Using machine learning algorithms to predict response and toxicity to immune checkpoint inhibitors (ICIs) in melanoma patients

2581

Background: There is growing interest in optimizing patient selection for treatment with immune checkpoint inhibitors (ICIs). We postulate that phenotypic features present in metastatic melanoma tissue reflect the biology of tumor cells, immune cells, and stromal tissue, and hence can provide predictive information about tumor behavior. Here, we test the hypothesis that machine learning algorithms can be trained to predict the likelihood of response and/or toxicity to ICIs. Methods: We examined 124 stage III/IV melanoma patients who received anti-CTLA-4 (n = 81), anti-PD-1 (n = 25), or combination (n = 18) therapy as first line. The tissue analyzed was resected before treatment with ICIs. In total, 340 H&E slides were digitized and annotated for three regions of interest: tumor, lymphocytes, and stroma. The slides were then partitioned into training (n = 285), validation (n = 26), and test (n = 29) sets. Slides were tiled (299x299 pixels) at 20X magnification. We trained a deep convolutional neural network (DCNN) to automatically segment the images into each of the three regions and then deconstruct images into their component features to detect non-obvious patterns with objectivity and reproducibility. We then trained the DCNN for two classifications: 1) complete/partial response versus progression of disease (POD), and 2) severe versus no immune-related adverse events (irAEs). Predictive accuracy was estimated by area under the curve (AUC) of receiver operating characteristics (ROC). Results: The DCNN identified tumor within LN with AUC 0.987 and within ST with AUC 0.943. Prediction of POD based on ST-only always performed better than prediction based on LN-only (AUC 0.84 compared to 0.61, respectively). The DCNN had an average AUC 0.69 when analyzing only tumor regions from both LN and ST data sets and AUC 0.68 when analyzing tumor and lymphocyte regions. Severe irAEs were predicted with limited accuracy (AUC 0.53). Conclusions: Our results support the potential application of machine learning on pre-treatment histologic slides to predict response to ICIs. It also revealed their limited value in predicting toxicity. We are currently investigating whether the predictive capability of the algorithm can be further improved by incorporating additional immunologic biomarkers.

Baseline antibody profiles predict toxicity in melanoma patients treated with immune checkpoint inhibitors

Background

Immune checkpoint inhibitors (anti-CTLA-4, anti-PD-1, or the combination) enhance anti-tumor immune responses, yielding durable clinical benefit in several cancer types, including melanoma. However, a subset of patients experience immune-related adverse events (irAEs), which can be severe and result in treatment termination. To date, no biomarker exists that can predict development of irAEs.

Methods

We hypothesized that pre-treatment antibody profiles identify a subset of patients who possess a sub-clinical autoimmune phenotype that predisposes them to develop severe irAEs following immune system disinhibition. Using a HuProt human proteome array, we profiled baseline antibody levels in sera from melanoma patients treated with anti-CTLA-4, anti-PD-1, or the combination, and used support vector machine models to identify pre-treatment antibody signatures that predict irAE development.

Results

We identified distinct pre-treatment serum antibody profiles associated with severe irAEs for each therapy group. Support vector machine classifier models identified antibody signatures that could effectively discriminate between toxicity groups with > 90% accuracy, sensitivity, and specificity. Pathway analyses revealed significant enrichment of antibody targets associated with immunity/autoimmunity, including TNFα signaling, toll-like receptor signaling and microRNA biogenesis.

Conclusions

Our results provide the first evidence supporting a predisposition to develop severe irAEs upon immune system disinhibition, which requires further independent validation in a clinical trial setting.

Electronic supplementary material

The online version of this article (10.1186/s12967-018-1452-4) contains supplementary material, which is available to authorized users.

The garlic component, allicin, prevents disease caused by Aeromonas hydrophila in rainbow trout, Oncorhynchus mykiss (Walbaum)

Allicin was fed at 0 (= control), 0.5 and 1.0 mL of Allimed liquid 100 g(-1) of feed for 14 days to rainbow trout, Oncorhynchus mykiss (Walbaum), fingerlings before infection with Aeromonas hydrophila with a resultant reduction in mortalities from 80% in the controls to 8% [relative percentage survival (RPS) = 90%] and 0% (RPS = 100%) among the treated fish. Allicin was strongly antibacterial compared to the control, with a minimum inhibitory concentration (MIC) of >400 microL mL(-1) of Allimed liquid. Use of allicin led to a lower number of white blood cells (132.0 +/- 0.4 x 10(3)) compared to 175.0 +/- 0.1 x 10(3) in the controls, but elicited increased phagocytic activity, i.e. a phagocytic value of 39.2% compared to 13.6% in the controls, and serum lysozyme activity, which showed significant (P > 0.05) differences compared to the control at 15 and 30 min after the first reading at 0 min of incubation.

First Report of Crown Gall, Caused by Agrobacterium tumefaciens on Geraldton Wax (Chamelaucium uncinatum) in South Africa

Geraldton wax (Chamelaucium uncinatum, family Myrtaceae) plants are grown as cutflowers for the export market in South Africa. In July 1998, gall-like structures were observed on collars and roots of Geraldton wax plants in commercial fields in Wellington. The galls were observed after plants exhibited poor growth. The galls varied in size and in texture from soft and spongy to hard. Secondary symptoms involved poor root development and browning of stem tissues near galls. Isolations from the galls yielded nearly pure cultures of a Gram negative, rod-shaped bacterium on Roy Sauer medium (2), typical of an Agrobacterium sp. Carbon source utilization testing with the Biolog GN Bacterial Identification System (version 3.50) confirmed the bacterium as a biovar of A. tumefaciens with a similarity of 0.88. Pathogenicity was confirmed by injecting 4- to 6-week old tomato and tobacco plants and 1-year-old Geraldton wax plants with approximately 5 μl of the bacterial suspension (10⁸ CFU/ml) in sterile, distilled water. Inoculated plants were then transferred to a greenhouse at 25°C. Galls developed 1 month after inoculation. The bacterium was readily reisolated from the inoculated plants. A. tumefaciens is endemic to South Africa and has a very wide host range that includes several ornamentals (1). This is the first report of A. tumefaciens on Geraldton wax plants in South Africa. References: (1) J. F. Bradbury. 1986. Guide to Plant Pathogenic Bacteria. CAB Int., Slough, U.K. (2) N. W. Schaad. 1988. Laboratory Guide for Identification of Plant Pathogenic Bacteria. The American Phytopathological Society, St. Paul, MN.