Page 5 - BladderCancerAbstracts

Podiums

Quantitative nuclear prognostics to improve grading for patients Exploring the impact of microbiome in the response of combined
with non-invasive bladder cancer radiation with immune checkpoint blockade in muscle-invasive
Katherine Lindale , Ava Slotman , Minqi Xu , Céline Hardy , bladder cancer
1,2
1
1,2
1,2
Lina Chen , Chelsea L. Jackson , D. Robert Siemens , Robert J. Eva Michaud , Sabina Fehric , Cynthia Faubert , Irah King , José Joao
1
2
1,2
2,3
1
1
5,6
Gooding , Amber L. Simpson 4,5,7 , David M. Berman 1,2 Mansure , Wassim Kassouf 1,4
1
2,3
1 Department of Pathology and Molecular Medicine, Queen’s University, 1 Urologic Oncology Research Division, McGill University Health Centre,
Kingston, ON, Canada; Queen’s Cancer Research Institute, Kingston, ON, Montreal, QC,Canada; McGill Interdisciplinary Initiative in Infection
2
2
Canada; Department of Physics, Engineering Physics, and Astronomy, and Immunity, Montreal, QC, Canada; Meakins-Christie Laboratories,
3
3
Queen’s University, Kingston, ON, Canada; School of Computing, Department of Medicine, McGill University Health Centre, Montreal, QC,
4
Queen’s University, Kingston, ON, Canada; Department of Biomedical Canada; Division of Urology, Department of Surgery, McGill University
5
4
and Molecular Sciences, Queen’s University, Kingston, ON, Canada; Health Centre, Montreal, QC, Canada
6 Department of Urology, Queen’s University, Kingston, ON, Canada; Introduction: Radiation therapy (RT) is a promising bladder-sparing option
7 Centre for Health Innovation, Queen’s University, Kingston, ON, Canada for MIBC treatment, yet 30% of patients do not respond and half will
Introduction: The qualitative nature of the non-muscle-invasive bladder later die of metastasis. Improved antitumor responses when RT is com-
cancer (NMIBC) grading system limits its reproducibility and ability to bined with PD-1/ PD-L1 blockade (CT) have been described in mice, yet
optimize prognostic value, compromising the potential for data-driven determinants of CT success remain flagrantly misunderstood. As such, gut
care decisions. Using machine learning (ML)-based image analysis, we microbiome composition influences PD-1 blockade efficacy and its modi-
constructed models that establish reproducible quantitative thresholds and fication potentiates combined RT and PD-L1 blockade activity. To add,
feature importance for grade classification. We leveraged these tools to responding patients with a favorable gut microbiome (i.e., enrichment in
optimize grade for its ability to stratify risk of recurrence. A. muciniphila, Bifidobacterium, and Faecalibacterium) have enhanced
Methods: Small (1.0 mm diameter) histopathology images were obtained systemic and antitumor immunity. We thus aimed to document the role
for 371 patients with stage Ta NMIBC, with clinical timelines for 163 of patients’ microbiome in polarizing antitumor immune responses to CT
patients. Nuclear measurements of 19 grade-based histological features and use its composition as a predicting factor of CT success in MIBC.
were extracted using Visiopharm image analysis software (Hoersholm, Methods: Fecal material from a responder (R) and non-responder (NR)
Denmark). Quantitative grading models built using these features were MIBC patient was gavaged into 20 germ-free mice. Three weeks after the
analyzed to determine the variables that best classify histological grade. last gavage, MB49 cells were delivered subcutaneously. Once tumors
Cox proportional hazards (CPH) models for time to first bladder recur- reached 0.1–0.15 cm , mice were randomized into four groups: control;
3
rence were cross-validated and tested for grade and quantified nuclear anti-PD-L1; RT; RT+anti-PD-L1. Seven days later, tumors were dissociated
features (QNFs). for single-cell immune sequencing and stools collected for 16S sequen-
Results: The standard deviation of nuclear area and mitotic index were the cing. Correlation networks were built (TransNet, Microbiome R packages)
QNFs that best predicted grade as univariate models (balanced accuracy and visualized in cytoscape.
up to 82%) and when used in combination with shape-related variables Results: We show feasibility and robust engraftment of human FMT to
(mean ellipticalness, mean solidity, and standard deviation of form fac- germ-free mice in a MIBC tumor model. FMT from NR lessened the
tor) in statistical and ML models (balanced accuracy up to 88%). Mitotic known beneficial effects of RT in the MB49 model compared to FMT
index, mean lesser diameter (size, shape), and mean variance HEM (tex- from R. Transkingdom analysis of sc-RNA-seq and 16Sseq shows robust
ture) carried the most prognostic value. Upon validation, the CPH model statistical interactions between immunosuppression and enrichment in
constructed using these QNFs achieved a C-index of 0.73 (CI 0.56–0.88) microbes associated to poor outcome humans.
compared to 0.55 (CI 0.40–0.69) using diagnostic grade alone. Conclusions: To our knowledge, this is the first study to use FMT as a
Conclusions: Histological features in NMIBC can be quantified and used modulator of response in the context of RT combinations in MIBC. These
in grading classification algorithms, addressing the irreproducibility asso- findings could be used to select patients who will benefit most from a
ciated with grading. The prognostic value of QNFs supports grade’s util- personalized therapeutic approach.
ity; however, findings indicate the opportunity to improve grading by
reprioritizing these features. Establishing prognostically driven feature
measurement thresholds and importance weights can optimize grade’s
contribution to risk scoring and clinical decisions.

CUAJ • MARCH 2023 • VOLUME 17, ISSUE 3(SUPPL1) S3