Furthermore, a high expression level of HER2, GATA3, PR, and ER was associated with a prolonged overall survival (p≥0.002 each).

Conclusion

Conclusion: A deep learning-based framework for automated breast cancer identification using BLEACH&STAIN multiplex fluorescence IHC facilitates automated prognosis marker quantification in breast cancer.

Cadherin-16 (CDH16) Immunohistochemistry: A Novel Diagnostic Tool for Renal Cell Carcinoma and Papillary Carcinomas of the Thyroid

Introduction/Objective

Introduction: Cadherin-16 (CDH16), also termed kidney specific cadherin (ksp-cadherin), is a membrane-associated glycoprotein with a role in the embryonal development of tubules in kidney and thyroid. Downregulation of CDH16 RNA was found in papillary carcinomas of the thyroid.

Methods/Case Report

Methods: A set of tissue microarrays containing 14,978 samples from 149 different tumor types and subtypes as well as 608 samples of 76 different normal tissue types was analyzed by immunohistochemistry to determine the expression of CDH16 in cancer and to assess the diagnostic utility of immunohistochemical CDH16 analysis.

Results (if a Case Study enter NA)

Results: Among normal tissues, a membranous CDH16 immunostaining predominated in thyroid, kidney, cauda epididymis, and in mesonephric remnants. In the thyroid, CDH16 staining was seen in 100% of normal samples, 83% of follicular adenomas, 58% of follicular carcinomas, but in only 9% of papillary carcinomas (p<0.0001). Among non-thyroidal tumors, CDH16 positivity was particularly frequent in nephrogenic adenomas (100%), oncocytomas (98%), chromophobe (97%), clear cell (85%), and papillary (76%) renal cell carcinomas (RCCs), clear cell (56%), mucinous (36%), and endometroid (16%) carcinomas as well as carcinosarcomas (18%) of the ovary, adenocarcinomas of the cervix uteri (40%), serous (33%), clear cell (33%), and endometroid carcinomas (18%) of the endometrium and in various subtypes of neuroendocrine neoplasms (4-26%).

Conclusion

Given the massive loss of CDH16 expression in >90% of papillary carcinomas of the thyroid, CDH16 is a highly useful diagnostic marker for these tumors. CDH16 immunohistochemistry is also useful for the identification of nephrogenic adenomas and the distinction of renal cell carcinomas from other neoplasms.

A Comparative Analysis of CPA1, bcl10 and Chymotrypsin for the Distinction of Pancreatic Acinar Cell Carcinomas

Introduction/Objective

Introduction: Pancreatic acinar cell carcinoma (PACC) is a rare tumor of the pancreas with an intermediate prognosis as compared to pancreatic neuroendocrine tumors (PNE) and pancreatic ductal adenocarcinoma (PDAC) from which it may be difficult to distinguish by morphology alone.

Methods/Case Report

Methods: To study was the efficiency of immunohistochemical markers, 18 PACCs, 531 PDACs, 64 PNEs, 117 extra pancreatic neuroendocrine neoplasms (EPNN), 826 colorectal carcinomas (CRC) and 252 gastric carcinomas (GC) were analyzed with antibodies for CPA1 (MSVA-601M), bcl10 (Santa Cruz sc5273), and chymotrypsin (Biorad 2100-0657) in a tissue microarray format.

Results (if a Case Study enter NA)

Results: CPA1 was positive in 18 of 18 (100%) of PACCs, 0 of 49 (0%) of PNEs, 0 of 88 (0%) of EPNNs, 10 of 404 (2.5%) of CRCs, and 0 of 178 (0%) of GCs. Chymotrypsin was positive in 16 (87,5%) PACCs, 1 (2%) PNEs, 2 (2.3%) EPNNs, 10 (2.5%) CRCs, and 1 (0.6%) GCs. Bcl10 was positive in 18 (100%) PACCs, 2 (4.1%) PNEs, 5 (1%) EPNNs, 109 (27%) CRCs, and 18 (10%) GCs. These data resulted in a sensitivity and specificity of 100%/99.2% for CPA1, 100%/88.4% for bcl10, and 94.4%/98.6% for chymotrypsin.

Conclusion

CPA1 and chymotrypsin are both highly specific and sensitive for ACC while bcl10 is sensitive but has markedly lower specificity. Because all “false positive” cases identified by CPA1 were CRCs that only showed a positive staining in goblet cells and an identical staining pattern was observed in all these cases for chymotrypsin and bcl10, a pancreatic origin of the mucus in these goblet cells is concluded.

Semi-automated validation and quantification of CTLA-4 in 90 different Tumor entities using multiple antibodies and artificial intelligence

Introduction/Objective

Introduction: CTLA-4 is an inhibitory immune checkpoint receptor and a negative regulator of anti-tumor T-cell function. This study aimed at a comparative analysis of CTLA-4+ entities. cells between different tumor

Methods/Case Report

Methods: To quantify CTLA-4+ cells, 4,582 tumor samples from 90 different tumor entities as well as 608 samples of 76 different normal tissue types were analyzed by immunohistochemistry in a tissue microarray format. Two different antibody clones (MSVA-152R and CAL49) were validated and quantified using a deep learning framework for automated exclusion of unspecific immunostaining.

Results (if a Case Study enter NA)

Results: Comparing both CTLA-4 antibodies revealed a clone dependent unspecific staining pattern in adrenal cortical adenoma (63%) for MSVA-152R and in pheochromocytoma (67%) as well as hepatocellular carcinoma (36%) for CAL49. After automated exclusion of non-specific staining reaction (3.6%), a strong correlation was observed for the densities of CTLA-4+ lymphocytes obtained by both antibodies (r=0.87; p<0.0001). The mean density of CTLA-4+cells was 674±1482 cells/ mm2 and ranged from 71±175 cells/mm2 in leiomyoma to 5916±3826 cells/mm2 in Hodgkin’s lymphoma. Within epithelial tumors, the density of CTLA-4+ lymphocytes were higher in squamous cell (421±467 cells/ mm2) and urothelial carcinomas (419±347 cells/ mm2) than in adenocarcinomas (269±375 cells/ mm2) and renal cell neoplasms (256±269 cells/ mm2). A high CTLA-4+ cell density was linked to low pT category (p<0.0001), absent lymph node metastases (p=0.0354), and PD-L1 expression in tumor cells or inflammatory cells (p<0.0001 each). A high CTLA-4/CD3-ratio was linked to absent lymph node metastases (p=0.0295) and to PD-L1 positivity on immune cells (p<0.0026).

Conclusion

Marked differences exist in the number of CTLA-4+ lymphocytes between tumors. Analyzing two independent antibodies by a deep learning framework can facilitate automated quantification of immunohistochemically analyzed target proteins such as CTLA-4.

P02.03 Automated cell type specific PD-L1 quantification by artificial intelligence using high throughput bleach & stain 15-marker multiplex fluorescence immunohistochemistry in human cancers

Background

The quantification of PD-L1 (programmed cell death ligand 1) has been used to predict patient’s survival, to characterize the tumor immune microenvironment, and to predict response to immune checkpoint therapies. However, a framework to assess the PD-L1 status with a high interobserver reproducibility on tumor cells and different types of immune cells has yet to be established.

Materials and Methods

To study the impact of PD-L1 expression on the tumor immune microenvironment and patient outcome, a framework for fully automated PD-L1 quantification on tumor cells and immune cells was established and validated. Automated PD-L1 quantification was facilitated by incorporating three different deep learning steps for the analysis of more than 80 different neoplasms from more than 10’000 tumor specimens using a bleach & stain 15-marker multiplex fluorescence immunohistochemistry panel (i.e., PD-L1, PD-1, CTLA-4, panCK, CD68, CD163, CD11c, iNOS, CD3, CD8, CD4, FOXP3, CD20, Ki67, CD31). Clinicopathological parameter were available for more than 30 tumor entities and overall survival data were available for 1517 breast cancer specimens.

Results

Comparing the automated deep-learning based PD-L1 quantification with conventional brightfield PD-L1 data revealed a high concordance in tumor cells (p<0.0001) as well as immune cells (p<0.0001) and an accuracy of the automated PD-L1 quantification ranging from 90% to 95.2%. Across all tumor entities, the PD-L1 expression level was significantly higher in distinct macrophage/dendritic cell (DC) subsets (identified by CD68, CD163, CD11c, iNOS; p<000.1) and in macrophages/DCs located in the Stroma (p<0.0001) as compared to intratumoral macrophages/DC subsets. Across all different tumor entities, the PD-L1 expression was highly variable and distinct PD-L1 driven immune phenotypes were identified based on the PD-L1 intensity on both tumor and immune cells, the distance between non-exhausted T-cell subsets (i.e. PD-1 and CTLA-4 expression on CD3+CD8+ cytotoxic T-cells, CD3+CD4+ T-helper cells, CD3+CD4+FOXP3+ regulatory T-cells) and tumor cells as well as macrophage/(DC) subtypes. In breast cancer, the PD-L1 fluorescence intensity on tumor cells showed a significantly higher predictive performance for overall survival with an area under receiver operating curves (AUC) of 0.72 (p<0.0001) than the percentage of PD-L1+ tumor cells (AUC: 0.54). In PD-L1 positive as well as negative breast cancers a close spatial relationship between T- cell subsets (CD3+CD4±CD8±FOXP3±PD-1±CTLA-4±) and Macrophage/DC subsets (CD68±CD163±CD11c±iNOS) was found prognostic relevant (p<0.0001).

Conclusions

In conclusion, multiplex immunofluorescence PD-L1 assessment provides cutoff-free/continuous PD-L1 data which are superior to the conventional percentage of PD-L1+ tumor cells and of high prognostic relevance. The combined analysis of spatial PD-L1/PD-1 data and more than 20 different immune cell subtypes of the immune tumor microenvironment revealed distinct PD-L1 immune phenotypes.

Disclosure Information

N.C. Blessin: None. E. Bady: None. T. Mandelkow: None. C. Yang: None. J. Raedler: None. R. Simon: None. C. Fraune: None. M. Lennartz: None. S. Minner: None. E. Burandt: None. D. Höflmayer: None. G. Sauter: None. S.A. Weidemann: None.

MMR Deficiency is Homogeneous in Pancreatic Carcinoma and Associated with High Density of Cd8-Positive Lymphocytes

BACKGROUND

Microsatellite instability (MSI) has emerged as a predictive biomarker for immune checkpoint inhibitor therapy. Cancer heterogeneity represents a potential obstacle for the analysis of predicitive biomarkers. MSI has been reported in pancreatic cancer, but data on the possible extent of intratumoral heterogeneity are lacking.

METHODS

To study MSI heterogeneity in pancreatic cancer, a tissue microarray (TMA) comprising 597 tumors was screened by immunohistochemistry with antibodies for the mismatch repair (MMR) proteins MLH1, PMS2, MSH2, and MSH6.

RESULTS

In six suspicious cases, large section immunohistochemistry and microsatellite analysis (Bethesda panel) resulted in the identification of 4 (0.8%) validated MSI cases out of 480 interpretable pancreatic ductal adenocarcinomas. MSI was absent in 55 adenocarcinomas of the ampulla of Vater and 7 acinar cell carcinomas. MMR deficiency always involved MSH6 loss, in three cases with additional loss of MSH2 expression. Three cancers were MSI-high and one case with isolated MSH6 loss was MSS in PCR analysis. The analysis of 44 cancer-containing tumor blocks revealed that the loss of MMR protein expression was always homogeneous in affected tumors. Automated digital image analysis of CD8 immunostaining demonstrated markedly higher CD8 + tumor infiltrating lymphocytes in tumors with (mean = 685, median = 626) than without (mean = 227; median = 124) MMR deficiency (p < 0.0001), suggesting a role of MSI for immune response.

CONCLUSIONS

Our data suggest that MSI occurs early in a small subset of ductal adenocarcinomas of the pancreas and that immunohistochemical MMR analysis on limited biopsy or cytology material may be sufficient to estimate MMR status of the entire cancer mass.