Logo: to the web site of Uppsala University

The histopathological evaluation of biopsies by human experts is a gold standard in clinical disease diagnosis. While recent artificial intelligence-based (AI) approaches have reached human expert-level performance, they often display shortcomings caused by variations in sample preparation, limiting clinical applicability. This study investigates the impact of data variation on AI-based histopathological grading and explores algorithmic approaches that confer prediction robustness. To evaluate the impact of data variation in histopathology, we collected a multicentric, retrospective, observational prostate cancer (PCa) trial consisting of six cohorts in 3 countries with 25,591 patients, 83,864 images. This includes a high-variance dataset of 8,157 patients and 28,236 images with variations in section thickness, staining protocol, and scanner. This unique training dataset enabled the development of an AI-based PCa grading framework by training on patient outcome, not subjective grading. It was made robust through several algorithmic adaptations, including domain adversarial training and credibility-guided color adaptation. We named the final grading framework PCAI. We compare PCAI to a BASE model and human experts on three external test cohorts, comprising 2,255 patients and 9,437 images. Variations in sample processing, particularly section thickness and staining time, significantly reduced the performance of AI-based PCa grading by up to 8.6 percentage points in the event-ordered concordance index (EOC-Index) thus highlighting serious risks for AI-based histopathological grading. Algorithmic improvements for model robustness, credibility, and training on high-variance data as well as outcome-based severity prediction give rise to robust models with grading performance surpassing experienced pathologists. We demonstrate how our algorithmic enhancements for greater robustness lead to significantly better performance, surpassing expert grading on EOC-Index and 5-year AUROC by up to 21.2 percentage points.

Background: Prostate cancer (PCa) is among the most common cancers in men and itsdiagnosis requires the histopathological evaluation of biopsies by human experts. Whileseveral recent artificial intelligence-based (AI) approaches have reached human expert-levelPCa grading, they often display significantly reduced performance on external datasets. Thisreduced performance can be caused by variations in sample preparation, for instance thestaining protocol, section thickness, or scanner used. Another limiting factor of contemporaryAI-based PCa grading is the prediction of ISUP grades, which leads to the perpetuation ofhuman annotation errors.

Methods: We developed the prostate cancer aggressiveness index (PCAI), an AI-based PCadetection and grading framework that is trained on objective patient outcome, rather thansubjective ISUP grades. We designed PCAI as a clinical application, containing algorithmicmodules that offer robustness to data variation, medical interpretability, and a measure ofprediction confidence. To train and evaluate PCAI, we generated a multicentric, retrospective,observational trial consisting of six cohorts with 25,591 patients, 83,864 images, and 5 yearsof median follow-up from 5 different centers and 3 countries. This includes a high-variancedataset of 8,157 patients and 28,236 images with variations in sample thickness, stainingprotocol, and scanner, allowing for the systematic evaluation and optimization of modelrobustness to data variation. The performance of PCAI was assessed on three external testcohorts from two countries, comprising 2,255 patients and 9,437 images.

Findings: Using our high-variance datasets, we show how differences in sample processing,particularly slide thickness and staining time, significantly reduce the performance ofAI-based PCa grading by up to 6.2 percentage points in the concordance index (C-index). Weshow how a select set of algorithmic improvements, including domain adversarial training,conferred robustness to data variation, interpretability, and a measure of credibility to PCAI.These changes lead to significant prediction improvement across two biopsy cohorts and oneTMA cohort, systematically exceeding expert ISUP grading in C-index and AUROC by up to22 percentage points.

Interpretation: Data variation poses serious risks for AI-based histopathological PCagrading, even when models are trained on large datasets. Algorithmic improvements formodel robustness, interpretability, credibility, and training on high-variance data as well asoutcome-based severity prediction gives rise to robust models with above ISUP-level PCagrading performance.

Background: Amyloid-beta (A beta) is one of the hallmark lesions of Alzheimer's disease (AD). During the disease process, A beta undergoes biochemical changes, producing toxic beta variants, proposed to be detected within the neurons. Idiopathic normal pressure hydrocephalus (iNPH) causes cognitive impairment, gait, and urinary symptoms in elderly, that can be reversed by a ventriculo-peritoneal shunt. Majority of iNPH subjects display different A beta variants in their brain biopsies, obtained during shunting. Objective: To study the cellular compartmentalization of different A beta variants in brain biopsies from iNPH subjects. Methods: We studied the cellular localization of different proteoforms of A beta using antibodies towards different amino acid sequences or post-translational modifications of A beta, including clones 4G8, 6F/3D, unmodified- (7H3D6), pyroglutamylated-(N3pE), phosphorylated-(1E4E11) A beta and A beta protein precursor (A beta PP), in brain biopsies from 3 iNPH subjects, using immunohistochemistry and light microscopy (LM), light microscopy on semi-thin sections (LMst), and electron microscopy (EM). Results: In LM all A beta variants were detected. In LMst and EM, the A beta 4G8, 6F/3D, and the pyroglutamylated A beta were detected. The A beta PP was visualized by all methods. The A beta labelling was located extracellularly with no specific signal within the intracellular compartment, whereas the A beta PP was seen both intra- and extracellularly. Conclusions: TheA beta markers displayed extracellular localization when visualized by three assessment techniques, reflecting the pathological extracellular accumulation of A beta in the human brain. No intracellular A beta pathology was seen. A beta PP was visualized in intra- and extracellularly, which corresponds to the localization of the protein in the membranes of cells and organelles.

Background Improved molecular diagnosis is needed in prostate cancer (PC). Fine needle aspiration (FNA) is a minimally invasive biopsy technique, less traumatic compared to core needle biopsy, and could be useful for diagnosis of PC. Molecular biomarkers (BMs) in FNA-samples can be assessed for prediction, eg of immunotherapy efficacy before treatment as well as at treatment decision time points during disease progression.

Methods In the present pilot study, the expression levels of 151 BM proteins were analysed by proximity extension assay in FNA-samples from 16 patients, including benign prostate lesions (n = 3) and cancers (n = 13). An ensemble data analysis strategy was applied using several machine learning models.

Results Twelve potentially predictive BM proteins correlating with International Society of Urological Pathology grade groups were identified, among them vimentin, tissue factor pathway inhibitor 2, and integrin beta-5. The validity of the results was supported by network analysis that showed functional associations between most of the identified putative BMs. We also showed that multiple immune checkpoint targets can be assessed (eg PD-L1, CD137, and Galectin-9), which may support the selection of immunotherapy in advanced PC. Results are promising but need further validation in a larger cohort.

Conclusions Our pilot study represents a "proof of concept" and shows that multiplex profiling of potential diagnostic and predictive BM proteins is feasible on tumour material obtained by FNA sampling of prostate cancer. Moreover, our results demonstrate that an ensemble data analysis strategy may facilitate the identification of BM signatures in pilot studies when the patient cohort is limited.

BACKGROUND: Prostate cancer (PCa) is a highly heterogeneous, multifocal disease, and identification of clinically significant lesions is challenging, which complicates the choice of adequate treatment. The Prostatype® score (P-score) is intended to guide treatment decisions for newly diagnosed PCa patients based on a three-gene signature (IGFBP3, F3, and VGLL3) and clinicopathological information obtained at diagnosis. This study evaluated association of the P-score measured in preoperative magnetic resonance imaging/transrectal ultrasound fusion-guided core needle biopsies (CNBs) and the P-score measured in radical prostatectomy (RP) specimens of PCa patients. We also evaluated the P-score association with the pathology of RP specimens. Furthermore, concordance of the P-score in paired CNB and RP specimens, as well as in index versus concomitant nonindex tumor foci from the same RP was investigated.

METHODS: The study included 100 patients with localized PCa. All patients were diagnosed by CNB and underwent RP between 2015 and 2018. Gene expression was assessed with the Prostatype® real-time quantitative polymerase chain reaction kit and the P-score was calculated. Patients were categorized into three P-score risk groups according to previously defined cutoffs.

RESULTS: For 71 patients, sufficient CNB tumor material was available for comparison with the RP specimens. The CNB-based P-score was associated with the pathological T-stage in RP specimens (p = 0.02). Moreover, the CNB-based P-score groups were in substantial agreement with the RP-based P-score groups (weighted κ score: 0.76 [95% confidence interval, 95% CI: 0.60-0.92]; Spearman's rank correlation coefficient r = 0.83 [95% CI: 0.74-0.89]; p < 0.0001). Similarly, the P-score groups based on paired index tumor and concomitant nonindex tumor foci (n = 64) were also in substantial agreement (weighted κ score: 0.74 [95% CI: 0.57-0.91]; r = 0.83 [95% CI: 0.73-0.89], p < 0.0001).

CONCLUSIONS: Our findings suggest that the P-score based on preoperative CNB accurately reflects the pathology of the whole tumor, highlighting its value as a decision support tool for newly diagnosed PCa patients.

Background: MRI and fusion guided biopsy have an increased role in the diagnosis of prostate cancer. Purpose: To demonstrate the possible advantages with Bi-parametric MRI fusion-guided repeat biopsy over systematic 10-12-core biopsy for the diagnosis of prostate cancer. Material and Methods: Four hundred and twenty-three consecutive men, with previous systematic 10-12-core TRUS-guided biopsy, and with suspicion of, or diagnosis of, low-risk prostate cancer underwent fusion-guided prostate biopsy between February 2015 and February 2017. The material was retrospectively assessed. In 220 cases no previous cancer was diagnosed, and in 203 cases confirmatory fusion guided biopsy was performed prior to active monitoring. MRI was classified according to PI-RADS. Systematic biopsy was compared to fusion guided biopsy for the detection of cancer, and PI-RADS was compared to the Gleason score. Results: Fusion guided biopsy detected significantly more cancers than systematic (p < .001). Gleason scores were higher in the fusion biopsy group (p < .001). Anterior tumors were present in 54% of patients. Fusion biopsy from these lesions showed cancer in 53% with previously negative biopsy in systematic biopsies and 66% of them were upgraded from low risk to intermediate or high-risk cancers. Conclusion: These results show superior detection rate and grading of bi-parametric MRI/TRUS fusion targeted repeat biopsy over systematic 10-12 core biopsies. Fusion guided biopsy detects more significant cancers despite using fewer cores. The risk group was changed for many patients initially selected for active surveillance due to upgrading of tumors. Bi-parametric MRI shows promising results in detecting anterior tumors in patients with suspected prostate cancer.

Cancer is associated with systemic pathologies that contribute to mortality, such as thrombosis and distant organ failure. The aim of this study was to investigate the potential role of neutrophil extracellular traps (NETs) in myocardial inflammation and tissue damage in treatment-naïve individuals with cancer. Mice with mammary carcinoma (MMTV-PyMT) had increased plasma levels of NETs measured as H3Cit-DNA complexes, paralleled with elevated coagulation, compared to healthy littermates. MMTV-PyMT mice displayed upregulation of pro-inflammatory markers in the heart, myocardial hypertrophy and elevated cardiac disease biomarkers in the blood, but not echocardiographic heart failure. Moreover, increased endothelial proliferation was observed in hearts from tumor-bearing mice. Removal of NETs by DNase I treatment suppressed the myocardial inflammation, expression of cardiac disease biomarkers and endothelial proliferation. Compared to a healthy control group, treatment-naïve cancer patients with different malignant disorders had increased NET formation, which correlated to plasma levels of the inflammatory marker CRP and the cardiac disease biomarkers NT-proBNP and sTNFR1, in agreement with the mouse data. Altogether, our data indicate that NETs contribute to inflammation and myocardial stress during malignancy. These findings suggest NETs as potential therapeutic targets to prevent cardiac inflammation and dysfunction in cancer patients.

Samples in biobanks are generally preserved by formalin-fixation and paraffin-embedding (FFPE) and/or optimal cutting temperature compound (OCT)-embedding and subsequently frozen. Mass spectrometry (MS)-based analysis of these samples is now available via developed protocols, however, the differences in results with respect to preservation methods needs further investigation. Here we use bladder urothelial carcinoma tissue of two different tumor stages (Ta/T1-non-muscle invasive bladder cancer (NMIBC), and T2/T3-muscle invasive bladder cancer (MIBC)) which, upon sampling, were divided and preserved by FFPE and OCT. Samples were parallel processed from the two methods and proteins were analyzed with label-free quantitative MS. Over 700 and 1200 proteins were quantified in FFPE and OCT samples, respectively. Multivariate analysis indicates that the preservation method is the main source of variation, but also tumors of different stages could be differentiated. Proteins involved in mitochondrial function were overrepresented in OCT data but missing in the FFPE data, indicating that these proteins are not well preserved by FFPE. Concordant results for proteins such as HMGCS2 (uniquely quantified in Ta/T1 tumors), and LGALS1, ANXA5 and plastin (upregulated in T2/T3 tumors) were observed in both FFPE and OCT data, which supports the use of MS technology for biobank samples and encourages the further evaluation of these proteins as biomarkers.

Bladder cancer, one of the most prevalent malignancies worldwide, remains hard to classify due to a staggering molecular complexity. Despite a plethora of diagnostic tools and therapies, it is hard to outline the key steps leading up to the transition from high-risk non-muscle-invasive bladder cancer (NMIBC) to muscle-invasive bladder cancer (MIBC). Carcinogen-induced murine models can recapitulate urothelial carcinogenesis and natural anti-tumor immunity. Herein, we have developed and profiled a novel model of progressive NMIBC based on 10 weeks of OH-BBN exposure in hepatocyte growth factor/cyclin dependent kinase 4 (R24C) (Hgf-Cdk4(R24C)) mice. The profiling of the model was performed by histology grading, single cell transcriptomic and proteomic analysis, while the derivation of a tumorigenic cell line was validated and used to assess in vivo anti-tumor effects in response to immunotherapy. Established NMIBC was present in females at 10 weeks post OH-BBN exposure while neoplasia was not as advanced in male mice, however all mice progressed to MIBC. Single cell RNA sequencing analysis revealed an intratumoral heterogeneity also described in the human disease trajectory. Moreover, although immune activation biomarkers were elevated in urine during carcinogen exposure, anti-programmed cell death protein 1 (anti-PD1) monotherapy did not prevent tumor progression. Furthermore, anti-PD1 immunotherapy did not control the growth of subcutaneous tumors formed by the newly derived urothelial cancer cell line. However, treatment with CpG-oligodeoxynucleotides (ODN) significantly decreased tumor volume, but only in females. In conclusion, the molecular map of this novel preclinical model of bladder cancer provides an opportunity to further investigate pharmacological therapies ahead with regards to both targeted drugs and immunotherapies to improve the strategies of how we should tackle the heterogeneous tumor microenvironment in urothelial bladder cancer to improve responses rates in the clinic.

Background: Survival of metastatic colorectal cancer (mCRC) patients has improved, but mainly for trial patients. New predictive and prognostic biomarkers validated in the general mCRC population are needed. Caudal-type homeobox 2 (CDX2) is an intestine-specific transcription factor with potential prognostic and predictive effect, but the importance in mCRC has not been fully investigated. Methods: Immunohistochemistry analysis of CDX2 was performed in a Scandinavian population-based cohort of mCRC (n = 796). Frequency, clinical and tumor characteristics, response rate, progression-free survival, and overall survival (OS) were estimated. Results: Loss of CDX2 expression was found in 87 (19%) of 452 stained cases, in 53% if BRAF mutated (BRAFmut) and in 9% if KRAS mutated (KRASmut). CDX2 loss was associated with microsatellite instability, BRAFmut, and poor differentiation and inversely associated with KRASmut. Patients with CDX2 loss received less first-line (53 vs. 64%, p = 0.050) and second-line (23 vs. 39%, p = 0.006) chemotherapy and secondary surgery (1 vs. 9%, p = 0.019). Median progression-free survival and OS for patients given first-line combination chemotherapy was 4 and 10 months if CDX2 loss vs. 9 and 24 months if CDX2 expressed (p = 0.001, p < 0.001). Immediate progression on first-line combination chemotherapy was seen in 35% of patients with CDX2 loss vs. 10% if CDX2 expressed (p = 0.003). Median OS in patients with BRAFmut or KRASmut and CDX2 expressed in tumor (both 21 months) was comparable to wild-type patients (27 months). However, if CDX2 loss, median OS was only 8 and 11 months in BRAFmut and KRASmut cases, respectively, and 10 months in double wild-type patients. In multivariate analysis, CDX2 loss (hazard ratio: 1.50, p = 0.027) and BRAFmut (hazard ratio: 1.62, p = 0.012) were independent poor prognostic markers for OS. Conclusion: In a population-based cohort of mCRC patients, CDX2 loss is an independent poor prognostic marker. Expression of CDX2 defines a new subgroup of BRAFmut cases with a much better prognosis. Loss of CDX2 defines a small group of KRASmut cases with a worse prognosis. Patients with CDX2 loss receive less palliative chemotherapy with less benefit and rarely reach secondary surgery.