Development of an automated combined positive score prediction pipeline using artificial intelligence on multiplexed immunofluorescence images.

Immunotherapy targeting immune checkpoint proteins, such as programmed cell death ligand 1 (PD-L1), has shown impressive outcomes in many clinical trials but only 20%-40% of patients benefit from it. Utilizing Combined Positive Score (CPS) to evaluate PD-L1 expression in tumour biopsies to identify patients with the highest likelihood of responsiveness to anti-PD-1/PD-L1 therapy has been approved by the Food and Drug Administration for several solid tumour types. Current CPS workflow requires a pathologist to manually score the two-colour PD-L1 chromogenic immunohistochemistry image. Multiplex immunofluorescence (mIF) imaging reveals the expression of an increased number of immune markers in tumour biopsies and has been used extensively in immunotherapy research. Recent rapid progress of Artificial Intelligence (AI)-based imaging analysis, particularly Deep Learning, provides cost effective and high-quality solutions to healthcare. In this article, we propose an imaging pipeline that utilizes three-colour mIF images (DAPI, PD-L1, and Pan-cytokeratin) as input and predicts the CPS using AI techniques. Our novel pipeline is composed of three modules employing algorithms of image processing, machine learning, and deep learning techniques. The first module of quality check (QC) detects and removes the image regions contaminated with sectioning and staining artefacts. The QC module ensures that only image regions free of the three common artefacts are used for downstream analysis. The second module of nuclear segmentation uses deep learning to segment and count nuclei in the DAPI images wherein our specialized method can accurately separate touching nuclei. The third module of cell phenotyping calculates CPS by identifying and counting PD-L1 positive cells and tumour cells. These modules are data-efficient and require only few manual annotations for training purposes. Using tumour biopsies from a clinical trial, we found that the CPS from the AI-based models shows a high Spearman correlation (78%, p = 0.003) to the pathologist-scored CPS.

Leveraging gene expression subgroups to classify DLBCL patients and select for clinical benefit from a novel agent.

Diffuse large B-cell lymphoma (DLBCL) is a heterogeneous disease, commonly described by cell-of-origin (COO) molecular subtypes. We sought to identify novel patient subgroups through an unsupervised analysis of a large public dataset of gene expression profiles from newly diagnosed de novo DLBCL patients, yielding 2 biologically distinct subgroups characterized by differences in the tumor microenvironment. Pathway analysis and immune deconvolution algorithms identified higher B-cell content and a strong proliferative signal in subgroup A and enriched T-cell, macrophage, and immune/inflammatory signals in subgroup B, reflecting similar biology to published DLBCL stratification research. A gene expression classifier, featuring 26 gene expression scores, was derived from the public dataset to discriminate subgroup A (classifier-negative, immune-low) and subgroup B (classifier-positive, immune-high) patients. Subsequent application to an independent series of diagnostic biopsies replicated the subgroups, with immune cell composition confirmed via immunohistochemistry. Avadomide, a CRL4CRBN E3 ubiquitin ligase modulator, demonstrated clinical activity in relapsed/refractory DLBCL patients, independent of COO subtypes. Given the immunomodulatory activity of avadomide and the need for a patient-selection strategy, we applied the gene expression classifier to pretreatment biopsies from relapsed/refractory DLBCL patients receiving avadomide (NCT01421524). Classifier-positive patients exhibited an enrichment in response rate and progression-free survival of 44% and 6.2 months vs 19% and 1.6 months for classifier-negative patients (hazard ratio, 0.49; 95% confidence interval, 0.280-0.86; P = .0096). The classifier was not prognostic for rituximab, cyclophosphamide, doxorubicin, vincristine, prednisone or salvage immunochemotherapy. The classifier described here discriminates DLBCL tumors based on tumor and nontumor composition and has potential utility to enrich for clinical response to immunomodulatory agents, including avadomide.

Multiplex immunofluorescence staining and image analysis assay for diffuse large B cell lymphoma.

With the explosion of immuno-oncology and the approval of many immune checkpoint therapies by regulatory agencies in the last few years, understanding the tumor microenvironment (TME) in the context of patients' immune status has become essential. Among available immune profiling techniques, multiplex immunofluorescence (mIF) assays offer the unique advantage of preserving the architectural features of the tumor and revealing the spatial relationships between tumor cells and immune cells. A number of mIF and image analysis assays have been described for solid tumors but most are not sufficiently suitable in lymphoma, where the lack of clear tumor-stromal boundaries and high tumor density present significant challenges. Here we describe the development and optimization of a reliable workflow using Akoya Opal staining kits to label and analyze 6 markers per slide in diffuse large B-cell lymphoma (DLBCL) tissue sections. Five panels totaling 30 markers were developed to characterize infiltrating immune cells and relevant check-point proteins such as PD1, PD-L1, ICOS, SIRP-alpha and Lag3 on 70 DLBCL sections. Multiplexed sections were scanned using an Akoya multispectral scanner. An image analysis workflow using InForm and Matlab was developed to overcome challenges inherent to the DLBCL environment. Using the assays and workflows detailed here, we were able to quantify cell densities of subsets of infiltrating immune cells and observe their spatial patterns within the tumors. We highlight heterogeneous distribution of cytotoxic T cells across tumors with similar T cell density to underscores the importance of considering spatial context when studying the effects of immunological therapies in DLBCL.

SALL4 mediates teratogenicity as a thalidomide-dependent cereblon substrate.

Targeted protein degradation via small-molecule modulation of cereblon offers vast potential for the development of new therapeutics. Cereblon-binding therapeutics carry the safety risks of thalidomide, which caused an epidemic of severe birth defects characterized by forelimb shortening or phocomelia. Here we show that thalidomide is not teratogenic in transgenic mice expressing human cereblon, indicating that binding to cereblon is not sufficient to cause birth defects. Instead, we identify SALL4 as a thalidomide-dependent cereblon neosubstrate. Human mutations in SALL4 cause Duane-radial ray, IVIC, and acro-renal-ocular syndromes with overlapping clinical presentations to thalidomide embryopathy, including phocomelia. SALL4 is degraded in rabbits but not in resistant organisms such as mice because of SALL4 sequence variations. This work expands the scope of cereblon neosubstrate activity within the formerly 'undruggable' C2H2 zinc finger family and offers a path toward safer therapeutics through an improved understanding of the molecular basis of thalidomide-induced teratogenicity.

Development of a mathematical model to estimate intra-tumor oxygen concentrations through multi-parametric imaging.

BACKGROUND: Tumor hypoxia is involved in every stage of solid tumor development: formation, progression, metastasis, and apoptosis. Two types of hypoxia exist in tumors-chronic hypoxia and acute hypoxia. Recent studies indicate that the regional hypoxia kinetics is closely linked to metastasis and therapeutic responses, but regional hypoxia kinetics is hard to measure. We propose a novel approach to determine the local pO2 by fusing the parameters obtained from in vivo functional imaging through the use of a modified multivariate Krogh model. METHODS: To test our idea and its potential to translate into an in vivo setting through the use of existing imaging techniques, simulation studies were performed comparing the local partial oxygen pressure (pO2) from the proposed multivariate image fusion model to the referenced pO2 derived by Green's function, which considers the contribution from every vessel segment of an entire three-dimensional tumor vasculature to profile tumor oxygen with high spatial resolution. RESULTS: pO2 derived from our fusion approach were close to the referenced pO2 with regression slope near 1.0 and an r2 higher than 0.8 if the voxel size (or the spatial resolution set by functional imaging modality) was less than 200 µm. The simulation also showed that the metabolic rate, blood perfusion, and hemoglobin concentration were dominant factors in tissue oxygenation. The impact of the measurement error of functional imaging to the pO2 precision and accuracy was simulated. A Gaussian error function with FWHM equal to 20 % of blood perfusion or fractional vascular volume measurement contributed to average 7 % statistical error in pO2. CONCLUSION: The simulation results indicate that the fusion of multiple parametric maps through the biophysically derived mathematical models can monitor the intra-tumor spatial variations of hypoxia in tumors with existing imaging methods, and the potential to further investigate different forms of hypoxia, such as chronic and acute hypoxia, in response to cancer therapies.

Development of Follicle-Stimulating Hormone Receptor Binding Probes to Image Ovarian Xenografts.

The Follicle-Stimulating Hormone Receptor (FSHR) is used as an imaging biomarker for the detection of ovarian cancer (OC). FSHR is highly expressed on ovarian tumors and involved with cancer development and metastatic signaling pathways. A decapeptide specific to the FSHR extracellular domain is synthesized and conjugated to fluorescent dyes to image OC cells in vitro and tumors xenograft model in vivo. The in vitro binding curve and the average number of FSHR per cell are obtained for OVCAR-3 cells by a high resolution flow cytometer. For the decapeptide, the measured EC50 was 160 µM and the average number of receptors per cell was 1.7 × 107. The decapeptide molecular imaging probe reached a maximum tumor to muscle ratio five hours after intravenous injection and a dose-dependent plateau after 24-48 hours. These results indicate the potential application of a small molecular weight imaging probe specific to ovarian cancer through binding to FSHR. Based on these results, multimeric constructs are being developed to optimize binding to ovarian cells and tumors.

Circulating Markers Reflect Both Anti- and Pro-Atherogenic Drug Effects in ApoE-Deficient Mice.

BACKGROUND: Current drug therapy of atherosclerosis is focused on treatment of major risk factors, e.g. hypercholesterolemia while in the future direct disease modification might provide additional benefits. However, development of medicines targeting vascular wall disease is complicated by the lack of reliable biomarkers. In this study, we took a novel approach to identify circulating biomarkers indicative of drug efficacy by reducing the complexity of the in vivo system to the level where neither disease progression nor drug treatment was associated with the changes in plasma cholesterol. RESULTS: ApoE-/- mice were treated with an ACE inhibitor ramipril and HMG-CoA reductase inhibitor simvastatin. Ramipril significantly reduced the size of atherosclerotic plaques in brachiocephalic arteries, however simvastatin paradoxically stimulated atherogenesis. Both effects occurred without changes in plasma cholesterol. Blood and vascular samples were obtained from the same animals. In the whole blood RNA samples, expression of MMP9, CD14 and IL-1RN reflected pro-and anti-atherogenic drug effects. In the plasma, several proteins, e.g. IL-1beta, IL-18 and MMP9 followed similar trends while protein readout was less sensitive than RNA analysis. CONCLUSION: In this study, we have identified inflammation-related whole blood RNA and plasma protein markers reflecting anti-atherogenic effects of ramipril and pro-atherogenic effects of simwastatin in a mouse model of atherosclerosis. This opens an opportunity for early, non-invasive detection of direct drug effects on atherosclerotic plaques in complex in vivo systems.