BackgroundUntargeted plasma metabolomic profiling combined with machine learning (ML) may lead to discovery of metabolic profiles that inform our understanding of pediatric CKD causes. We sought to identify metabolomic signatures in pediatric CKD based on diagnosis: FSGS, obstructive uropathy (OU), aplasia/dysplasia/hypoplasia (A/D/H), and reflux nephropathy (RN).MethodsUntargeted metabolomic quantification (GC-MS/LC-MS, Metabolon) was performed on plasma from 702 Chronic Kidney Disease in Children study participants (n: FSGS=63, OU=122, A/D/H=109, and RN=86). Lasso regression was used for feature selection, adjusting for clinical covariates. Four methods were then applied to stratify significance: logistic regression, support vector machine, random forest, and extreme gradient boosting. ML training was performed on 80% total cohort subsets and validated on 20% holdout subsets. Important features were selected based on being significant in at least two of the four modeling approaches. We additionally performed pathway enrichment analysis to identify metabolic subpathways associated with CKD cause.ResultsML models were evaluated on holdout subsets with receiver-operator and precision-recall area-under-the-curve, F1 score, and Matthews correlation coefficient. ML models outperformed no-skill prediction. Metabolomic profiles were identified based on cause. FSGS was associated with the sphingomyelin-ceramide axis. FSGS was also associated with individual plasmalogen metabolites and the subpathway. OU was associated with gut microbiome–derived histidine metabolites.ConclusionML models identified metabolomic signatures based on CKD cause. Using ML techniques in conjunction with traditional biostatistics, we demonstrated that sphingomyelin-ceramide and plasmalogen dysmetabolism are associated with FSGS and that gut microbiome–derived histidine metabolites are associated with OU.

The development of immunotherapy drugs, such as immune checkpoint inhibitors (ICIs) has changed the environment of cancer treatment tremendously by providing efficacious therapeutic options for many cancer patients. However, only a minority of patients experience durable clinical benefit and increasing evidence has linked the efficacy of ICIs to tumor cell heterogeneity, the complex tumor immune microenvironment and their interactions, which remains poorly understood, particularly in the tissue context. Recent advances in spatially resolved transcriptomics (SRT) has provided great opportunity to better understand spatial tumor-immune interactions. In this study, we obtained public and in-house SRT data on a total of 136 tissue sections across 11 different cancer types, representing to date, the largest collection of SRT on human cancer. Sample and spot-level quality filters were applied, batch effects were assessed and properly handled. High-quality SRT data were pre-processed uniformly to comprehensively interrogate the spatial heterogeneity and architectures of the 3 major compartments (malignant, stroma, and immune) as well as relationships among them. Transcriptome data was integrated with region- and/or spot-level annotations from pathologists. For regions enriched with malignant cells, we inferred somatic copy number alterations, clonal structure of tumor cells, and profiled the transcriptional hallmarks of intra-tumor heterogeneity (ITH) including a number of curated gene sets and meta-programs, and systematically characterized tumor cell heterogeneity under the spatial modality. We observed a great variation in aneuploidy levels and transcriptome profiles within and across patients and cancer types and notably in molecular processes regulating tumor cell responses to stress, hypoxia and interferon signals and other key processes such as epithelial-mesenchymal transition. In addition, we performed cell-type deconvolution analysis using available tools including RCTD and cell2location, based on expression of curated cell-type specific gene signatures, we inferred levels of immune infiltration in each tissue section in both tumor core and invasive edges, and classified tumors into “cold”, “warm”, “hot”, and “mixed” immune phenotypes. We further quantified the abundance and spatial distribution of stromal cells and key TME structures such as tertiary lymphoid structures and lympho-myeloid aggregates, as well as their spatial neighbors with oncogenic features which revealed multiple interesting interplay patterns. Together, this study provide novel insights into our improved understanding of spatial tumor heterogeneity and tumor-immune interactions and revealed potential exploitable targets, and great resource for the community. Citation Format: Guangsheng Pei, Jingjing Wu, Enyu Dai, Yunhe Liu, Guangchun Han, Jian Hu, Fuduan Peng, Kyung S. Cho, Jiahui Jiang, Daiwei Zhang, Ansam F. Sinjab, Boyu Zhang, Shumei Song, Junya Fujimoto, Luisa M. Solis Soto, Anirban Maitra, Jaffer Ajani, Mingyao Li, Humam Kadara, Linghua Wang. Pan-cancer characterization of tumor-immune interactions using spatially resolved transcriptomics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6766.

The tumor microenvironment (TME) contains networks of cells and structures that surround tumor cells. Cell populations in the TME, including their abundance, composition, and spatial location are critical determinants of the occurrence, growth, and metastasis of a tumor. A comprehensive analysis of the multiple exchanges between tumor cells and their TME is essential for understanding the underlying mechanisms of tumor growth and response to anti-cancer therapies. Recent advances in spatially resolved transcriptomics (SRT) techniques have enabled gene expression profiling while preserving location information in tissues, which innovates a promising avenue to study the TME in a spatial context. With the power of SRT, we aim to provide a detailed annotation of tumor structure and different lymphocytes by integrating gene expression information and cell morphology features in the complemented high-resolution histology image obtained from the same tissue section. A major challenge that hinders gene expression and histology integration in SRT data is the relatively low resolution of gene expression data compared to pixel-resolution histology images. As gene expression is only measured in discrete spots that are separated by tissue gaps, a large proportion of the tissue area remains unmeasured (e.g., >50% in 10x Visium). The incomplete coverage of gene expression in ST has prevented the deciphering of detailed TME structures such as the tertiary lymphoid structure (TLS).To overcome this challenge, we present TESLA (Tumor Edge Structure and Lymphocyte multi-level Annotation), a machine learning framework that integrates gene expression and histology image in SRT to investigate the detailed strictures in TME. TESLA first fills in the gene expression for unmeasured areas and generates gene expression images in the same resolution as the histology image. Next, TESLA integrates the gene images with the histology image to annotate different tumor/TME cell types at pixel resolution. Some specific tumor-infiltrating lymphocytes structures, such as TLSs, can also be detected by colocalization analysis of different lymphocytes. TESLA is also able to characterize high-resolution cellular and molecular spatial structures of tumor by separating tumor into different subtype regions and elucidating differential transcriptome programs. The detailed multi-level annotations performed by TESLA provide a comprehensive understanding of the spatial context and the nature of cellular heterogeneity of the TME. Citation Format: Jian Hu, Kyle Coleman, Edward B. Lee, Humam Kadara, Linghua Wang, Mingyao Li. Deciphering tumor ecosystems at super-resolution from spatial transcriptomics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 747.

Transcriptomic analysis has substantially advanced our understanding of human diseases, but the complex nature of tissues is often ignored. Recent development of single-cell RNA-seq (scRNA-seq) technologies has made it possible to characterize cellular heterogeneity of solid tissues. Along with analyzing gene expression patterns on a single-cell level, there is a critical need to explore the spatial patterns exhibited by the genes across the tissue sample. Spatially resolved transcriptomics (SRT) is key to understanding cellular functions in their morphological state. However, current barcoding-based SRT technologies, such as 10x Genomics Visium, lack single-cell resolution, greatly hindering the investigation of the spatial differences in gene expression profiles as failure to account for cell-type variations can lead to an obscured understanding of the spatial patterns detected across the tissue. The overall goal of this research is to integrate spatial transcriptomics and scRNA-seq data in order to infer the gene expression profiles for each of the bulk-level spots in the spatially resolved data. This will allow us to study spatial patterns of genes with cell-type level resolution. SPACER expands the spatial transcriptomics data into the individual cell types by utilizing the spatial similarities and high-resolution histology information as a weight in a non-negative least squares regression. Having this integrated understanding allows us to gain an additional level of information about the gene activity for each cell type present in the tissue. We have performed benchmark evaluations for our method based on data generated from the 10x Visium platform and have seen promising results. The evaluations have shown high correlations for the gene expression patterns predicted by SPACER for varying cell types on the benchmark evaluations. We next analyzed spatial transcriptomics data for pancreatic cancer, breast cancer, and melanoma tissue samples to better study the complex nature of cancerous tissues. A tumor is consistently interacting with its microenvironment, which can impact tumor growth and cell proliferation. Thus, it is important to study the spatial gene expression patterns for different cell types in cancerous tissues. During our analysis, we identify the boundary of the tumor region in the bulk-spatial transcriptomics data and perform differential expression analysis on the SPACER results to compare the expression patterns of the core tumor region to the infiltrated tumor regions for each individual cell type. Identifying genes that are differentially expressed for specific cell types in varying regions of the cancerous tissue can link gene expressions to clinically important morphological features, provide critical information about how cells are interacting with neighboring cells in their macroenvironment, and be used to target different tumor cells therapeutically. Citation Format: Amelia R. Schroeder, Kyle Coleman, Jian Hu, Mingyao Li. Modeling spatially resolved cell-type-specific gene expression by weighted regression with SPACER [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1205.