Yang, Ping

Abstract Background Electronic health record (EHR)–based prognostic modeling is increasingly used in oncology, yet incorporating pharmacogenomic (PGx) knowledge derived from experimental systems into clinical prediction frameworks remains challenging. This gap is driven by fundamental mismatches between controlled drug–mutation assays and heterogeneous, incomplete real-world clinical data. Methods We propose a representation transfer framework that integrates PGx embeddings learned from large-scale in vitro pharmacogenomic screens into patient-level EHR models. A frozen pharmacogenomic encoder is used to generate interaction-aware embeddings from patient mutation profiles and administered therapies, which are aggregated into a fixed-length PGx Complementarity Representation. These representations are incorporated into multimodal survival prediction models alongside standard clinical features. Performance was evaluated using systematic modality ablation analyses, attribution analyses, and exploratory unsupervised representation analyses. Results Integrating PGx embeddings yielded consistent performance improvements across all evaluated modality combinations. Relative gains were largest in modality-sparse settings, where baseline EHR features encode limited biological context, and were attenuated—but remained significant—in biologically enriched configurations. Attribution analyses indicated that PGx embeddings contributed non-redundant predictive signal beyond standard clinical features. Exploratory unsupervised analyses further demonstrated that the learned representations exhibit interpretable association patterns aligned with known therapeutic exposures and pathway-level associations. Conclusion These findings suggest that externally learned pharmacogenomic representations can be transferred into real-world EHR models as a context-dependent, non-redundant augmentation. By framing PGx knowledge as an interaction-aware representation rather than a mechanistic model, this work provides an informatics framework for integrating experimental pharmacogenomic data into clinical prediction tasks in a reproducible and interpretable manner.

Long intergenic non-coding RNA (lincRNA) is a family of gene transcripts, the functions of which are largely unknown. Although cigarette smoking is the main cause for lung cancer, lung cancer in non-smokers is a separate entity and its underlying cause is little known. Growing evidence suggests lincRNAs play a significant role in cancer development and progression; however, such data is lacking for lung cancer in non-smokers, or those who have never smoked. This study conducted comprehensive profiling of lincRNAs from RNA sequencing (RNA-seq) data of non-smoker patients with lung adenocarcinoma. Both known and novel lincRNAs distinctly segregated tumors from normal tissues. Approximately one third of lincRNAs were differentially expressed between tumors and normal samples and most of them were coordinated with their putative protein gene targets. More importantly, lincRNAs defined two clusters of tumors that were associated with tumor aggressiveness and patient survival. We identified a subset of lincRNAs that were differentially expressed and also associated with patient survival. Very high concordance (R2 = 0.9) was observed for the differentially expressed lincRNAs in the Cancer Genome Atlas (TCGA) validation set of 85 transcriptomes and the lincRNAs associated with survival from the discovery set were similarly predictive in the validation set. These lincRNAs warrant further investigation as potential diagnostic and prognostic markers.

Abstract Purpose: miRNA plays an important role in human disease and cancer. We seek to investigate the expression status, clinical relevance, and functional role of miRNA in non–small cell lung cancer. Experimental Design: We conducted miRNA expression profiling in matched lung adenocarcinoma and uninvolved lung using 56 pairs of fresh-frozen (FF) and 47 pairs of formalin-fixed, paraffin-embedded (FFPE) samples from never smokers. The most differentially expressed miRNA genes were evaluated by Cox analysis and log-rank test. Among the best candidate, miR-708 was further examined for differential expression in two independent cohorts. Functional significance of miR-708 expression in lung cancer was examined by identifying its candidate mRNA target and through manipulating its expression levels in cultured cells. Results: Among the 20 miRNAs most differentially expressed between tested tumor and normal samples, high expression level of miR-708 in the tumors was most strongly associated with an increased risk of death after adjustments for all clinically significant factors including age, sex, and tumor stage (FF cohort: HR, 1.90; 95% CI, 1.08–3.35; P = 0.025 and FFPE cohort: HR, 1.93; 95% CI, 1.02–3.63; P = 0.042). The transcript for TMEM88 gene has a miR-708 binding site in its 3′ UTR and was significantly reduced in tumors high of miR-708. Forced miR-708 expression reduced TMEM88 transcript levels and increased the rate of cell proliferation, invasion, and migration in culture. Conclusions: miRNA-708 acts as an oncogene contributing to tumor growth and disease progression by directly downregulating TMEM88, a negative regulator of the Wnt signaling pathway in lung cancer. Clin Cancer Res; 18(13); 3658–67. ©2012 AACR.

Abstract Background: Genetic susceptibility for cancer can differ substantially among families. We use trait-related covariates to identify a genetically homogeneous subset of families with the best evidence for linkage in the presence of heterogeneity. Methods: We performed a genome-wide linkage screen in 93 families. Samples and data were collected by the familial lung cancer recruitment sites of the Genetic Epidemiology of Lung Cancer Consortium. We estimated linkage scores for each family by the Markov chain Monte Carlo procedure using SimWalk2 software. We used ordered subset analysis (OSA) to identify genetically homogenous families by ordering families based on a disease-associated covariate. We performed permutation tests to determine the relationship between the trait-related covariate and the evidence for linkage. Results: A genome-wide screen for lung cancer loci identified strong evidence for linkage to 6q23–25 and suggestive evidence for linkage to 12q24 using OSA, with peak logarithm of odds (LOD) scores of 4.19 and 2.79, respectively. We found other chromosomes also suggestive for linkages, including 5q31–q33, 14q11, and 16q24. Conclusions: Our OSA results support 6q as a lung cancer susceptibility locus and provide evidence for disease linkage on 12q24. This study further increased our understanding of the inheritability for lung cancer. Validation studies using larger sample size are needed to verify the presence of several other chromosomal regions suggestive of an increased risk for lung cancer and/or other cancers. Impact: OSA can reduce genetic heterogeneity in linkage study and may assist in revealing novel susceptibility loci. Cancer Epidemiol Biomarkers Prev; 19(12); 3157–66. ©2010 AACR.

Abstract A common variant on chromosomal region 15q24-25.1, marked by rs1051730, was found to be associated with lung cancer risk. Here, we attempted to confirm the second variant on 15q24-25.1 in several large sporadic lung cancer populations and determined what percentage of additional risk for lung cancer is due to the genetic effect of the second variant. SNPs rs1051730 and rs481134 were genotyped in 2,818 lung cancer cases and 2,766 controls from four populations. Joint analysis of these two variants (rs1051730 and rs481134) on 15q24-25.1 identified three major haplotypes (G_T, A_C, and G_C) and provided stronger evidence for association of 15q24-25.1 with lung cancer (P = 9.72 × 10−9). These two variants represent three levels of risk associated with lung cancer. The most common haplotype G_T is neutral; the haplotype A_C is associated with increased risk for lung cancer with 5.0% higher frequency in cases than in controls [P = 1.68 × 10−7; odds ratio (OR), 1.24; 95% confidence interval (95% CI), 1.14–1.35]; whereas the haplotype G_C is associated with reduced risk for lung cancer with 4.4% lower frequency in cases than in controls (P = 7.39 × 10−7; OR, 0.80; 95% CI, 0.73–0.87). We further showed that these two genetic variants on 15q24-25.1 independently influence lung cancer risk (rs1051730: P = 4.42 × 10−11; OR, 1.60; 95% CI, 1.46–1.74; rs481134: P = 7.01 × 10−4; OR, 0.81; 95% CI, 0.72–0.92). The second variant on 15q24-25.1, marked by rs481134, explains an additional 13.2% of population attributable risk for lung cancer. Cancer Res; 70(8); 3128–35. ©2010 AACR.

Abstract Cigarette smoking is the major cause for lung cancer, but genetic factors also affect susceptibility. We studied families that included multiple relatives affected by lung cancer. Results from linkage analysis showed strong evidence that a region of chromosome 6q affects lung cancer risk. To characterize the effects that this region of chromosome 6q region has on lung cancer risk, we identified a haplotype that segregated with lung cancer. We then performed Cox regression analysis to estimate the differential effects that smoking behaviors have on lung cancer risk according to whether each individual carried a risk-associated haplotype or could not be classified and was assigned unknown haplotypic status. We divided smoking exposures into never smokers, light smokers (<20 pack-years), moderate smokers (20 to <40 pack-years), and heavy smokers (≥40 pack-years). Comparing results according to smoking behavior stratified by carrier status, compared with never smokers, there was weakly increasing risk for increasing smoking behaviors, with the hazards ratios being 3.44, 4.91, and 5.18, respectively, for light, moderate, or heavy smokers, whereas among the individuals from families without the risk haplotype, the risks associated with smoking increased strongly with exposure, the hazards ratios being, respectively, 4.25, 9.17, and 11.89 for light, moderate, and heavy smokers. The never smoking carriers had a 4.71-fold higher risk than the never smoking individuals without known risk haplotypes. These results identify a region of chromosome 6q that increases risk for lung cancer and that confers particularly higher risks to never and light smokers. Cancer Res; 70(6); 2359–67

Abstract Background: Genetic factors play important roles in lung cancer susceptibility. In this study, we replicated the association of 5p15.33 and 6p21.33 with familial lung cancer. Taking into account the previously identified genetic susceptibility variants on 6q23-25/RGS17 and 15q24-25.1, we further determined the cumulative association of these four genetic regions and the population attributable risk percent of familial lung cancer they account for. Methods: One hundred ninety-four case patients and 219 cancer-free control subjects from the Genetic Epidemiology of Lung Cancer Consortium were used for the association analysis. Each familial case was chosen from one high-risk lung cancer family that has three or more affected members. Single nucleotide polymorphisms (SNP) on chromosomal regions 5p15.33, 6p21.33, 6q23-25/RGS17, and 15q24-25.1 were assessed for their associations with familial lung cancer. The cumulative association of the four chromosomal regions with familial lung cancer was evaluated with the use of a linear logistic model. Population attributable risk percent was calculated for each SNP using risk ratio. Results: SNP rs31489 showed the strongest evidence of familial lung cancer association on 5p15.33 (P = 2 × 10−4; odds ratio, 0.57; 95% confidence interval, 0.42-0.77), whereas rs3117582 showed a weak association on 6p21.33 (P = 0.09; odds ratio, 1.47; 95% confidence interval, 0.94-2.31). Analysis of a combination of SNPs from the four regions provided a stronger cumulative association with familial lung cancer (P = 6.70 × 10−6) than any individual SNPs. The risk of lung cancer was increased to 3- to 11-fold among those subjects who had at least one copy of risk allele at each region compared with subjects without any of the risk factors. These four genetic regions contribute to a total of 34.6% of familial lung cancer in smokers. Conclusions: The SNPs in four chromosomal regions have a cumulative and significant association with familial lung cancer and account for about one-third of the population attributable risk for familial lung cancer. Cancer Epidemiol Biomarkers Prev; 19(2); 517–24

Fifteen percent of lung cancer cases occur in never-smokers and show characteristics that are molecularly and clinically distinct from those in smokers. Epidermal growth factor receptor ( EGFR ) gene mutations, which are correlated with sensitivity to EGFR-tyrosine kinase inhibitors (EGFR-TKIs), are more frequent in never-smoker lung cancers. In this study, microRNA (miRNA) expression profiling of 28 cases of never-smoker lung cancer identified aberrantly expressed miRNAs, which were much fewer than in lung cancers of smokers and included miRNAs previously identified (e.g., up-regulated miR-21) and unidentified (e.g., down-regulated miR-138) in those smoker cases. The changes in expression of some of these miRNAs, including miR-21, were more remarkable in cases with EGFR mutations than in those without these mutations. A significant correlation between phosphorylated-EGFR ( p -EGFR) and miR-21 levels in lung carcinoma cell lines and the suppression of miR-21 by an EGFR-TKI, AG1478, suggest that the EGFR signaling is a pathway positively regulating miR-21 expression. In the never-smoker–derived lung adenocarcinoma cell line H3255 with mutant EGFR and high levels of p -EGFR and miR-21, antisense inhibition of miR-21 enhanced AG1478-induced apoptosis. In a never-smoker–derived adenocarcinoma cell line H441 with wild-type EGFR , the antisense miR-21 not only showed the additive effect with AG1478 but also induced apoptosis by itself. These results suggest that aberrantly increased expression of miR-21, which is enhanced further by the activated EGFR signaling pathway, plays a significant role in lung carcinogenesis in never-smokers, as well as in smokers, and is a potential therapeutic target in both EGFR -mutant and wild-type cases.

Abstract Purpose: We have previously mapped a major susceptibility locus influencing familial lung cancer risk to chromosome 6q23-25. However, the causal gene at this locus remains undetermined. In this study, we further refined this locus to identify a single candidate gene, by fine mapping using microsatellite markers and association studies using high-density single nucleotide polymorphisms (SNP). Experimental Design: Six multigenerational families with five or more affected members were chosen for fine-mapping the 6q linkage region using microsatellite markers. For association mapping, we genotyped 24 6q-linked cases and 72 unrelated noncancer controls from the Genetic Epidemiology of Lung Cancer Consortium resources using the Affymetrix 500K chipset. Significant associations were validated in two independent familial lung cancer populations: 226 familial lung cases and 313 controls from the Genetic Epidemiology of Lung Cancer Consortium, and 154 familial cases and 325 controls from Mayo Clinic. Each familial case was chosen from one high-risk lung cancer family that has three or more affected members. Results: A region-wide scan across 6q23-25 found significant association between lung cancer susceptibility and three single nucleotide polymorphisms in the first intron of the RGS17 gene. This association was further confirmed in two independent familial lung cancer populations. By quantitative real-time PCR analysis of matched tumor and normal human tissues, we found that RGS17 transcript accumulation is highly and consistently increased in sporadic lung cancers. Human lung tumor cell proliferation and tumorigenesis in nude mice are inhibited upon knockdown of RGS17 levels. Conclusion: RGS17 is a major candidate for the familial lung cancer susceptibility locus on chromosome 6q23-25.