Prediction of prognostic signatures in triple-negative breast cancer based on the differential expression analysis via NanoString nCounter immune panel

Abstract

Background Triple-Negative Breast Cancer (TNBC) is an aggressive and complex subtype of breast cancer. The current biomarkers used in the context of breast cancer treatment are highly dependent on the targeting of oestrogen receptor, progesterone receptor, or HER2, resulting in treatment failure and disease recurrence and creating clinical challenges. Thus, there is still a crucial need for the improvement of TNBC treatment; the discovery of effective biomarkers that can be easily translated to the clinics is essential. Methods We report an approach for the discovery of biomarkers that can predict tumour relapse and pathologic complete response (pCR) in TNBC on the basis of mRNA expression quantified using the NanoString nCounter Immunology Panel. To overcome the limited sample size, prediction models based on random Forest were constructed using the differentially expressed genes (DEGs) as selected features. We also evaluated the differences between pre- and post-treatment groups aiming for the combinatorial assessment of pCR and relapse using additive models in edgeR. Results We identify nine and 13 DEGs strongly associated with pCR and relapse, respectively, from 579 immune genes in a small number of samples (n = 55) using edgeR. An additive model for the comparison of pre- and post-treatment groups via the adjustment of the independent subject in the relapse group revealed associations for 41 genes. Comprehensive analysis indicated that our prediction models outperformed those constructed using features extracted from the existing feature selection model Elastic Net in terms of accuracy. The prediction models were assessed using a randomization test to validate the robustness (empirical P for the model of pCR = 0.015 and empirical P for the model of relapse = 0.018). Furthermore, three DEGs (FCER1A, EDNRB, and TGFBI) in the model of relapse showed prognostic significance for predicting the survival of patients with cancer through Cox proportional hazards regression model-based survival analysis. Conclusion Gene expression quantified via the NanoString nCounter Immunology Panel can be seamlessly analysed using edgeR, even considering small sample sizes. Our approach provides a scalable framework that can easily be applied for the discovery of biomarkers based on the NanoString nCounter Immunology Panel. Data availability The source code will be available from github at https://github.com/sungheep/nanostring.

Identification of a minimum number of genes to predict triple-negative breast cancer subgroups from gene expression profiles

“…Indeed, Matrilysin is a validated target of several compounds that could be proposed to personalise BL1 TNBC. At the same time, PGBD5 levels were found significantly higher in basal-like BC [ 24 ], and the same goes for CALML5, one of the top expressed genes in TNBC samples [ 25 ], PADI2 [ 26 ] and KLRG2 [ 27 ]. Gong et al demonstrated that the upregulation of MGP promotes the proliferation of cancer which probably makes it a novel biomarker or therapeutic target for TNBC patients [ 28 ].…”

Incorporating knowledge of disease-defining hub genes and regulatory network into a machine learning-based model for predicting treatment response in lupus nephritis after the first renal flare

“…This panel is therefore ideal for studying immune-related conditions and diseases. A key advantage of this NanoString technology is the ability to directly quantify molecules of interest in the absence of an amplification step, which prevents introduction of artificial bias [ 25 ]. Gene expression data were pre-processed and normalized using author-defined methods.…”

MetaNorm: Incorporating Meta-analytic Priors into Normalization of NanoString nCounter Data

“…The parameters s jk are assumed to have a probe specific mean t j and a shared variance τ 2 , as the points by probe in Figure 2(d) have centers clearly distinct from one another but have similar variability. We further enforce the constraints shown in (10), to avoid issues with identifiability, making s jk dependent on other probe effects from the same dataset. We include these constraints to stabilize our distribution in the absence of strong prior information, relying on the fact that the log-transformed positive probe data have a strong linear relationship with the log-transformed control RNA expression levels in each study alone.…”

“…This has led to a ubiquity of FFPE samples and has made the nCounter system an invaluable resource in medical research. Among other areas, the nCounter framework has been used to analyze FFPE samples in studies of colon, breast, and lung cancer [5, 10, 17]. However, a significant downside to using FFPE samples is the level of mRNA “modification” during the preservation and storage of the sample [11], causing higher levels of variability and uncertainty in the readings.…”