Artificial Intelligence-based Detection of FGFR3 Mutational Status Directly from Routine Histology in Bladder Cancer: A Possible Preselection for Molecular Testing?

BACKGROUND: Fibroblast growth factor receptor (FGFR) inhibitor treatment has become the first clinically approved targeted therapy in bladder cancer. However, it requires previous molecular testing of each patient, which is costly and not ubiquitously available. OBJECTIVE: To determine whether an artificial intelligence system is able to predict mutations of the FGFR3 gene directly from routine histology slides of bladder cancer. DESIGN, SETTING, AND PARTICIPANTS: We trained a deep learning network to detect FGFR3 mutations on digitized slides of muscle-invasive bladder cancers stained with hematoxylin and eosin from the Cancer Genome Atlas (TCGA) cohort (n = 327) and validated the algorithm on the "Aachen" cohort (n = 182; n = 121 pT2-4, n = 34 stroma-invasive pT1, and n = 27 noninvasive pTa tumors). OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: The primary endpoint was the area under the receiver operating curve (AUROC) for mutation detection. Performance of the deep learning system was compared with visual scoring by an uropathologist. RESULTS AND LIMITATIONS: In the TCGA cohort, FGFR3 mutations were detected with an AUROC of 0.701 (p < 0.0001). In the Aachen cohort, FGFR3 mutants were found with an AUROC of 0.725 (p < 0.0001). When trained on TCGA, the network generalized to the Aachen cohort, and detected FGFR3 mutants with an AUROC of 0.625 (p = 0.0112). A subgroup analysis and histological evaluation found highest accuracy in papillary growth, luminal gene expression subtypes, females, and American Joint Committee on Cancer (AJCC) stage II tumors. In a head-to-head comparison, the deep learning system outperformed the uropathologist in detecting FGFR3 mutants. CONCLUSIONS: Our computer-based artificial intelligence system was able to detect genetic alterations of the FGFR3 gene of bladder cancer patients directly from histological slides. In the future, this system could be used to preselect patients for further molecular testing. However, analyses of larger, multicenter, muscle-invasive bladder cancer cohorts are now needed in order to validate and extend our findings. PATIENT SUMMARY: In this report, a computer-based artificial intelligence (AI) system was applied to histological slides to predict genetic alterations of the FGFR3 gene in bladder cancer. We found that the AI system was able to find the alteration with high accuracy. In the future, this system could be used to preselect patients for further molecular testing.

Refining genotype-phenotype correlations in 304 patients with autosomal recessive polycystic kidney disease and PKHD1 gene variants.

Autosomal recessive polycystic kidney disease (ARPKD) is a severe disease of early childhood that is clinically characterized by fibrocystic changes of the kidneys and the liver. The main cause of ARPKD are variants in the PKHD1 gene encoding the large transmembrane protein fibrocystin. The mechanisms underlying the observed clinical heterogeneity in ARPKD remain incompletely understood, partly due to the fact that genotype-phenotype correlations have been limited to the association of biallelic null variants in PKHD1 with the most severe phenotypes. In this observational study we analyzed a deep clinical dataset of 304 patients with ARPKD from two independent cohorts and identified novel genotype-phenotype correlations during childhood and adolescence. Biallelic null variants frequently show severe courses. Additionally, our data suggest that the affected region in PKHD1 is important in determining the phenotype. Patients with two missense variants affecting amino acids 709-1837 of fibrocystin or a missense variant in this region and a null variant less frequently developed chronic kidney failure, and patients with missense variants affecting amino acids 1838-2624 showed better hepatic outcome. Variants affecting amino acids 2625-4074 of fibrocystin were associated with poorer hepatic outcome. Thus, our data expand the understanding of genotype-phenotype correlations in pediatric ARPKD patients and can lay the foundation for more precise and personalized counselling and treatment approaches.

Predicting Mutational Status of Driver and Suppressor Genes Directly from Histopathology With Deep Learning: A Systematic Study Across 23 Solid Tumor Types.

In the last four years, advances in Deep Learning technology have enabled the inference of selected mutational alterations directly from routine histopathology slides. In particular, recent studies have shown that genetic changes in clinically relevant driver genes are reflected in the histological phenotype of solid tumors and can be inferred by analysing routine Haematoxylin and Eosin (H&E) stained tissue sections with Deep Learning. However, these studies mostly focused on selected individual genes in selected tumor types. In addition, genetic changes in solid tumors primarily act by changing signaling pathways that regulate cell behaviour. In this study, we hypothesized that Deep Learning networks can be trained to directly predict alterations of genes and pathways across a spectrum of solid tumors. We manually outlined tumor tissue in H&E-stained tissue sections from 7,829 patients with 23 different tumor types from The Cancer Genome Atlas. We then trained convolutional neural networks in an end-to-end way to detect alterations in the most clinically relevant pathways or genes, directly from histology images. Using this automatic approach, we found that alterations in 12 out of 14 clinically relevant pathways and numerous single gene alterations appear to be detectable in tissue sections, many of which have not been reported before. Interestingly, we show that the prediction performance for single gene alterations is better than that for pathway alterations. Collectively, these data demonstrate the predictability of genetic alterations directly from routine cancer histology images and show that individual genes leave a stronger morphological signature than genetic pathways.

Comparative genomic profiling of glandular bladder tumours.

Primary glandular bladder tumours (bladder adenocarcinoma [BAC], urachal adenocarcinoma [UAC], urothelial carcinoma with glandular differentiation [UCg]) are rare malignancies with histological resemblance to colorectal adenocarcinoma (CORAD) in the majority of this subgroup. Definite case numbers are very low, molecular data are limited and the pathogenesis remains poorly understood. Therefore, this study was designed to complement current knowledge by in depth analysis of BAC (n = 12), UAC (n = 13), UCg (n = 11) and non-invasive glandular lesions (n = 19). In BAC, in addition to known alterations in TP53, Wnt, MAP kinase and MTOR pathway, mutations in SMAD4, ARID1A and BRAF were identified. Compared to published data on muscle invasive bladder cancer (BLCA) and CORAD, UCg exhibited frequent "urothelial" like alterations while BAC and UAC were characterised by a more "colorectal" like mutational pattern. Immunohistochemically, there was no evidence of DNA mismatch repair deficiency or PD-L1 tumour cell positivity in any sample. Depending on the used antibody 0-45% of BAC, 0-30% of UCg and 0% UAC cases exhibited PD-L1 expressing tumour associated immune cells. A single BAC (9%, 1/11) showed evidence of ARID1A protein loss, and two cases of UCg (20%, 2/10) showed loss of SMARCA1 and PBRM1, respectively. Taken together, our data suggest at least in part involvement of similar pathways driving tumourigenesis of adenocarcinomas like BAC, UAC and CORAD independent of their tissue origin. Alterations of TERT and FBXW7 in single cases of intestinal metaplasia further point towards a possible precancerous character in line with previous reports.

Challenges in establishing genotype-phenotype correlations in ARPKD: case report on a toddler with two severe PKHD1 mutations.

BACKGROUND: Autosomal recessive polycystic kidney disease (ARPKD) constitutes an important cause of pediatric end stage renal disease and is characterized by a broad phenotypic variability. The disease is caused by mutations in a single gene, Polycystic Kidney and Hepatic Disease 1 (PKHD1), which encodes a large transmembrane protein of poorly understood function called fibrocystin. Based on current knowledge of genotype-phenotype correlations in ARPKD, two truncating mutations are considered to result in a severe phenotype with peri- or neonatal mortality. Infants surviving the neonatal period are expected to carry at least one missense mutation. CASE-DIAGNOSIS/TREATMENT: We report on a female patient with two truncating PKHD1 mutations who survived the first 30 months of life without renal replacement therapy. Our patient carries not only a known stop mutation, c.8011C>T (p.Arg2671*), but also the previously reported c.51A>G PKHD1 sequence variant of unknown significance in exon 2. Using functional in vitro studies we have confirmed the pathogenic nature of c.51A>G, demonstrating activation of a new donor splice site in intron 2 that results in a frameshift mutation and generation of a premature stop codon. CONCLUSIONS: This case illustrates the importance of functional mutation analyses and also raises questions regarding the current belief that the presence of at least one missense mutation is necessary for perinatal survival in ARPKD.

Epidemiology of autosomal-dominant polycystic kidney disease: an in-depth clinical study for south-western Germany.

BACKGROUND: As we emerge into the genomic medicine era, the epidemiology of diseases is taken for granted. Accurate prevalence figures, especially of rare diseases (RDs, ≤50/100,000), will become even more important for purposes of health care and societal planning. We noticed that the numbers of affected individuals in regionally established registries for mainly hereditary RDs do not align with published estimated and expected prevalence figures. We therefore hypothesized that such non-population-based means overestimate RDs and sought to address this by recalculating prevalence for an important 'common' hereditary disease, autosomal-dominant polycystic kidney disease (ADPKD) whereby presumed-prevalence is 100-250/100,000 METHODS: The Else-Kroener-Fresenius-ADPKD-Study in south-west Germany with a population of 2,727,351 inhabitants was established with the cooperation of all nephrology centres. Furthermore, general practitioners, internists, urologists, human geneticists and neurosurgery centres were contacted with questionnaires for demographic, family and kidney function data. Germline-mutation screening of susceptibility genes PKD1 and PKD2 was offered. Official population data for 2010 were used for overall and kidney function-adjusted prevalence estimations. RESULTS: A total of 891 subjects, 658 index-cases and 233 relatives, aged 10-89 (mean 52), were registered, with >90% response rate, 398 by nephrologists and 493 by non-nephrologists. Molecular-genetic analyses contributed to confirmation of the diagnosis in 57%. The overall prevalence of ADPKD was 32.7/100,000 reaching a maximum of 57.3/100,000 in the 6th decade of life. CONCLUSIONS: Prevalence of ADPKD is overestimated by 2- to 5-fold and close to the limit of RDs which may be of broad clinical, logistic and policy implications.