Peters, Ulrike

Abstract We used deep learning (DL) to predict multiple genetic biomarkers of colorectal cancer (CRC) from routine histopathology images, aiming to identify morphological changes associated with the mutational profile of tumors. A new Transformer-based prediction model was developed which can simultaneously predict multiple biomarkers from a single histopathology image. We compared the performance of the multi-target approach to conventional single-target models. We analyzed five patient cohorts from the Genetics and Epidemiology of CRC Consortium (GECCO) with colorectal cancer (N=1,385 patients). Our model was trained on an internal data set (N=739) to predict the presence of >200 genetic alterations, focusing on genes with non-silent mutations and mutational signatures, which were assessed using targeted sequencing. We used a 7-fold cross validation and validated the model on an independent, external data set (N=646). Morphological features of microsatellite instability (MSI) are detectable by DL models in histopathology images, making MSI a potential confounding factor. Thus, we also assessed whether our model can predict genetic alterations regardless of MSI status, or whether the model identifies the MSI phenotype. For the majority of biomarkers, the multi-target transformer reached a higher prediction performance as measured by the mean Area Under the Receiver Operating Characteristic (AUROC) Curve and lower standard deviation than single-target DL models. For the external validation set, the mean AUROC of the multi-target versus the single-target model was 0.78 (+/- 0.01) versus 0.72 (+/- 0.06) for BRAF mutations, 0.88 (+/- 0.01) versus 0.86 (+/- 0.03) for hypermutation status, 0.94 (+/- 0.01) versus 0.91 (+/- 0.02) for MSI, 0.86 (+/- 0.01) versus 0.80 (+/- 0.05) for RNF43 mutations and 0.72 (+/- 0.02) versus 0.69 (+/- 0.05) for TP53 mutations, respectively. However, our analysis of the individual prediction scores suggests that the model does not identify the phenotype of the alterations themselves. Instead, the model seems to determine the status of specific genetic alterations based on their correlation with MSI. Consequently, biomarkers which are associated with microsatellite stability have prediction scores negatively correlated with predicted MSI scores. Conversely, biomarkers which are associated with MSI have prediction scores positively correlated with predicted MSI scores. Our results demonstrate potential benefits of multi-target DL models in improving the predictive power and efficacy of histology-based biomarker extraction compared to single-target DL models. However, our results crucially highlight the importance of considering potential confounders, such as MSI status in CRC, and emphasize the need for extensive analysis beyond AUROC in studies focusing on DL-based biomarker detection. Citation Format: Marco Gustav, Marko van Treeck, Zunamys I. Carrero, Chiara M. Loeffler, Nic G. Reitsam, Bruno Märkl, Asier Rabasco Meneghetti, Lisa A. Boardman, Amy J. French, Ellen L. Goode, Andrea Gsur, Stefanie Brezina, Marc J. Gunter, Neil Murphy, Paul Limburg, Stephen Thibodeau, Sebastian Foersch, Robert Steinfelder, Tabitha Harrison, Ulrike Peters, Amanda Phipps, Jakob N. Kather. Assessing microsatellite instability dominance in colorectal cancer phenotype: A multi-study initiative using multi-target transformers for genomic biomarker prediction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4924.

Abstract Background: The genotoxin colibactin causes a tumor single-base substitution (SBS) mutational signature, SBS88. It is unknown whether epidemiologic factors’ association with colorectal cancer risk and survival differs by SBS88. Methods: Within the Genetic Epidemiology of Colorectal Cancer Consortium and Colon Cancer Family Registry, we measured SBS88 in 4,308 microsatellite stable/microsatellite instability low tumors. Associations of epidemiologic factors with colorectal cancer risk by SBS88 were assessed using multinomial regression (N = 4,308 cases, 14,192 controls; cohort-only cases N = 1,911), and with colorectal cancer–specific survival using Cox proportional hazards regression (N = 3,465 cases). Results: 392 (9%) tumors were SBS88 positive. Among all cases, the highest quartile of fruit intake was associated with lower risk of SBS88-positive colorectal cancer than SBS88-negative colorectal cancer [odds ratio (OR) = 0.53, 95% confidence interval (CI) 0.37-0.76; OR = 0.75, 95% CI 0.66–0.85, respectively, Pheterogeneity = 0.047]. Among cohort studies, associations of body mass index (BMI), alcohol, and fruit intake with colorectal cancer risk differed by SBS88. BMI ≥30 kg/m2 was associated with worse colorectal cancer–specific survival among those SBS88-positive [hazard ratio (HR) = 3.40, 95% CI 1.47–7.84], but not among those SBS88-negative (HR = 0.97, 95% CI 0.78–1.21, Pheterogeneity = 0.066). Conclusions: Most epidemiologic factors did not differ by SBS88 for colorectal cancer risk or survival. Higher BMI may be associated with worse colorectal cancer–specific survival among those SBS88-positive; however, validation is needed in samples with whole-genome or whole-exome sequencing available. Impact: This study highlights the importance of identification of tumor phenotypes related to colorectal cancer and understanding potential heterogeneity for risk and survival.

AbstractWe conducted the first large genome‐wide association study to identify novel genetic variants that predict better (or poorer) prognosis in colorectal cancer patients receiving standard first‐line oxaliplatin‐based chemotherapy vs chemotherapy without oxaliplatin. We used data from two phase III trials, NCCTG N0147 and NCCTG N9741 and a population‐based patient cohort, DACHS. Multivariable Cox proportional hazards models were employed, including an interaction term between each SNP and type of treatment for overall survival (OS) and progression‐free survival. The analysis was performed for studies individually, and the results were combined using fixed‐effect meta‐analyses separately for resected stage III colon cancer (3098 patients from NCCTG N0147 and 549 patients from DACHS) and mCRC (505 patients from NCCTG N9741 and 437 patients from DACHS). We further performed gene‐based analysis as well as in silico bioinformatics analysis for CRC‐relevant functional genomic annotation of identified loci. In stage III colon cancer patients, a locus on chr22 (rs11912167) was associated with significantly poorer OS after oxaliplatin‐based chemotherapy vs chemotherapy without oxaliplatin (Pinteraction < 5 × 10−8). For mCRC patients, three loci on chr1 (rs1234556), chr12 (rs11052270) and chr15 (rs11858406) were found to be associated with differential OS (P < 5 × 10−7). The locus on chr1 located in the intronic region of RCSD1 was replicated in an independent cohort of 586 mCRC patients from ALGB/SWOG 80405 (Pinteraction = .04). The GWA gene‐based analysis yielded for RCSD1 the most significant association with differential OS in mCRC (P = 6.6 × 10−6). With further investigation into its biological mechanisms, this finding could potentially be used to individualize first‐line treatment and improve clinical outcomes.

Abstract Background: DEPendency of association on the number of Top Hits (DEPTH) is an approach to identify candidate susceptibility regions by considering the risk signals from overlapping groups of sequential variants across the genome. Methods: We applied a DEPTH analysis using a sliding window of 200 SNPs to colorectal cancer data from the Colon Cancer Family Registry (CCFR; 5,735 cases and 3,688 controls), and Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO; 8,865 cases and 10,285 controls) studies. A DEPTH score &gt; 1 was used to identify candidate susceptibility regions common to both analyses. We compared DEPTH results against those from conventional genome-wide association study (GWAS) analyses of these two studies as well as against 132 published susceptibility regions. Results: Initial DEPTH analysis revealed 2,622 (CCFR) and 3,686 (GECCO) candidate susceptibility regions, of which 569 were common to both studies. Bootstrapping revealed 40 and 49 candidate susceptibility regions in the CCFR and GECCO data sets, respectively. Notably, DEPTH identified at least 82 regions that would not be detected using conventional GWAS methods, nor had they been identified by previous colorectal cancer GWASs. We found four reproducible candidate susceptibility regions (2q22.2, 2q33.1, 6p21.32, 13q14.3). The highest DEPTH scores were in the human leukocyte antigen locus at 6p21 where the strongest associated SNPs were rs762216297, rs149490268, rs114741460, and rs199707618 for the CCFR data, and rs9270761 for the GECCO data. Conclusions: DEPTH can identify candidate susceptibility regions for colorectal cancer not identified using conventional analyses of larger datasets. Impact: DEPTH has potential as a powerful complementary tool to conventional GWAS analyses for discovering susceptibility regions within the genome.

Abstract Background: There are conflicting data on whether nonalcoholic fatty liver disease (NAFLD) is associated with susceptibility to pancreatic cancer. Using Mendelian randomization (MR), we investigated the relationship between genetic predisposition to NAFLD and risk for pancreatic cancer. Methods: Data from genome-wide association studies (GWAS) within the Pancreatic Cancer Cohort Consortium (PanScan; cases n = 5,090, controls n = 8,733) and the Pancreatic Cancer Case Control Consortium (PanC4; cases n = 4,163, controls n = 3,792) were analyzed. We used data on 68 genetic variants with four different MR methods [inverse variance weighting (IVW), MR-Egger, simple median, and penalized weighted median] separately to predict genetic heritability of NAFLD. We then assessed the relationship between each of the four MR methods and pancreatic cancer risk, using logistic regression to calculate ORs and 95% confidence intervals (CI), adjusting for PC risk factors, including obesity and diabetes. Results: No association was found between genetically predicted NAFLD and pancreatic cancer risk in the PanScan or PanC4 samples [e.g., PanScan, IVW OR, 1.04; 95% confidence interval (CI), 0.88–1.22; MR-Egger OR, 0.89; 95% CI, 0.65–1.21; PanC4, IVW OR, 1.07; 95% CI, 0.90–1.27; MR-Egger OR, 0.93; 95% CI, 0.67–1.28]. None of the four MR methods indicated an association between genetically predicted NAFLD and pancreatic cancer risk in either sample. Conclusions: Genetic predisposition to NAFLD is not associated with pancreatic cancer risk. Impact: Given the close relationship between NAFLD and metabolic conditions, it is plausible that any association between NAFLD and pancreatic cancer might reflect host metabolic perturbations (e.g., obesity, diabetes, or metabolic syndrome) and does not necessarily reflect a causal relationship between NAFLD and pancreatic cancer.

Abstract Background: The genotoxin colibactin, produced by Escherichia coli containing the pks island, is associated with a tumor single base substitution (SBS) mutational signature SBS88. It is unknown whether lifestyle and environmental factors related to gut dysbiosis impact colorectal cancer (CRC) risk and survival differently by colibactin signature status. Methods: Utilizing the resources of the Genetic Epidemiology of Colorectal cancer Consortium and Colon Cancer Family Registry (GECCO-CCFR), a consortium including both case-control and cohort study designs, SBS88 tumor mutational signature was measured in 4308 CRC cases who were microsatellite stable (MSS) and had at least five somatic single nucleotide variants for signature decomposition. Associations of 13 lifestyle and environmental factors with CRC risk by colibactin signature status were assessed using logistic and multinomial regression; associations with CRC survival were assessed using Cox proportional hazards regression. Results: 392 (9%) of the 4308 MSS tumors were positive for SBS88. Having a BMI ≥30 kg/m2 (versus a BMI 18.5-24.9 kg/m2) was strongly associated with worse CRC-specific survival among those with the SBS88 signature, but not among those with SBS88 negative CRC [further adjusted hazard ratio (HR) = 3.40 95% confidence interval (CI) 1.47, 7.84, and HR = 0.97 95% CI 0.78, 1.21, respectively, P for heterogeneity = 0.066]. Associations of fruit intake with CRC risk may be differential by SBS88 status. In minimally adjusted models, those in the highest quartile of fruit intake (versus the first quartile) had 0.53 (95% CI 0.37, 0.76) times the risk of SBS88 positive CRC, as opposed to 0.75 (95% CI 0.66, 0.85) times the risk of SBS88 negative CRC (P for heterogeneity = 0.047). Among cohort studies specifically, associations between CRC risk and BMI, alcohol, and fruit intake also differed by SBS88 status (P for heterogeneity ranging 0.02-0.05). Conclusions: Higher BMI may be associated with worse CRC-specific survival among those with SBS88 positive tumors. BMI, alcohol, and dietary factors may be differentially associated with CRC risk based on SBS88 status. Citation Format: Claire Elizabeth Thomas, Peter Georgeson, Conghui Qu, Robert S. Steinfelder, Daniel D. Buchanan, Mark A. Jenkins, Andrea Gsur, Jane C. Figueiredo, Polly A. Newcomb, Peter T. Campbell, Christina Newton, Caroline Y. Um, Lisa Boardman, Victor Moreno, Jenny Chang-Claude, Michael Hoffmeister, Marc J. Gunter, Andrew T. Chan, Shuji Ogino, Ellen L. Goode, Brigid Lynch, Yohannes A. Melaku, Steven Gallinger, Sonja I. Berndt, Wen-Yi Huang, Ulrike Peters, Amanda I. Phipps. Lifestyle and environmental factors in relation to colorectal cancer risk and survival by colibactin tumor mutational signature status [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6453.

Abstract Colorectal cancer risk can be impacted by genetic, environmental, and lifestyle factors, including diet and obesity. Gene-environment interactions (G × E) can provide biological insights into the effects of obesity on colorectal cancer risk. Here, we assessed potential genome-wide G × E interactions between body mass index (BMI) and common SNPs for colorectal cancer risk using data from 36,415 colorectal cancer cases and 48,451 controls from three international colorectal cancer consortia (CCFR, CORECT, and GECCO). The G × E tests included the conventional logistic regression using multiplicative terms (one degree of freedom, 1DF test), the two-step EDGE method, and the joint 3DF test, each of which is powerful for detecting G × E interactions under specific conditions. BMI was associated with higher colorectal cancer risk. The two-step approach revealed a statistically significant G×BMI interaction located within the Formin 1/Gremlin 1 (FMN1/GREM1) gene region (rs58349661). This SNP was also identified by the 3DF test, with a suggestive statistical significance in the 1DF test. Among participants with the CC genotype of rs58349661, overweight and obesity categories were associated with higher colorectal cancer risk, whereas null associations were observed across BMI categories in those with the TT genotype. Using data from three large international consortia, this study discovered a locus in the FMN1/GREM1 gene region that interacts with BMI on the association with colorectal cancer risk. Further studies should examine the potential mechanisms through which this locus modifies the etiologic link between obesity and colorectal cancer. Significance: This gene-environment interaction analysis revealed a genetic locus in FMN1/GREM1 that interacts with body mass index in colorectal cancer risk, suggesting potential implications for precision prevention strategies.

Abstract Background: Colorectal cancer (CRC) incidence has increased in those 50 years or younger compared to older individuals in recent decades. While the intra-tumoral microbiome is a potent facilitator of tumorigenesis, the role of the microbiome in early onset (eo)CRC is unclear. Identification of the intra-tumoral microbiome may enhance prediction of adverse outcomes to improve survival of CRC overall age groups and eoCRC. Methods: We performed targeted sequencing of tumor and matched normal DNA samples from 3695 colorectal cancer cases within the Genetics and Colorectal Cancer Consortium (GECCO) and the Colon Cancer Family Registry (CCFR). Tumor DNA was extracted from formalin fixed paraffin embedded (FFPE) or fresh frozen tissue. Normal tissue DNA was extracted from blood or normal surrounding colorectal tissue. We designed probes for the bacterial pathogens enterotoxigenic Bacteroides fragilis (ETBF), polyketide synthase positive E. coli (pks+ EC), and Fusobacterium nucleatum (Fn). We used logistic regression to assess tumor-associated bacteria and clinical parameters, tumor molecular characteristics, and tumor stage adjusted for age, sex, and study and a Cox promotional hazard to assess association with CRC-specific survival, overall and in eoCRC, adjusted for age at diagnosis, sex, mutational burden, and microsatellite instability (MSI) and stratified baseline hazards by study. Survival data were available for 2357 and 308 patients, overall and in eoCRC, respectively. Results: The prevalence of different bacteria in tumors differed slightly overall vs. in eoCRC tumors: ETBF (5.2%, 6.3%), pks+ EC (12.9%, 14.8%), and Fn (10.4%, 8.2%). Overall, men had a lower probability of tumors with Fn (odds ratio (OR)=0.8, 95% confidence interval (CI)=0.64-0.99, p=0.04) but higher odds of tumors with pks+ EC (OR=1.38, CI=1.14-1.68, p=0.001), compared to women. The odds of prevalent Fn were higher for proximal vs distal tumors (OR=1.76, CI=1.38-2.26, p&lt;0.001), but there was no association between tumor site and pks+ EC or ETBF. MSI-high tumors were more likely to carry Fn (OR=2.79, CI=2.08-3.75, p&lt;0.001) and less likely to carry pks+ EC (OR=0.63, CI=0.44-0.88, p=0.008) when compared to patients with MSS tumors. In patients with non-hypermutated tumors, we observed that pks+ EC positive tumors were associated with better CRC-specific survival (HR=0.73, CI=0.57-0.93, p=0.01), whereas ETBF positive tumors (HR=1.99, CI=1,18-3.36,p=0.009) and Fn positive tumors (HR=1.33, CI=1.04-1.72, p=0.03) were associated with poorer survival. In eo-CRC patients, Fn positive status was associated with poorer survival (HR=2.4, CI= 1.056 - 5.492, p=0.026). Conclusions: We showed that the intra-tumoral microbiome differs in distribution across patient populations and patient survival over all age groups and, specifically, in eoCRC. Citation Format: Meredith A. Hullar, Keith R. Curtis, Tabitha Harrison, Yi Lin, Robert Steinfelder, Sonja I. Berndt, Daniel D. Buchanan, Andrew T. Chan, David A. Drew, Jane Figueiredo, Amy J. French, Tabitha A. Harrison, Mark A. Jenkins, Johannes Melaku, Victor Moreno, Tomotaka Ugai, Shuji Ogino, Conghui Qu, Chenxu Qu, Steven Thibodeau, Syed Zaidi, Amanda I. Phipps, Ulrike Peters, on behalf of GECCO and CCFR investigators. Evaluation of intra-tumoral pks+ E. coli, enterotoxigenic B. fragilis and Fusobacterium nucleatum, overall and in early onset disease, in colorectal cancer cases [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 3039.

Abstract Background Mendelian randomization (MR) studies are susceptible to metadata errors (e.g. incorrect specification of the effect allele column) and other analytical issues that can introduce substantial bias into analyses. We developed a quality control (QC) pipeline for the Fatty Acids in Cancer Mendelian Randomization Collaboration (FAMRC) that can be used to identify and correct for such errors. Methods We collated summary association statistics from fatty acid and cancer genome-wide association studies (GWAS) and subjected the collated data to a comprehensive QC pipeline. We identified metadata errors through comparison of study-specific statistics to external reference data sets (the National Human Genome Research Institute-European Bioinformatics Institute GWAS catalogue and 1000 genome super populations) and other analytical issues through comparison of reported to expected genetic effect sizes. Comparisons were based on three sets of genetic variants: (i) GWAS hits for fatty acids, (ii) GWAS hits for cancer and (iii) a 1000 genomes reference set. Results We collated summary data from 6 fatty acid and 54 cancer GWAS. Metadata errors and analytical issues with the potential to introduce substantial bias were identified in seven studies (11.6%). After resolving metadata errors and analytical issues, we created a data set of 219 842 genetic associations with 90 cancer types, generated in analyses of 566 665 cancer cases and 1 622 374 controls. Conclusions In this large MR collaboration, 11.6% of included studies were affected by a substantial metadata error or analytical issue. By increasing the integrity of collated summary data prior to their analysis, our protocol can be used to increase the reliability of downstream MR analyses. Our pipeline is available to other researchers via the CheckSumStats package (https://github.com/MRCIEU/CheckSumStats).

INTRODUCTION: Early-onset colorectal cancer diagnosed before the age of 50 years has been increasing. Likely reflecting the pathogenic role of the intestinal microbiome, which gradually changes across the entire colorectal length, the prevalence of certain tumor molecular characteristics gradually changes along colorectal subsites. Understanding how colorectal tumor molecular features differ by age and tumor location is important in personalized patient management. METHODS: Using 14,004 cases with colorectal cancer including 3,089 early-onset cases, we examined microsatellite instability (MSI), CpG island methylator phenotype (CIMP), and KRAS and BRAF mutations in carcinomas of the cecum, ascending colon, transverse colon, descending colon, sigmoid colon, and rectum and compared early-onset cases with later-onset cases. RESULTS: The proportions of MSI-high, CIMP-high, and BRAF -mutated early-onset tumors were lowest in the rectum (8.8%, 3.4%, and 3.5%, respectively) and highest in the ascending colon (46% MSI-high; 15% CIMP-high) or transverse colon (8.6% BRAF -mutated) (all P trend <0.001 across the rectum to ascending colon). Compared with later-onset tumors, early-onset tumors showed a higher prevalence of MSI-high status and a lower prevalence of CIMP-high status and BRAF mutations in most subsites. KRAS mutation prevalence was higher in the cecum compared with that in the other subsites in both early-onset and later-onset tumors ( P < 0.001). Notably, later-onset MSI-high tumors showed a continuous decrease in KRAS mutation prevalence from the rectum (36%) to ascending colon (9%; P trend <0.001), followed by an increase in the cecum (14%), while early-onset MSI-high cancers showed no such trend. DISCUSSION: Our findings support biogeographical and pathogenic heterogeneity of colorectal carcinomas in different colorectal subsites and age groups.