U133a microarray

Gene expression data and clinical information of the MD Anderson Cancer Center cohort [18 (link)] (Affymetrix U133A microarrays) was downloaded from the GEO repository (GSE25066). A total of 320 tumors were HR+/HER2− and patients received neoadjuvant chemotherapy and adjuvant endocrine treatment if HR+. HR positivity was defined as any number of stained cells. We applied an additional filter based on gene expression of ESR1 (probeset 205225_at) based on its bimodal distribution (> 10.45). In total, 267 patients met these criteria. Additional file 3: Table S2 lists the patient characteristics. HLA-A, HLA-B, and HLA-C metagenes were calculated as mean expression of the respective probesets (HLA-A: 215313_x_at, 213932_x_at; HLA-B: 211911_x_at, 209140_x_at, 208729_x_at, HLA-C: 208812_x_at, 216526_x_at, 214459_x_at, 211799_x_at). Unsupervised cut-offs were chosen by assigning the same fraction of cases to the groups with high (60%) and low (40%) expression as in the GeparTrio dataset (HLA-A > 14.24, HLA-B > 13.63, and HLA-C > 13.69). Probesets 205225_at (estrogen receptor 1) and 208079_s_at (Aurora kinase A) were used to evaluate their association with HLA-A. Immune cell metagenes were calculated as the mean expression of cell-type-specific genes [16 (link)].

PD-L1 positive tumor cells were treated with different IFN fusions for 72h, and IP-10 expression from MC38, LL/2, GEO and NCI-H747 cells was determined using Meso-Scale technology (V-PLEX ELISA based assay). Expression of IP-10 and CD3D in tumor clinical specimen were downloaded from Oncomine Powertool by pooling multiple published datasets using Affymetrix U133p2 or U133A microarrays, all data sets were log-transformed and median-centered per array, and standard deviations were normalized to one per array.

The GOBO database is a web-based tool that enables a variety of analyzes of gene expression data in a merged 1881-sample breast tumor data set of 11 different publicly available datasets, all generated on Affymetrix U133A microarrays [40 (link)]. Four TFPI specific probe sets (one for total TFPI (α + β), two for TFPIα, one for TFPIβ) and one probe set for TF gene expression were identified (Additional file 4: Table S3).
Using the Gene Set Analysis application, the TFPI and TF gene expressions were dichotomized to levels above the median (high expression) or below the median (low expression) before Kaplan-Meier plots were created and univariate analyzes (log-rank) were performed to predict 10-year censored overall and relapse free survival. This was conducted in the complete 1881-sample merged clinical data set (hereafter termed “all tumors”) as well as in clinical subgroups. Multivariate Cox proportional hazards regression analyzes were carried out to account for possible survival effects of the following covariates; tumor size, age, histological tumor grade, lymph node status and ER-status. In addition, GOBO was used to assess the distribution of TFPI and TF gene expressions across clinical subgroups.

A workflow for this study was presented in Figure 1. The datasets of GSE59867, GSE1869, and GSE42955 were obtained from the GEO database (http://www.ncbi.nlm.nih.gov/geo/). In the GSE59867 dataset (Maciejak et al., 2015 (link)), details on the development of HF during the 6-month follow-up were recorded for 65 samples obtained from 17 patients with AMI. The transcriptional profiling of peripheral blood mononuclear cells (PBMCs) in these 17 patients, which were performed at admission, discharge (4–6 days), 1 month, and 6 months after AMI, were selected for further analysis. There were no significant differences observed in the baseline demographic and clinical characteristics between HF (n = 9) and non-HF (n = 8) patients. This data were sequenced using the GPL6244 platform [Affymetrix Human Gene 1.0 ST Array, transcript (gene) version]. The GSE1869 dataset included 10 patients with HF post-AMI and 6 non-HF patients, and was performed using the platform GPL96 (Affymetrix U133A microarray) (Kittleson et al., 2005 (link)). The GSE42955 dataset included 12 patients with HF post-AMI and 5 non-HF patients, and was performed using the platform GPL6244 (Molina-Navarro et al., 2013 (link)).

A total of 1460 breast cancer profiles from eight breast cancer datasets that were uniformly profiled on the Affymetrix U133A microarray were obtained form the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/). Because of partial overlap in several of these datasets, the selected unique samples from each GEO dataset are provided in Additional file 3: (Affy_sample_info1460.txt). A summary of the combined data set is given in Table 2.

We performed an observational retrospective cohort analysis of women with breast cancer with publicly available clinical and primary tumor mRNA expression and mutation data from the METABRIC study. Data was retrieved using the cBioPortal implementation in R, package CGDS-R [25 (link), 26 (link), 70 (link)]. The METABRIC study included over 2,000 fresh-frozen breast cancer specimens and a subset of normal breast tissue from tumor banks in the UK and Canada. The METABRIC study reported 2,136 primary tumors with expression array (Affymetrix U133A microarray) data and 2,433 primary tumors with somatic mutation testing by sequencing for 173 genes. We also analyzed RNA sequencing data for mRNA expression from 1,100 primary breast cancer tumors from The Cancer Genome Atlas (TCGA) to validate our results for differentially expressed genes from the METABRIC study [27 (link)].