C. Omics & Fusion Biology [C. Omics & Fusion Biology] C-1 PPARγ Targets-Derived Diagnostic and Prognostic Index for Papillary Thyroid Cancer Jaehyung Kim¹ , Soo Young Kim² , Shi-Xun Ma³, Su-Jin Shin⁵, Yong Sang Lee⁴, Hojin Chang⁴, Hang-Seok # # Chang⁴, Cheong Soo Park⁶, Seok-Mo Kim⁴*, Su Bin Lim¹* ¹Department of Biochemistry & Molecular Biology, Ajou University School of Medicine, Suwon 16499, Korea, ²Department of Surgery, Ajou University School of Medicine, Suwon 16499, Korea, ³Department of Neurology, Institute for Cell Engineering, Johns Hopkins University School of Medicine, Baltimore MD
21205, USA, ⁴Department of Surgery, Yonsei University College of Medicine, Seoul 06273, Korea, ⁵Department of Pathology, Yonsei University College of Medicine, Seoul 06273, Korea In most cases, papillary thyroid cancer (PTC) is highly curable and associated with an excellent prognosis. Yet, there are several clinicopathological features that lead to a poor prognosis, underscoring the need for a better genomic strategy to refine prognostication and patient management. We hypothesized that PPARγ targets could be potential markers for better diagnosis and prognosis due to the variants found in
PPARG in three pairs of monozygotic twins with PTC. Here, we developed a 10-gene personalized prognostic index, designated PPARGi, based on gene expression of 10 PPARγ targets. Through scRNA-seq data analysis of PTC tissues derived from patients, we found that PPARGi genes were predominantly expressed in macrophages and epithelial cells. Machine learning algorithms showed a near-perfect performance of PPARGi in deciding the presence of the disease and in selecting a small subset of patients with poor disease-specific survival in TCGA-THCA and newly developed merged microarray data (MMD)
consisting exclusively of thyroid cancers and normal tissues. 2022 KSBMB Winter Workshop PPARγ Targets-Derived Diagnostic and Prognostic Index for Papillary Thyroid Cancer 4, 5 3 Jaehyung Kim 1,†,# , Soo Young Kim , Shi-Xun Ma , Seok-Mo Kim * , Su-Jin Shin , Yong Sang Lee , Hojin Chang , Hang-Seok Chang , Cheong Soo Park and Su Bin Lim * 2,† 4 1, 6 4 4 1 Department of Biochemistry & Molecular Biology, Ajou University School of Medicine, Suwon 16499, Korea, 2 Department of Surgery, Ajou University School of Medicine, Suwon 16499, Department of Neurology, 3 Institute for Cell Engineering, Johns
Hopkins University School of Medicine, Baltimore, MD 21205, USA, Thyroid Cancer Center, Department of Surgery, Institute of Refractory Thyroid Cancer, 4 Gangnam Severance Hospital, Yonsei University College of Medicine, Seoul 06273, Korea, Department of Pathology, Gangnam Severance Hospital, Yonsei University College of Medicine, 5 6 Seoul 06273, Korea, CHA Ilsan Medical Center, Department of Surgery, Goyang-si 10414, Korea, * Correspondence: seokmokim@yuhs.ac (S.-M.K.); sblim@ajou.ac.kr (S.B.L.); † These authors contributed equally to this work. # E-mail: stride@ajou.ac.kr , Homepage:
https://sites.google.com/view/subinlab/ ABSTRACT In most cases, papillary thyroid cancer (PTC) is highly curable and associated with an excellent prognosis. Yet, there are several clinicopathological features that lead to a poor prognosis, underscoring the need for a better genomic strategy to refine prognostication and patient management. We hypothesized that PPARγ targets could be potential markers for better diagnosis and prognosis due to the variants found in PPARG in three pairs of monozygotic twins with PTC. Here, we developed a 10- gene personalized prognostic index, designated PPARGi,
based on gene expression of 10 PPARγ targets. Through scRNA-seq data analysis of PTC tissues derived from patients, we found that PPARGi genes were predominantly expressed in macrophages and epithelial cells. Machine learning algorithms showed a near-perfect performance of PPARGi in deciding the presence of the disease and in selecting a small subset of patients with poor disease-specific survival in TCGA-THCA and newly developed merged microarray data (MMD) consisting exclusively of thyroid cancers and normal tissues. RESULTS Figure 3. Analysis of PPARGi-comprising genes at the single-cell
level in PTC. (A) UMAP 0o.2f0404.25 Q0.C30-p0a.3s5se0d.40cells depicting GeneRatio 0.0100 15 cell clusters. (B) Heatmap of top 5 differentially expressed features across the identified cell clusters. The number (n) and proportion (%) of cells in each cluster are stated. (C) Distribution of PPARGi across different cell clusters. Kruskal– Wallis p-value (p) is stated. (D) Dot plot depicting average expression of PPARGi-comprising genes across different cell clusters. (E) Heatmap showing expression levels of PPARGi-comprising genes and PPARGi in macrophages (left) and epithelial cells (right).
The data are sorted in increasing PPARGi. (F) Dot plot showing top enriched GO terms from PPARGi-comprising genes. Figure 4. ML algorithms applied to expression profiles of PPARGi-comprising genes. A) The t-distributed stochastic neighbor embedding (t-SNE) visualization of cancers (TT) and normal (TT) tissues in TCGA-THCA (left), MMD-THCA (PTC, middle), and MMD-THCA (ATC, right). (B) Confusion matrices of LASSO and SVM models in TCGA-THCA (top right), MMD-PTC (bottom left), and MMD-ATC (bottom right). (C) ROC curves and (D) confusion matrix showing classifying performance for normal (left) and
tumor (right) tissues in Figure 1. Diagnostic and prognostic performance of PPARGi in TCGA-THCA. MMD-PTC using kNN model developed from TDM-transformed TCGA-PTC. (E) t-SNE visualization of PPARGihigh (colored in blue) and PPARGiow (colored in red) tumors in TCGA-THCA. (F) Confusion matrix of SVM model in classifying PPARGihigh from PPARGilow tumors. (G) ROC curve (A) Venn diagrams showing the number of the SnpEff-predicted variants in three pairs of twins (TP1–3). Genomic positions of mutations, including showing classifying performance for PPARGiow tumors. chromosome (chr) number, and gene
symbols are shown in the table (right). (B) Expression heatmap (left) and forest plot (right) of PPARGi- comprising genes. The data are sorted in increasing PPARGi. The horizontal axis of the forest plot represents hazard ratio (HR) with 95% confidence Table 1.ML models for disease selection in thyroid cancer intervals (CI) estimated using a Cox proportional hazards model. The regression coefficients (Coef) and the Wald statistic p-values (p) are stated. (C) PPARGi computation for patient stratification. (D) PPARGi of normal (n = 59) and tumor tissues (n = 505). Mann–Whitney–Wilcoxon test
p-values (p) and the num- ber of samples (n) are stated. (E) The area under the ROC curve (AUC) of the PPARGi classifier. The AUC value and the optimal cut-off are stated. (F) The hazard ratio (HR), log-rank p-value, and the number of patients successfully stratified (n) determined from univariate Cox regression analysis are shown on the survival Kaplan–Meier (KM) curve. Black and red KM curves represent predicted PPARGilow and PPARGihigh group, respectively. (G) Volcano plot depicting differentially expressed (red and blue) and non-significant (gray) genes in the PPARGi-stratified groups. The
number of genes (n) are stated. (H) GSEA plot showing the enrichment of thyroid hormone generation gene set (GO:0006590) in PPARGilow tumors. The cumulative enrichment score (ES) is plotted as the green curve, which is the running sum of the weighted ES as the analysis walks down the limma-generated ranked list. The vertical black lines on the horizontal axis of the plot indicate the position of query genes in the ranked list of genes. The bottom plot shows the value of the fold change (log2-base) as the computation goes down the limma-generated ranked list. Normalized ES (NES) and adjusted
p-values (padj) are stated. (I) Dot plot showing top gene sets (downregulated) in PPARGilow tumors. (J) Dot plot showing Pearson’s correlations between PPARGi and CIBERSORT-estimated proportion of immune cell populations. Table 2. Cross-platform evaluation results of the ML models. Table 3. ML models for risk stratification in thyroid cancer. CONCLUSION It remains to be investigated whether the identified intronic variants found in monozy- gotic twins in the LBD of the PPARG gene would induce loss-of-function effects on PPARγ or impaired PPARγ-RXRα signaling pathway. Further, our sample size
was small and might only represent a small subset of PTC cohort in Korea, although the expression levels and prognostic performance of PPARGi-comprising Figure 2. PPARGi in MMD-THCA. genes were extensively validated using public transcriptomic databases. We thus aim to assess the (A) UMAP representation of the uncorrected (left) and corrected (right) MMD. Data are colored by study (top) and tissue type (bottom). The number of predictive power of PPARGi in a larger validation patient cohort or in a prospectively conducted study. samples (n) are stated. (B) Disease-gene associations depicting
the linkages of genes and the enriched diseases as a network. The color and size of the node represent the value of fold change (FC) and gene count, respectively. (C) Expression heatmap of PPARGi-comprising genes. The data are Altogether, the functional PPARGi personalized scoring system may represent a powerful and effective sorted in increasing PPARGi. (D) PPARGi of normal and tumor tissues (ATC, PDTC, and PTC). The asterisks represent the statistical significance genomic tool to improve patient management in PTC. assessed by Mann–Whitney–Wilcoxon test (**** p ≤ 0.0001, ** p ≤ 0.01, ns: p >
0.05). Kruskal–Wallis p-value (p) is stated. (E) The area under the ROC curve (AUC) of the PPARGi classifier. The AUC values are stated for ATC (left), PDTC (middle), and PTC (right). (F) Volcano plot depicting ACKNOWLEDGMENTS differentially expressed (red and blue) and non-significant (gray) genes in the PPARGi-stratified groups in PTC. The number of genes (n) are stated. (G) GSEA plot showing the enrichment of thyroid hormone generation gene set (GO:0006590) in PPARGilow tumors. The cumulative enrichment score is plotted as the green curve, which is the running sum of the weighted ES as the
analysis walks down the limma-generated ranked list. The This research was funded by the National Research Foundation of Korea (NRF) grant funded by the vertical black lines on the horizontal axis of the plot indicate the position of query genes in the ranked list of genes. The bottom plot shows the value of Korea government (MSIT) (2021R1F1A1064122 and 2017R1E1A1A03070345) and a faculty research the fold change (log2-base) as the computation goes down the limma-generated ranked list. Normalized ES (NES) and adjusted p-values (padj) are stated. (H) Dot plot showing Pearson’s correlations
between PPARGi and CIBERSORT-estimated proportion of immune cell populations. fund of Ajou University School of Medicine and Yonsei University College of Medicine (6-2019-0092). [C. Omics & Fusion Biology] C-2 Autophagy context-based drug repurposing approach with Parkinson's disease-specific network # # Seunghwan Jung¹,² , Junseok Park¹,² , Jaywook Han¹,², Doheon Lee¹,²* ¹Department of Bio and Brain Engineering, KAIST, Daejeon 34141, Korea, ²Bio-Synergy Research Center, Bio- Synergy Research Center, Daejeon 34141, Korea Parkinson's disease is a major long-term degenerative disease that
affects the motor system. Parkinson's disease has been increasing over the past 30 years, despite ongoing research and efforts at drug development. The reason is that previous drug development for Parkinson's disease has mainly focused on recovering dopamine levels, which only provide short-term relief. Recent studies prove that misfolded protein aggregation causes mitochondria dysfunction and neuroinflammation so that elimination of the misfolded protein should offer an effective long-term treatment for Parkinson's disease. A promising mechanism to eliminate the misfolded protein is
autophagy. We developed a pipeline to identify a
