Introduction
Cancer is a genetic disease characterized by genomic abnormalities that alter the transcriptome and influence the pathways that control proliferation and survival (1). The application of next-generation sequencing technology and single-cell sequencing in oncology has provided evidence that cellular heterogeneity is common in cancer (2). In the majority of tumor studies, key information may be disregarded owing to the tissue-centric nature of research. Malignant solid tumor tissues consist mainly of tumor cells, but also contain tumor-associated stromal, immune and vascular cells. Although non-tumor cells constitute a relatively small proportion of the cancer tissue, their role as potent tumor promoters has been previously indicated (3). A previous study, using a network approach to identify the functional gene modules in cancer cells, verified the presence of immune, stromal and vascular gene modules in cancer tissues (4).
The majority of genomic and transcriptomic studies into cancer do not explicitly consider genetic heterogeneity, and the generated inferences usually refer to mixed cell populations (5). However, the experimental isolation of single cells from tissues is expensive and may affect cell physiology. Additionally, the single-cell sequencing of a large cohort is unrealistic. An efficient solution to this limitation may be the de-convolution of genomic data from heterogeneous samples. Publicly available transcriptome databases can provide resources that allow for this type of analysis. To date, only one method, referred to as ‘Estimation of STromal and Immune cells in MAlignant Tumors using Expression data’ (ESTIMATE) has been described that can be used to score the stromal and immune fraction in transcriptomic data of cancer tissue (5,6). However, to the best of our knowledge, the association of the proportion of immune and stromal fraction with patient survival has not been thoroughly investigated.
Several studies have examined the microenvironment-associated transcriptional tumor profile using transcriptomic data. Calon et al (7) identified transforming growth factor-β (TGF-β) response signatures in tumor-associated stromal cells that could predict disease relapse in colorectal cancer (CRC). Cheng et al (8) used principle component analysis and clustering methods to identify a signature of stromal activation that was associated with late recurrence in breast cancer. Teschendorff et al (9) described an immune response gene expression module associated with a good prognostic subtype in estrogen receptor negative breast cancer. Finak et al (10) used laser capture microdissection (LCM) to compare the gene expression profiles of tumor stroma from primary breast tumors and derived signatures that were strongly associated with the clinical outcome by clustering. Isella et al (11) used Gene Set Enrichment Analysis (GSEA) and examined the gene signatures of subtypes for expression in stromal cell subpopulations vs. CRC cells. Wu et al (12) identified a stromal gene super-module associated with gastric cancer patient survival using gene co-expression network analysis. Furthermore, extensive experimental research has indicated the role served by stromal and immune cells in breast cancer (8,10,13), CRC (7,11,14), lymphoma (15) and drug resistance (16,17).
Transcriptome-based subtyping of cancer identifies different subtypes by clustering; however, non-tumor components are usually ignored (18). The ESTIMATE algorithm scores stromal and immune cells that form the major non-tumor components of tumor samples. In the present study, the scoring of stromal and immune cells in healthy and cancerous tissues, as well as in disease prognosis and drug resistance was investigated. The scores were associated with the clinicopathological characteristics of various cancer types and chemotherapeutic drug resistance. The results of the present study indicated that ESTIMATE could be used as a metric for patient prognosis assessment.
Discussion
To the best of our knowledge, this is the first study in which ESTIMATE scoring was used to differentiate between tissues. Stromal and immune scores were associated with the clinical outcome of the patient and chemotherapy drug resistance. The prognostic value of the two scores was validated using multiple microarray platforms. Either the stromal or the immune score was associated with patient survival, relapse and metastasis in multiple types of cancer. Furthermore, the two scores were associated with chemotherapeutic drug response. The present study has indicated a microenvironment view of tissue-based tumor transcriptomic data and highlighted the contribution of stromal and immune cells in carcinogenesis. However, the precise molecular mechanism underlying this phenomenon should be investigated in future studies.
Tissues are composed of a mixture of cell types. Currently, tissue-based transcriptomic data do not reflect the information from multiple cell types. Even though a number of technological approaches, including flow cytometry and LCM, have been developed, their application in cancer research is limited owing to the expensive cost. Therefore, it is crucial to study the information hidden in available public datasets. Functional modules in cancer transcriptomic datasets have been previously identified by gene co-expression network analysis (4). In the present study, two major components of the cancer microenvironment, stromal and immune cells, were investigated, and it was demonstrated that the ESTIMATE scores for these components could predict patient clinical outcomes. For instance, the stroma score was identified as a predictor for survival in patients with gastric cancer (a lower stroma score was indicative of better patient prognosis). Notably, the ESTIMATE-derived stromal score was found to be more efficient than the recently described stromal super-module-based method (12). However, it should be mentioned that the stroma might serve distinct roles in different tissues. In the present study it was demonstrated that a higher stroma score indicated later relapse in breast cancer, which is consistent with a clustering-derived gene signature method (8,10). The immune score was also found to predict survival in basal breast cancer, in agreement with a previous study that used the Profile Analysis using the Clustering and Kurtosis method (9). Similarly, a higher immune score was associated with a longer survival time in melanoma and ovarian cancer, but earlier recurrence in hepatocellular carcinoma-adjacent tissue. The degree of tumour-infiltrating lymphocytes in particular activated CD8+ T-cells, within melanoma positively correlates with better prognosis (26). The decreased recruitment of tumor-infiltrating lymphocytes may lead to poor prognosis in high-risk ovarian cancer patients (27). It has been previously demonstrated that the poor prognostic signature in hepatocellular carcinoma-adjacent tissue involved genes associated with inflammation, including interferon signaling and activation of nuclear factor-κB and tumor necrosis factor (28). These results indicated that non-tumor liver tissue could serve a prognostic role in patients with early-stage disease. These results may suggest the dual host-protective and tumor-promoting roles of immune cells in different tumor types (29,30).
The results of the present study indicated that research into microenvironment-associated cells is warranted in patients with cancer. Understanding the effect of the microenvironment on drug sensitivity may improve the efficiency of targeted therapies (31). For instance, an association has been demonstrated between the initiation of metastasis in CRC and the TGF-β stromal program (7). Even though the ESTIMATE algorithm is based on cancer tissue data, it was found to be effective in assessing cellular data as well (Fig. 4). In the present study, the ESTIMATE algorithm was used on cell line data, identifying a positive correlation between 5-FU treatment response and the immune score and indicating the potential mechanism of the drug (25). A lower immune score was observed in p53-mutant CRC patients, indicating that 5-FU may not be the optimal treatment choice for p53 mutant patients. Indeed, several clinical studies have reported that CRC patients with wild-type p53 benefit from 5-FU-based chemotherapy, but those with mutant TP53 do not (32,33). Thus, robust patient stratification using microenvironment data may aid the development and application of cancer therapies (34).
It is reasonable to apply the ESTIMATE scoring to a specific tissue-based transcriptomic dataset, as the sampling criteria are identical for every specimen within a study. To the best of our knowledge, the present study is the first to demonstrate that ESTIMATE scores are indicative of patient survival, relapse, metastasis and chemotherapeutic drug resistance. The two scores may have a prognostic value, indicating that stromal and immune cells contribute to tumor clinical outcome. It was further demonstrated that immune cells were associated with CRC development and that the ESTIMATE scores may become useful indicators of tissue-based patient prognosis.