Prediction of homologous recombination deficiency from routine histology with attention-based multiple instance learning in nine different tumor types.

BACKGROUND: Homologous recombination deficiency (HRD) is recognized as a pan-cancer predictive biomarker that potentially indicates who could benefit from treatment with PARP inhibitors (PARPi). Despite its clinical significance, HRD testing is highly complex. Here, we investigated in a proof-of-concept study whether Deep Learning (DL) can predict HRD status solely based on routine hematoxylin & eosin (H&E) histology images across nine different cancer types. METHODS: We developed a deep learning pipeline with attention-weighted multiple instance learning (attMIL) to predict HRD status from histology images. As part of our approach, we calculated a genomic scar HRD score by combining loss of heterozygosity (LOH), telomeric allelic imbalance (TAI), and large-scale state transitions (LST) from whole genome sequencing (WGS) data of n = 5209 patients across two independent cohorts. The model's effectiveness was evaluated using the area under the receiver operating characteristic curve (AUROC), focusing on its accuracy in predicting genomic HRD against a clinically recognized cutoff value. RESULTS: Our study demonstrated the predictability of genomic HRD status in endometrial, pancreatic, and lung cancers reaching cross-validated AUROCs of 0.79, 0.58, and 0.66, respectively. These predictions generalized well to an external cohort, with AUROCs of 0.93, 0.81, and 0.73. Moreover, a breast cancer-trained image-based HRD classifier yielded an AUROC of 0.78 in the internal validation cohort and was able to predict HRD in endometrial, prostate, and pancreatic cancer with AUROCs of 0.87, 0.84, and 0.67, indicating that a shared HRD-like phenotype occurs across these tumor entities. CONCLUSIONS: This study establishes that HRD can be directly predicted from H&E slides using attMIL, demonstrating its applicability across nine different tumor types.

Automated curation of large-scale cancer histopathology image datasets using deep learning.

BACKGROUND: Artificial intelligence (AI) has numerous applications in pathology, supporting diagnosis and prognostication in cancer. However, most AI models are trained on highly selected data, typically one tissue slide per patient. In reality, especially for large surgical resection specimens, dozens of slides can be available for each patient. Manually sorting and labelling whole-slide images (WSIs) is a very time-consuming process, hindering the direct application of AI on the collected tissue samples from large cohorts. In this study we addressed this issue by developing a deep-learning (DL)-based method for automatic curation of large pathology datasets with several slides per patient. METHODS: We collected multiple large multicentric datasets of colorectal cancer histopathological slides from the United Kingdom (FOXTROT, N = 21,384 slides; CR07, N = 7985 slides) and Germany (DACHS, N = 3606 slides). These datasets contained multiple types of tissue slides, including bowel resection specimens, endoscopic biopsies, lymph node resections, immunohistochemistry-stained slides, and tissue microarrays. We developed, trained, and tested a deep convolutional neural network model to predict the type of slide from the slide overview (thumbnail) image. The primary statistical endpoint was the macro-averaged area under the receiver operating curve (AUROCs) for detection of the type of slide. RESULTS: In the primary dataset (FOXTROT), with an AUROC of 0.995 [95% confidence interval [CI]: 0.994-0.996] the algorithm achieved a high classification performance and was able to accurately predict the type of slide from the thumbnail image alone. In the two external test cohorts (CR07, DACHS) AUROCs of 0.982 [95% CI: 0.979-0.985] and 0.875 [95% CI: 0.864-0.887] were observed, which indicates the generalizability of the trained model on unseen datasets. With a confidence threshold of 0.95, the model reached an accuracy of 94.6% (7331 classified cases) in CR07 and 85.1% (2752 classified cases) for the DACHS cohort. CONCLUSION: Our findings show that using the low-resolution thumbnail image is sufficient to accurately classify the type of slide in digital pathology. This can support researchers to make the vast resource of existing pathology archives accessible to modern AI models with only minimal manual annotations.

Privacy-preserving large language models for structured medical information retrieval.

Most clinical information is encoded as free text, not accessible for quantitative analysis. This study presents an open-source pipeline using the local large language model (LLM) "Llama 2" to extract quantitative information from clinical text and evaluates its performance in identifying features of decompensated liver cirrhosis. The LLM identified five key clinical features in a zero- and one-shot manner from 500 patient medical histories in the MIMIC IV dataset. We compared LLMs of three sizes and various prompt engineering approaches, with predictions compared against ground truth from three blinded medical experts. Our pipeline achieved high accuracy, detecting liver cirrhosis with 100% sensitivity and 96% specificity. High sensitivities and specificities were also yielded for detecting ascites (95%, 95%), confusion (76%, 94%), abdominal pain (84%, 97%), and shortness of breath (87%, 97%) using the 70 billion parameter model, which outperformed smaller versions. Our study successfully demonstrates the capability of locally deployed LLMs to extract clinical information from free text with low hardware requirements.

Deep learning for dual detection of microsatellite instability and POLE mutations in colorectal cancer histopathology.

In the spectrum of colorectal tumors, microsatellite-stable (MSS) tumors with DNA polymerase ε (POLE) mutations exhibit a hypermutated profile, holding the potential to respond to immunotherapy similarly to their microsatellite-instable (MSI) counterparts. Yet, due to their rarity and the associated testing costs, systematic screening for these mutations is not commonly pursued. Notably, the histopathological phenotype resulting from POLE mutations is theorized to resemble that of MSI. This resemblance not only could facilitate their detection by a transformer-based Deep Learning (DL) system trained on MSI pathology slides, but also indicates the possibility for MSS patients with POLE mutations to access enhanced treatment options, which might otherwise be overlooked. To harness this potential, we trained a Deep Learning classifier on a large dataset with the ground truth for microsatellite status and subsequently validated its capabilities for MSI and POLE detection across three external cohorts. Our model accurately identified MSI status in both the internal and external resection cohorts using pathology images alone. Notably, with a classification threshold of 0.5, over 75% of POLE driver mutant patients in the external resection cohorts were flagged as "positive" by a DL system trained on MSI status. In a clinical setting, deploying this DL model as a preliminary screening tool could facilitate the efficient identification of clinically relevant MSI and POLE mutations in colorectal tumors, in one go.

Regression-based Deep-Learning predicts molecular biomarkers from pathology slides.

Deep Learning (DL) can predict biomarkers from cancer histopathology. Several clinically approved applications use this technology. Most approaches, however, predict categorical labels, whereas biomarkers are often continuous measurements. We hypothesize that regression-based DL outperforms classification-based DL. Therefore, we develop and evaluate a self-supervised attention-based weakly supervised regression method that predicts continuous biomarkers directly from 11,671 images of patients across nine cancer types. We test our method for multiple clinically and biologically relevant biomarkers: homologous recombination deficiency score, a clinically used pan-cancer biomarker, as well as markers of key biological processes in the tumor microenvironment. Using regression significantly enhances the accuracy of biomarker prediction, while also improving the predictions' correspondence to regions of known clinical relevance over classification. In a large cohort of colorectal cancer patients, regression-based prediction scores provide a higher prognostic value than classification-based scores. Our open-source regression approach offers a promising alternative for continuous biomarker analysis in computational pathology.

Multi-omic and functional analysis for classification and treatment of sarcomas with FUS-TFCP2 or EWSR1-TFCP2 fusions.

Linking clinical multi-omics with mechanistic studies may improve the understanding of rare cancers. We leverage two precision oncology programs to investigate rhabdomyosarcoma with FUS/EWSR1-TFCP2 fusions, an orphan malignancy without effective therapies. All tumors exhibit outlier ALK expression, partly accompanied by intragenic deletions and aberrant splicing resulting in ALK variants that are oncogenic and sensitive to ALK inhibitors. Additionally, recurrent CKDN2A/MTAP co-deletions provide a rationale for PRMT5-targeted therapies. Functional studies show that FUS-TFCP2 blocks myogenic differentiation, induces transcription of ALK and truncated TERT, and inhibits DNA repair. Unlike other fusion-driven sarcomas, TFCP2-rearranged tumors exhibit genomic instability and signs of defective homologous recombination. DNA methylation profiling demonstrates a close relationship with undifferentiated sarcomas. In two patients, sarcoma was preceded by benign lesions carrying FUS-TFCP2, indicating stepwise sarcomagenesis. This study illustrates the potential of linking precision oncology with preclinical research to gain insight into the classification, pathogenesis, and therapeutic vulnerabilities of rare cancers.

The future landscape of large language models in medicine.

Large language models (LLMs) are artificial intelligence (AI) tools specifically trained to process and generate text. LLMs attracted substantial public attention after OpenAI's ChatGPT was made publicly available in November 2022. LLMs can often answer questions, summarize, paraphrase and translate text on a level that is nearly indistinguishable from human capabilities. The possibility to actively interact with models like ChatGPT makes LLMs attractive tools in various fields, including medicine. While these models have the potential to democratize medical knowledge and facilitate access to healthcare, they could equally distribute misinformation and exacerbate scientific misconduct due to a lack of accountability and transparency. In this article, we provide a systematic and comprehensive overview of the potentials and limitations of LLMs in clinical practice, medical research and medical education.

Direct prediction of Homologous Recombination Deficiency from routine histology in ten different tumor types with attention-based Multiple Instance Learning: a development and validation study.

Background: Homologous Recombination Deficiency (HRD) is a pan-cancer predictive biomarker that identifies patients who benefit from therapy with PARP inhibitors (PARPi). However, testing for HRD is highly complex. Here, we investigated whether Deep Learning can predict HRD status solely based on routine Hematoxylin & Eosin (H&E) histology images in ten cancer types. Methods: We developed a fully automated deep learning pipeline with attention-weighted multiple instance learning (attMIL) to predict HRD status from histology images. A combined genomic scar HRD score, which integrated loss of heterozygosity (LOH), telomeric allelic imbalance (TAI) and large-scale state transitions (LST) was calculated from whole genome sequencing data for n=4,565 patients from two independent cohorts. The primary statistical endpoint was the Area Under the Receiver Operating Characteristic curve (AUROC) for the prediction of genomic scar HRD with a clinically used cutoff value. Results: We found that HRD status is predictable in tumors of the endometrium, pancreas and lung, reaching cross-validated AUROCs of 0.79, 0.58 and 0.66. Predictions generalized well to an external cohort with AUROCs of 0.93, 0.81 and 0.73 respectively. Additionally, an HRD classifier trained on breast cancer yielded an AUROC of 0.78 in internal validation and was able to predict HRD in endometrial, prostate and pancreatic cancer with AUROCs of 0.87, 0.84 and 0.67 indicating a shared HRD-like phenotype is across tumor entities. Conclusion: In this study, we show that HRD is directly predictable from H&E slides using attMIL within and across ten different tumor types.

Direct image to subtype prediction for brain tumors using deep learning.

Background: Deep Learning (DL) can predict molecular alterations of solid tumors directly from routine histopathology slides. Since the 2021 update of the World Health Organization (WHO) diagnostic criteria, the classification of brain tumors integrates both histopathological and molecular information. We hypothesize that DL can predict molecular alterations as well as WHO subtyping of brain tumors from hematoxylin and eosin-stained histopathology slides. Methods: We used weakly supervised DL and applied it to three large cohorts of brain tumor samples, comprising N = 2845 patients. Results: We found that the key molecular alterations for subtyping, IDH and ATRX, as well as 1p19q codeletion, were predictable from histology with an area under the receiver operating characteristic curve (AUROC) of 0.95, 0.90, and 0.80 in the training cohort, respectively. These findings were upheld in external validation cohorts with AUROCs of 0.90, 0.79, and 0.87 for prediction of IDH, ATRX, and 1p19q codeletion, respectively. Conclusions: In the future, such DL-based implementations could ease diagnostic workflows, particularly for situations in which advanced molecular testing is not readily available.

Signaling-induced systematic repression of miRNAs uncovers cancer vulnerabilities and targeted therapy sensitivity.

Targeted therapies are effective in treating cancer, but success depends on identifying cancer vulnerabilities. In our study, we utilize small RNA sequencing to examine the impact of pathway activation on microRNA (miRNA) expression patterns. Interestingly, we discover that miRNAs capable of inhibiting key members of activated pathways are frequently diminished. Building on this observation, we develop an approach that integrates a low-miRNA-expression signature to identify druggable target genes in cancer. We train and validate our approach in colorectal cancer cells and extend it to diverse cancer models using patient-derived in vitro and in vivo systems. Finally, we demonstrate its additional value to support genomic and transcriptomic-based drug prediction strategies in a pan-cancer patient cohort from the National Center for Tumor Diseases (NCT)/German Cancer Consortium (DKTK) Molecularly Aided Stratification for Tumor Eradication (MASTER) precision oncology trial. In conclusion, our strategy can predict cancer vulnerabilities with high sensitivity and accuracy and might be suitable for future therapy recommendations in a variety of cancer subtypes.

Inheritance of the Golgi Apparatus and Cytokinesis Are Controlled by Degradation of GBF1.

Although much is known about how chromosome segregation is coupled to cell division, how intracellular organelles partition during mitotic division is poorly understood. We report that the phosphorylation-dependent degradation of the ARFGEF GBF1 regulates organelle trafficking during cell division. We show that, in mitosis, GBF1 is phosphorylated on Ser292 and Ser297 by casein kinase-2 allowing recognition by the F-box protein ßTrCP. GBF1 interaction with ßTrCP recruits GBF1 to the SCFßTrCP ubiquitin ligase complex, triggering its degradation. Phosphorylation and degradation of GBF1 occur along microtubules at the intercellular bridge of telophase cells and are required for Golgi membrane positioning and postmitotic Golgi reformation. Indeed, expression of a non-degradable GBF1 mutant inhibits the transport of the Golgi cluster adjacent to the midbody toward the Golgi twin positioned next to the centrosome and results in defective Golgi reassembly and cytokinesis failure. These findings define a mechanism that controls postmitotic Golgi reassembly and inheritance.

ß-TrCP- and Casein Kinase II-Mediated Degradation of Cyclin F Controls Timely Mitotic Progression.

Orderly progressions of events in the cell division cycle are necessary to ensure the replication of DNA and cell division. Checkpoint systems allow the accurate execution of each cell-cycle phase. The precise regulation of the levels of cyclin proteins is fundamental to coordinate cell division with checkpoints, avoiding genome instability. Cyclin F has important functions in regulating the cell cycle during the G2 checkpoint; however, the mechanisms underlying the regulation of cyclin F are poorly understood. Here, we observe that cyclin F is regulated by proteolysis through ß-TrCP. ß-TrCP recognizes cyclin F through a non-canonical degron site (TSGXXS) after its phosphorylation by casein kinase II. The degradation of cyclin F mediated by ß-TrCP occurs at the G2/M transition. This event is required to promote mitotic progression and favors the activation of a transcriptional program required for mitosis.

Mutant p53 protects ETS2 from non-canonical COP1/DET1 dependent degradation.

Mutations in the tumor suppressor gene TP53 contribute to the development of approximately half of all human cancers. One mechanism by which mutant p53 (mtp53) acts is through interaction with other transcription factors, which can either enhance or repress the transcription of their target genes. Mtp53 preferentially interacts with the erythroblastosis virus E26 oncogene homologue 2 (ETS2), an ETS transcription factor, and increases its protein stability. To study the mechanism underlying ETS2 degradation, we knocked down ubiquitin ligases known to interact with ETS2. We observed that knockdown of the constitutive photomorphogenesis protein 1 (COP1) and its binding partner De-etiolated 1 (DET1) significantly increased ETS2 stability, and conversely, their ectopic expression led to increased ETS2 ubiquitination and degradation. Surprisingly, we observed that DET1 binds to ETS2 independently of COP1, and we demonstrated that mutation of multiple sites required for ETS2 degradation abrogated the interaction between DET1 and ETS2. Furthermore, we demonstrate that mtp53 prevents the COP1/DET1 complex from ubiquitinating ETS2 and thereby marking it for destruction. Mechanistically, we show that mtp53 destabilizes DET1 and also disrupts the DET1/ETS2 complex thereby preventing ETS2 degradation. Our study reveals a hitherto unknown function in which DET1 mediates the interaction with the substrates of its cognate ubiquitin ligase complex and provides an explanation for the ability of mtp53 to protect ETS2.

Regulation of nucleotide metabolism by mutant p53 contributes to its gain-of-function activities.

Mutant p53 (mtp53) is an oncogene that drives cancer cell proliferation. Here we report that mtp53 associates with the promoters of numerous nucleotide metabolism genes (NMG). Mtp53 knockdown reduces NMG expression and substantially depletes nucleotide pools, which attenuates GTP-dependent protein activity and cell invasion. Addition of exogenous guanosine or GTP restores the invasiveness of mtp53 knockdown cells, suggesting that mtp53 promotes invasion by increasing GTP. In addition, mtp53 creates a dependency on the nucleoside salvage pathway enzyme deoxycytidine kinase for the maintenance of a proper balance in dNTP pools required for proliferation. These data indicate that mtp53-harbouring cells have acquired a synthetic sick or lethal phenotype relationship with the nucleoside salvage pathway. Finally, elevated expression of NMG correlates with mutant p53 status and poor prognosis in breast cancer patients. Thus, mtp53's control of nucleotide biosynthesis has both a driving and sustaining role in cancer development.

Phosphorylation regulates coilin activity and RNA association.

The Cajal body (CB) is a domain of concentrated components found within the nucleus of cells in an array of species that is functionally important for the biogenesis of telomerase and small nuclear ribonucleoproteins. The CB is a dynamic structure whose number and size change during the cell cycle and is associated with other nuclear structures and gene loci. Coilin, also known as the marker protein for the CB, is a phosphoprotein widely accepted for its role in maintaining CB integrity. Recent studies have been done to further elucidate functional activities of coilin apart from its structural role in the CB in an attempt to explore the rationale for coilin expression in cells that have few CBs or lack them altogether. Here we show that the RNA association profile of coilin changes in mitosis with respect to that during interphase. We provide evidence of transcriptional and/or processing dysregulation of several CB-related RNA transcripts as a result of ectopic expression of both wild-type and phosphomutant coilin proteins. We also show apparent changes in transcription and/or processing of these transcripts upon coilin knockdown in both transformed and primary cell lines. Additionally, we provide evidence of specific coilin RNase activity regulation, on both U2 and hTR transcripts, by phosphorylation of a single residue, serine 489. Collectively, these results point to additional functions for coilin that are regulated by phosphorylation.

Coilin levels modulate cell cycle progression and γH2AX levels in etoposide treated U2OS cells.

Coilin is considered the Cajal body (CB) marker protein. In this report, we investigated the role of coilin in the DNA damage response and found that coilin reduction correlated with significantly increased levels of soluble γH2AX in etoposide treated U2OS cells. Additionally, coilin levels influenced the proliferation rate and cell cycle distribution of cells exposed to etoposide. Moreover, coilin overexpression inhibited nucleolar localization of endogenous coilin in etoposide treated U2OS cells. Collectively, these data provide additional evidence for coilin and CBs in the DNA damage response.

Coilin phosphomutants disrupt Cajal body formation, reduce cell proliferation and produce a distinct coilin degradation product.

Coilin is a nuclear phosphoprotein that accumulates in Cajal bodies (CBs). CBs participate in ribonucleoprotein and telomerase biogenesis, and are often found in cells with high transcriptional demands such as neuronal and cancer cells, but can also be observed less frequently in other cell types such as fibroblasts. Many proteins enriched within the CB are phosphorylated, but it is not clear what role this modification has on the activity of these proteins in the CB. Coilin is considered to be the CB marker protein and is essential for proper CB formation and composition in mammalian cells. In order to characterize the role of coilin phosphorylation on CB formation, we evaluated various coilin phosphomutants using transient expression. Additionally, we generated inducible coilin phosphomutant cell lines that, when used in combination with endogenous coilin knockdown, allow for the expression of the phosphomutants at physiological levels. Transient expression of all coilin phosphomutants except the phosphonull mutant (OFF) significantly reduces proliferation. Interestingly, a stable cell line induced to express the coilin S489D phosphomutant displays nucleolar accumulation of the mutant and generates a N-terminal degradation product; neither of which is observed upon transient expression. A N-terminal degradation product and nucleolar localization are also observed in a stable cell line induced to express a coilin phosphonull mutant (OFF). The nucleolar localization of the S489D and OFF coilin mutants observed in the stable cell lines is decreased when endogenous coilin is reduced. Furthermore, all the phosphomutant cells lines show a significant reduction in CB formation when compared to wild-type after endogenous coilin knockdown. Cell proliferation studies on these lines reveal that only wild-type coilin and the OFF mutant are sufficient to rescue the reduction in proliferation associated with endogenous coilin depletion. These results emphasize the role of coilin phosphorylation in the formation and activity of CBs.

Coilin participates in the suppression of RNA polymerase I in response to cisplatin-induced DNA damage.

Coilin is a nuclear phosphoprotein that concentrates within Cajal bodies (CBs) and impacts small nuclear ribonucleoprotein (snRNP) biogenesis. Cisplatin and γ-irradiation, which cause distinct types of DNA damage, both trigger the nucleolar accumulation of coilin, and this temporally coincides with the repression of RNA polymerase I (Pol I) activity. Knockdown of endogenous coilin partially overrides the Pol I transcriptional arrest caused by cisplatin, while both ectopically expressed and exogenous coilin accumulate in the nucleolus and suppress rRNA synthesis. In support of this mechanism, we demonstrate that both cisplatin and γ-irradiation induce the colocalization of coilin with RPA-194 (the largest subunit of Pol I), and we further show that coilin can specifically interact with RPA-194 and the key regulator of Pol I activity, upstream binding factor (UBF). Using chromatin immunoprecipitation analysis, we provide evidence that coilin modulates the association of Pol I with ribosomal DNA. Collectively, our data suggest that coilin acts to repress Pol I activity in response to cisplatin-induced DNA damage. Our findings identify a novel and unexpected function for coilin, independent of its role in snRNP biogenesis, establishing a new link between the DNA damage response and the inhibition of rRNA synthesis.

Coilin interacts with Ku proteins and inhibits in vitro non-homologous DNA end joining.

Coilin is a nuclear protein that plays a role in Cajal body formation. The function of nucleoplasmic coilin is unknown. Here we report that coilin interacts with Ku70 and Ku80, which are major players in the DNA repair process. Ku proteins compete with SMN and SmB' proteins for coilin interaction sites. The binding domain on coilin for Ku proteins cannot be localized to one discrete region, and only full-length coilin is capable of inhibiting in vitro non-homologous DNA end joining (NHEJ). Since Ku proteins do not accumulate in CBs, these findings suggest that nucleoplasmic coilin participates in the regulation of DNA repair.