Leveraging pharmacovigilance data to predict population-scale toxicity profiles of checkpoint inhibitor immunotherapy
Related Articles
High baseline levels of PD-L1 reduce the heterogeneity of immune checkpoint signature and sensitize anti-PD1 therapy in lung and colorectal cancers
Immune checkpoint blockade (ICB) therapy only induces durable responses in a subset of cancer patients. The underlying mechanisms of such selective efficacy remain largely unknown. By analyzing the expression profiles of immune checkpoint molecules in different statuses of murine tumors, we found that tumor progression generally randomly upregulated multiple immune checkpoints, thus increased the Heterogeneity of Immune checkpoint Signature (HIS) and resulted in immunotherapeutic resistance. Interestingly, overexpressing one pivotal immune checkpoint in a tumor hindered the upregulation of a majority of other immune checkpoint genes during tumor progression via suppressing interferon γ, resulting in HIS-low. Indeed, PD-L1 high-expression sensitized baseline large tumors to anti-PD1 therapy without altering the sensitivity of baseline small tumors. In line with these preclinical results, a retrospective analysis of a phase III study involving patients with non-small cell lung cancer (NSCLC) revealed that PD-L1 tumor proportion score (TPS) ≥ 50% more reliably predicted therapeutic response in NSCLC patients with baseline tumor volume (BTV)-large compared to patients with BTV-small. Notably, TPS combined with BTV significantly improved the predictive accuracy. Collectively, the data suggest that HIS reflects the dynamic features of tumor immune evasion and dictates the selective efficacy of ICB in a tumor size-dependent manner, providing a potential novel strategy to improve precision ICB. These findings highlight the application of ICB to earlier stages of cancer patients. The integration of PD-L1 with BTV may immediately improve patient stratification and prediction performance in the clinic.
Inhibition of GSK3β is synthetic lethal with FHIT loss in lung cancer by blocking homologous recombination repair
FHIT is a fragile site tumor suppressor that is primarily inactivated upon tobacco smoking. FHIT loss is frequently observed in lung cancer, making it an important biomarker for the development of targeted therapy for lung cancer. Here, we report that inhibitors of glycogen synthase kinase 3 beta (GSK3β) and the homologous recombination DNA repair (HRR) pathway are synthetic lethal with FHIT loss in lung cancer. Pharmacological inhibition or siRNA depletion of GSK3β selectively suppressed the growth of FHIT-deficient lung cancer tumors in vitro and in animal models. We further showed that FHIT inactivation leads to the activation of DNA damage repair pathways, including the HRR and NHEJ pathways, in lung cancer cells. Conversely, FHIT-deficient cells are highly dependent on HRR for survival under DNA damage stress. The inhibition of GSK3β in FHIT-deficient cells suppressed the ATR/BRCA1/RAD51 axis in HRR signaling via two distinct pathways and suppressed DNA double-strand break repair, leading to the accumulation of DNA damage and apoptosis. Small molecule inhibitors of HRR, but not NHEJ or PARP, induced synthetic lethality in FHIT-deficient lung cancer cells. The findings of this study suggest that the GSK3β and HRR pathways are potential drug targets in lung cancer patients with FHIT loss.
Intratumoral and peritumoral PET/CT-based radiomics for non-invasively and dynamically predicting immunotherapy response in NSCLC
We aimed to develop a machine learning model based on intratumoral and peritumoral 18F-FDG PET/CT radiomics to non-invasively and dynamically predict the response to immunotherapy in non-small cell lung cancer (NSCLC).
Assessment of polygenic risk score performance in East Asian populations for ten common diseases
Polygenic risk score (PRS) uses genetic variants to assess disease susceptibility. While PRS performance is well-studied in Europeans, its accuracy in East Asians is less explored. This study evaluated PRSs for ten diseases in the Health Examinees (HEXA) cohort (n = 55,870) in Korea. Single-population PRSs were constructed using PRS-CS, LDpred2, and Lassosum based on East Asian GWAS summary statistics (sample sizes: 51,442–341,204), while cross-population PRSs were developed using PRS-CSx and CT-SLEB by integrating European and East Asian GWAS data. PRS-CS consistently outperformed other single-population methods across key metrics, including the likelihood ratio test (LRT), odds ratio per standard deviation (perSD OR), net reclassification improvement (NRI), and area under the curve (AUC). Cross-population PRSs further improved predictive performance, with average increases of 1.08-fold (LRT), 1.07-fold (perSD OR), and 1.15-fold (NRI) across seven diseases with statistical significance, and a 1.01-fold improvement in AUC. Differences in R² between single- and cross-population PRSs were statistically significant for five diseases, showing an average increase of 1.13%. Cross-population PRSs achieved 87.8% of the predictive performance observed in European PRSs. These findings highlight the benefits of integrating European GWAS data while underscoring the need for larger East Asian datasets to improve prediction accuracy.
Multi-population GWAS detects robust marker associations in a newly established six-rowed winter barley breeding program
Genome-wide association study (GWAS) is a powerful tool for identifying marker-trait associations that can accelerate breeding progress. Yet, its power is typically constrained in newly established breeding programs where large phenotypic and genotypic datasets have not yet accumulated. Expanding the dataset by inclusion of data from well-established breeding programs with many years of phenotyping and genotyping can potentially address this problem. In this study we performed single- and multi-population GWAS on heading date and lodging in four barley breeding populations with varying combinations of row-type and growth habit. Focusing on a recently established 6-rowed winter (6RW) barley population, single-population GWAS hardly resulted in any significant associations. Nevertheless, the combination of the 6RW target population with other populations in multi-population GWAS detected four and five robust candidate quantitative trait loci for heading date and lodging, respectively. Of these, three remained undetected when analysing the combined populations individually. Further, multi-population GWAS detected markers capturing a larger proportion of genetic variance in 6RW. For multi-population GWAS, we compared the findings of a univariate model (MP1) with a multivariate model (MP2). While both models surpassed single-population GWAS in power, MP2 offered a significant advantage by having more realistic assumptions while pointing towards robust marker-trait associations across populations. Additionally, comparisons of GWAS findings for MP2 and single-population GWAS allowed identification of population-specific loci. In conclusion, our study presents a promising approach to kick-start genomics-based breeding in newly established breeding populations.
Responses