Background
Two of the most predictive measures of breast cancer (BC) patient mortality are tumor progression and immune infiltration [
1‐
3]. By decoding and recognizing the underlying molecular patterns of invasive breast tumors, clinicians may provide high-grade tumor patients with appropriate prognosis and treatment, while monitoring the potential progression of lower-grade cancers [
4,
5]. Breast tumor invasiveness and patient prognosis are related to molecular subtypes, which are currently classified through PAM50 mRNA expression or immunohistochemistry staining of hormone receptors [
6,
7]. BC patients with luminal tumors, defined by the expression of the estrogen and/or progesterone receptor (ER
+|−, PgR
+|−), are known to have the best overall outcome [
8,
9]. Luminal A type tumors are associated with a slightly better patient survival rate than luminal B tumors, which have high expression levels of Ki-67 (> 14%), and in some cases,
human epidermal growth factor receptor 2 (HER2) amplification [
8,
9]. Patients with estrogen- and progesterone receptor-negative (ER
−, PgR
−), Her2-amplified tumors, have poorer outcomes than those with luminal subtypes, even though this group of patients has been shown to respond well to targeted therapy [
10]. The basal-like and triple-negative breast cancer (TNBC) subtypes, which are largely overlapping and classified by the lack of hormone receptor expression (ER
−, PgR
−, Her2
−) [
11], have the poorest prognosis among the subtypes [
8,
9].
A precise characterization of the degree of breast tumor invasiveness, alongside the biological relevant pathways and underlying molecular mechanisms, hinges on the identification of a set of specific and sensitive biomarkers.
Recent studies suggest that circulating microRNAs may have great potentials as cancer progression markers [
12‐
14], partially due to their high stability in the plasma/blood [
15,
16]. Not only does the level of externalized miRNAs reflects the molecular events underlying tumor progression but, importantly, some studies point to a functional role of tumor-secreted circulating miRNA in intracellular communication and tumor reprogramming [
17‐
19]. Tumor cells may release micro-vesicles into the extracellular space, which may then be taken up by other cells (tumor, epithelial, or immune) via endocytosis [
20]. Some micro-vesicles have been found to not only contain mature miRNAs, but pre-miRNAs with accompanying RNA-induced silencing complexes (RISCs) [
21]. Uptake of the pre-miRNA exosomes by recipient cells resulted in an efficient silencing of target mRNAs and reprogramming of the cellular transcriptome [
22]. In accordance, it has been reported [
23] that the release of miRNAs within exosomes was not merely a reflection of the abundance of a given miRNA species, but a selective process facilitated by the tumor cells [
23,
24]. For example, exosome-mediated transport of miR-10b from BC cell lines has been shown to promote tumor cell invasiveness in other BC cell lines, which were otherwise not invasive [
24].
Circulating miRNAs may also be found free of exosomes, either in complexes with argonaute proteins [
25] or bound by high-density lipoprotein (HDL) [
26]. HDL-bound circulating miRNAs are delivered to recipient cells, via the scavenger receptor class B/type I-dependent and uptake of these results in targeting of mRNA reporters [
26]. MicroRNA silencing of gene targets is facilitated through the interaction of the mRNA 3′ UTR, with the ~ 8 nucleotide seed sequence within the miRNA [
27]. Mature miRNAs, which have identical seed sequences, are classified as belonging to the same miRNA family [
28]. Because seed sequences of family members are complementary to the same binding motifs, these miRNAs are thought to regulate the same target genes [
27,
28]. In a study from 2013, Hamilton et al. [
29] identified a pan-cancer oncogenic microRNA family, which was responsible for the co-regulation of central tumor suppressors and, in general, genes functioning within the same pathways. As members of a miRNA family can act as gene co-regulators, it is possible that these are also co-secreted by tumor cells. Studying the relationship between tumor externalized miRNAs should identify which biological processes and pathways are specifically modulated within the tumor cell itself and potentially targeted in recipient cells. Indeed, one very interesting aspect of tumor-secreted miRNA reprogramming is a possible connection to disease progression through the modification of both neighboring and distal tissues. Le et al. [
30] showed that exosomes containing miR-200 from metastatic human breast cancer cell lines were absorbed by non-metastatic cells, resulting in the promotion of mesenchymal-to-epithelial transition [
30]. The study found that miR-200-expressing tumors used extracellular vesicles to drive metastasis of otherwise weakly metastatic cells at distant sites, providing these cells with the ability to colonize distant tissues in a miR-200-dependent manner [
30]. Similarly, it has been reported that cancer cells can suppress glucose uptake by non-tumor cells in the pre-metastatic niche, by secreting micro-vesicles containing miR-122 [
31]. Repression of miR-122 restored the glucose uptake in distant organs, while decreasing the incidence of metastasis and disease progression [
31].
As such, identification of specifically secreted miRNAs may not only help to improve patient diagnosis/prognosis and disease monitoring, but could also relay information about which target genes are central for particular tumor properties, including invasiveness, and how these properties may be promoted by tumor cell communication.
Despite their potential usefulness, however, identification of robust circulating miRNA biomarkers is no trivial task, as a range of non-cancerous events may cause changes in the levels of biomolecules [
32]. Blood-based biomarkers are especially dynamic and can be affected by the time of sampling, patient diet, level of physical activity, medication, and other biological variances, which are extremely difficult to take into account [
33]. Furthermore, the serum/plasma may be considered a difficult starting material for marker discovery as cancer-related macro-molecules will be highly diluted and buried in a complex serum/plasma secretome [
34,
35].
In recent years, the importance of the tumor microenvironment has become a central area of cancer research, as multiple studies have shown how cancer cells modulate the mechanisms of the surrounding stromal cells in ways that enable the tumor to induce angiogenesis, sustain proliferation, and evade immune destruction [
36]. Cross-talk within the tumor stroma is facilitated by the tumor interstitial fluid (TIF), which forms the interface between circulating body fluids [
37]. In the local tumor environment, stromal cells and tumor cells are surrounded by TIF, allowing for the secretion and uptake of ions, miRNAs, proteins, and other signaling molecules [
38,
39]. As a result, TIF is thought to modulate the epigenetic program of non-malignant cells by tumor cells and vice versa, demonstrating the importance of local tumor milieu for cancer progression [
40,
41]. In addition to molecules secreted from tumor and healthy stromal and epithelial cells, TIF encompasses externalized biomolecules from immune cells in the tumor microenvironment [
42]. Tumor immune cell infiltration has been shown to be central for the prediction of patient response to treatment and overall survival [
2]. The relationship between lymphocyte infiltration and tumor progression is multifaceted [
43]. A number of studies have found that a higher degree of CD8+ T lymphocytes is associated with a better outcome for patients with BC [
43], especially for the TNBC and Her2-enriched subtypes [
44].
In contrast, tumor-infiltrating CD4+ T lymphocytes have been linked to a poorer overall survival. This may be related to the expression of PD-L1 (
programmed death-ligand 1) by some populations of tumor-infiltrating lymphocytes (TILs), as PD-L1 is a major inhibitor of an anti-tumor immune response [
45,
46]. In accordance, the degree of immune infiltration by PD-L1+ T lymphocytes was found to be correlated with large tumors, high-grade tumors, and positive lymph node status [
45,
46]. It should be noted that the role of PD-L1 in tumor immune escape is complex, with tumor cells themselves as well as some populations of immune cells displaying this protein, and contributes to anti-immunity in a context-dependent manner [
47]. Interstitial fluids provide a snapshot of circulating tumor molecules, as well as immune cell-secreted biomolecules associated with tumor properties such as growth and response to therapy [
2,
48]. As the concentration of cancer-specific biomolecules within the local tumor milieu is estimated to be 1000–1500 times that of blood, TIF is a unique resource for BC biomarker identification and a promising alternative to a highly diluted serum secretome [
37,
38].
In this study, we analyzed a set of secreted miRNAs from tumor and normal interstitial fluids acquired from 60 women with breast cancer [
49]. The availability of clinicopathological information, including tumor grade, receptor status, and BC subtypes classification as well as the characterization of immune infiltration of every biopsy, allowed us to investigate the relationship between interstitial fluid miRNA levels and patient clinical features. We subsequently identified the deregulated gene targets of IF miRNAs in tumor tissues from the same cohort of patients [
50]. Integration of miRNA and mRNA expression data helps to pinpoint the perturbed pathways responsible for breast cancer progression while strengthening biomarker selection by utilizing the combinatorial power of a bi-molecular expression profile.
Discussion
In this study, we integrated interstitial fluid miRNA abundances with expression levels of mRNA from paired tumor tissues. Our analysis allowed us to explore whether miRNAs secreted into the interstitium could be associated with differentially expressed gene targets and whether these targets were co-expressed and/or co-regulated. We partitioned the data based on sample molecular and clinical information to obtain sets of differentially expressed IF miRNA and their intracellular gene targets, hereby elucidating potential pathways and mechanisms underlying breast cancer.
As expected, we observed a good separation of BC subtypes based on intracellular mRNA expression; however, this was not the case for the interstitial fluid miRNAs. Although we did see some clustering of TIF from TNBC samples, this could just as well be related to the common higher immune status and higher tumor grades of these samples [
50]. Other studies on circulating miRNA expression in BC patients have found similar trends, with a poor distinction of different subtypes, except for TNBC (or basal-like) tumors [
108‐
110]. The majority of differentially abundant interstitial fluid miRNAs identified in our study were DE in the contrast of normal vs cancer or associated with immune infiltration and tumor grade. These results are in accordance with previously published literature on circulating miRNA BC from the serum/plasma, as these most often highlight miRNA profiles related to cancer progression, invasiveness, metastasis, and relapse [
111‐
114]. In more quantitative measures, this is supported by a PubMed search on titles and abstracts (the “
Materials and methods” section). A search on terms related to circulating miRNAs + cancer progression yielded 757 results (18 titles), whereas the search with terms circulating miRNAs + subtype only returned 106 results (four titles). A further look into the four articles with subtype terms in the title revealed none of them to find differences between PAM50 or immunohistochemistry subtypes.
When comparing the results of our miRNA analysis to those obtained in the original study by Halvorsen et al. [
49], results were highly variable. We hypothesize that discrepancies mainly arise from the following:
-
Choice of statistical framework. limma [
53], which was employed in our analysis, is likely to return a larger number of significant DA miRNAs, compared to the Kruskal-Wallis test used in the original analysis [
49]. This is due to limma’s underlying Bayesian properties, which help overcome issues relating to small sample sizes and miRNA-specific variances.
-
Correction for batch effects and confounders. Clustering of datasets revealed significant confounding of covariates; as such, we incorporated information on confounders into the design matrix for generalized linear modeling with limma.
-
Integrative analysis. As we performed an integrated analysis, including co-abundance analysis and collective analysis of both TIF miRNA data and paired intra-tumor mRNA data, we naturally curated our results based on miRNA abundances, as well as the relationship between miRNAs and predicted differentially expressed mRNA targets. As such, we obtained a very different set of miRNA top candidates for further analysis and validation.
In addition to the aforementioned, other differences may have contributed to varying results, (i) how comparisons were defined, (ii) cutoff for retaining a miRNA in the dataset, (iii) cutoff for significance (log fold change was added as a criterion in our analysis), and (iv) missing value imputation.
We believe that the solid bioinformatic framework and data integration implemented in our study have resulted in new and valuable biological insights while highlighting the major impact of data correction and choice of statistical set-up has for down-stream results.
Overall, our analysis revealed a set of miRNAs which were upregulated in tumor interstitial fluids from mainly TNBC patients with high-grade and high immune infiltration score tumors. Subsets of these miRNAs were predicted to have target genes, which were also differentially expressed in tumors from the same cohort of patients. Table
4 shows the overall best candidate interstitial fluid miRNAs and predicted gene targets, based on all analyses and database support.
In the following sections, we will discuss some of these most interesting miRNAs and genes, in greater detail.
Breast cancer subtypes
Common to subtypes
The most interconnected DE miRNAs from the comparisons of BC subtypes were miR-9, miR-15b, miR-17, miR-19a, and miR-30d. We found these to be depleted in tumor interstitial fluids from patients with luminal and Her2-enriched breast cancers, compared to samples from TNBC patients. Interestingly, all of these miRNAs have been shown to be highly abundant in the basal-like BC subtype, which is largely similar to TNBC [
115,
116]. Patients with basal-like/TNBC tumors are known to have the poorest prognosis, and this subtype is associated with high-grade and rate of metastasis [
8]. In accordance with this, miR-17 and miR-19a belong to the miR-17-92 cluster, also denoted OncomiR-1 (13q31.3) [
117]. We found these two miRNAs, along with other members of OncomiR-1 (miR-19b, miR-18a, and miR-20a), to be differentially co-expressed. The miR-17-92 cluster of miRNAs has been shown to target the well-studied tumor-suppressor PTEN (
phosphatase and tensin homolog), as well as key players involved in TGF-
β (
transforming growth factor beta) signaling [
118].
Multiple studies on miR-17 have found an association between the over-expression of this family of miRNAs with poor patient prognosis (poor disease-free survival and overall survival) [
119,
120] and, in connection with this, cancer cell migration and invasion in breast cancer [
121,
122].
Over-expression of miR-30d and miR-9 has been associated with an aggressive phenotype, shorter time to recurrence, and a poor prognosis in patients with breast cancer [
123,
124]. More specifically, miR-30d is proposed to be an inhibitory regulator of autophagy [
125], and the miR-30 family of miRNAs is thought to promote non-attachment growth of breast cancer cells [
126].
MiR-9, miR-15b, miR-17, miR-19a, and miR-30d were predicted to interact with a set of differentially expressed genes, some of which were common to the three subtype comparisons. Common genes were
AR (
androgen receptor),
CERS6 (
ceramide synthase 6),
FOXA1 (
forkhead box A1),
GPR160 (
G protein-coupled receptor 160),
KIAA1244 (
ARFGEF family member 3), KLK5 (kallikrein-related peptidase 5), SPDEF (
SAM pointed domain-containing ETS transcription factor), and
XBP1 (
X-box binding protein 1), all of which were upregulated in luminal types and Her2-enriched TIF samples vs TNBC. Three of these genes belonged to the PAM50 set:
AR,
FOXA1, and
GPR160 [
7], while the remaining genes had all been individually associated with breast cancer subtypes [
92‐
96].
Luminal subtypes
While some genes were common to the three contrasts, others were subtype-specific, such as
ESR1 (
estrogen receptor 1),
KIF3B (
kinesin family member 3B), KRT4 (
keratin 4), and
NFIB (
nuclear factor I B), which were associated with luminal types only.
KIF3B was upregulated in the luminal samples, and in accordance with this,
KIF3B has been shown to be over-expressed in ER-positive tumors, with estrogen directly inducing the expression of KIF3B [
127]. KRT4 and NFIB were downregulated in luminal subtypes compared to TNBC.
KRT4 and
NFIB have both been shown to be over-expressed in basal/TNBC tumors [
128,
129], supporting our findings. The expression levels of keratins change during metastatic progression of breast cancer, and over-expression of some keratins have been associated with poor patient survival [
130]. Of particular interest was NFIB, which has directly been proposed as a potential gene target for ER-negative breast tumors. NFIB was found to be over-expressed in TNBC compared to ER-positive tumors, and over-expression of this gene was associated with a high nuclear grade [
129].
ESR1 and
CERS6 (see the section above) were co-expressed in the green module, along with
C6orf211 (
ARMT1,
acidic residue methyltransferase 1),
CCND1 (
cyclin D1), and
THSD4 (
thrombospondin type 1 domain containing 4). This set of genes has been suggested as markers for a prognostic luminal signature [
104] and has more recently been highlighted as the key players in a novel,
FOXA1/ESR1-interacting pathway [
131], highlighting their association with estrogen receptor status.
Luminal A subtype
The gene
CCND1 was upregulated and highly interconnected in the luminal A vs TNBC comparison.
CCND1 is a well-studied breast cancer driver gene [
132], the amplification of which is more prevalent in luminal subtypes compared to Her2 and basal-like [
133]. Amplification of this gene has been found to be more prevalent in luminal B tumors compared to luminal A [
134]. However, as
CCND1 amplification is also associated with a more aggressive phenotype within both luminal subtypes, as well as in familial and sporadic tumors [
134], this might explain the slight discrepancy we observe here, e.g., different compositions and sizes of luminal sets, resulting in this gene not reaching significance in the luminal B vs TNBC comparison. A comparison of miRNA-gene pairs with MiRTarBase resulted in support for the
CCND1 gene and its predicted miRNA regulators.
Luminal B subtype
For the contrast of luminal B vs TNBC, the
ELOVL6 (
ELOVL fatty acid elongase 6) gene was found to be upregulated and interact with both a larger number of genes and miRNAs. A high level of
ELOVL6 (oncogene in prostate cancer) has been proposed to be a marker of poor prognosis in BC [
135], which is of great interest, as patients with luminal B type tumors generally have poorer outcomes than those with luminal A types [
8]. Dysregulated expression of genes involved in mammary gland fatty acid and phospholipid metabolism, such as the
ELOVL6 gene, have been proposed to characterize cell proliferation and differentiation state, and many of these have been linked to BC patient survival [
136]. STRING network analysis with the set of eight co-expressed from the red module (including
ELOVL6), returned the gene ontology term sphingolipid metabolism. Genes assigned to this term were
ELOVL6,
NEU1 (
neuraminidase 1), and
SERINC3 (
serine incorporator 3). A literature search revealed that another gene from this module,
FLOT1 (
flotillin), had recently been linked to the sphingolipid pathway, proposed to be a regulator of cellular sphingolipid distribution and signaling [
137]. All genes from the red module, except
CCN1, were specifically upregulated in luminal B vs TNBC, but not in luminal A type, indicating that over-expression of sphingolipid-related genes might be specific to luminal B tumors. Interestingly,
ELOVL6,
NEU1, and
SERINC3 were the predicted targets of miR-23a, which was also highly interconnected and downregulated in the luminal B vs TNBC comparison. A literature search for miR-23a revealed this miRNA to be a well-known oncogenic miRNA, and a recent study by Ma et al. [
138] found that over-expression of miR-23a induced EMT, migration, invasion, and metastasis of breast cancer both in vitro and in vivo [
138]. miR-23b has been proposed to be a circulating biomarker for BC diagnosis, subtyping, and disease recurrence [
139], many times over, highlighted by a novel review on this miRNA [
140].
Estrogen-positive tumors
The DE expression network generated for ER
+ vs ER
− tumors and interstitial fluids showed
ESR1,
GATA3 (
GATA binding protein 3), and
GREB1 (
growth-regulating estrogen receptor binding 1) to all be upregulated, while
ERBB2 (
Erb-B2 receptor tyrosine kinase 2) was downregulated. Both
GATA3 and
GREB1 have been linked to estrogen receptor-positive breast tumors and have been proposed as markers for patient response to hormone treatment [
141‐
143].
The most interesting miRNA from this network was miR-32-5p, which was over-expressed in ER
+ tumors vs ER
− tumors, and the most interconnected miRNA in the network. Not much is known about this miRNA in connection with breast cancer; interestingly, however, miR-32-5p interacts with genes
NFIB,
SOX11 (
SRY-box 11), and
DSC2 (
desmocollin 2) (downregulated in ER
+ vs ER
−), all three of which are known to be over-expressed in basal-like/TNBC/ER
− tumors and associated with poor survival [
129,
144,
145].
Her2-enriched subtype
Specific to the contrast Her2-enriched vs TNBC, were genes ERBB2, GRB7 (
growth factor receptor bound protein 7) and LASP1 (
LIM and SH3 protein 1), all of which were upregulated. These genes are well-supported central players in Her2-enriched cancers and belong to the Her2 amplicon (chromosome region 17q-12-21) [
98]. ERBB2 and GRB7 are both Pam50 genes [
7]. Another gene specific to the Her2 set was CPD, which overall had the most interactions in the miRNA-mRNA network. CPD (
carboxypeptidase D) is another gene known to be amplified in patients with Her2-enriched tumors on chromosome 17, right upstream of ERBB2 (chromosome region 17q-11-2) [
146].
Tumor-infiltrating lymphocyte scores and tumor grade
Analysis of the miRNA-mRNA pairs differentially expressed in high TILs (2, 3) vs low TILs (0,1) revealed the
NEDD4L (
NEDD4 like E3 ubiquitin protein ligase) gene, to be paired with the highest number of miRNAs.
NEDD4L, which was downregulated in samples with high tumor-infiltrating lymphocytes scores, has been shown to be a negative regulator of Wnt-signaling [
147]—a pathway often perturbed in cancer [
148]. Wnt signaling is central in the regulation of immunity and has been reported to facilitate immune evasion via dendritic cells and T regulatory cells [
149]. In a study by Ding et al. [
147] on
NEDD4L inhibitory effects on the Wnt signaling, it was noted that
NEDD4L is often found to be downregulated in cancers, while its Wnt-target Dvl (
disheveled), which is modified by
NEDD4L for proteasomal degradation, is often upregulated in the same cancers [
147]. This could indicate that the accumulation of Dvl contributes to an oncogenic type of Wnt signaling. Furthermore, the downregulation of
NEDD4L has been implicated in the initiation of breast tumor development, and this gene has been proposed as a prognostic lung cancer marker linked to histological grade, tumor stage, and lymph node metastasis [
86,
150]. These findings support the results of our analysis, as samples with high immune scores were also those with a high histological grade (grade 3 tumors). Additionally, our analysis revealed that NEDD4L interacts with PARD6B (
Par-6 family cell polarity regulator beta) and CGN (cingulin). These two genes were co-expressed in the green module and were downregulated in cluster 1 (high TILs and high-grade, mainly TNBC) vs cluster 2 (low TILs and lower grade, mainly luminal).
PARD6B,
CGN (
cingulin), and
NEDD4L belong to the KEGG pathway, tight junction (TJ) (hsa04530). Aberrant levels of tight junction proteins result in incorrect formation and maintenance of cellular polarity, contact inhibition, and proliferation, contributing to epithelial-mesenchymal-transition (EMT) [
151].
PARD6B expression is critical for TJ assembly, and decreased expression of this gene has been proposed to result in epithelial cell changes and tumor metastatic behavior [
152].
PARD6B has been shown to be amplified in breast cancer [
153]; however, in a comparison of BC subtypes, the expression of this protein was specifically proposed to be upregulated in the luminal type compared to basal-like and Her2-enriched tumors [
154]. This observation is in accordance with our findings, as it was had a higher expression level in cluster 2 than in cluster 1. Although
PARD6B is generally considered to be an oncogene, it has also been linked to suppression of cell proliferation in breast cancer, indicating that the role of this gene may be complex [
155].
PARD6B,
CGN (
cingulin),
and
NEDD4L were all predicted targets of the OncomiR-1 (13q31.3) cluster, or one of its paralogues 106a/363 (Xq26.2) and 106b/25 (7q22/1), miRNA included miR-17, miR-19a/b, miR-20a/a, and miR-106a/b. In accordance with this,
NEDD4L has experimentally been shown to be the gene target of the miR-106-25 cluster miRNAs [
86].
Tumor grade
Network analysis of miRNA-mRNA DE pairs high-grade tumors (grade 3) vs medium/low-grade tumors (grades 1, 2) revealed two genes of interest. One of these genes,
BTRC (
beta-transducin repeat containing E3 ubiquitin protein ligase) predicted to interact with miR-10a/b and miR-107. Interestingly, we found these three miRNAs to be co-abundant (module 1, Fig.
5). One study on miRNA-10b found that this miRNA was secreted via exosomes and that the uptake of these exosomes by recipient cells resulted in a decrease of target gene levels and induced invasiveness in otherwise non-malignant cells [
24]. Whereas miRNA-10b is generally considered to promote tumor progression and metastasis [
156], the role of miR-107 in breast cancer seems less straightforward. Some studies suggest that miR-107 has a tumor-suppressive role [
157], while others have found that over-expression of this miRNA promotes tumor progression, is associated with lymph node metastasis and poor patient prognosis [
113,
158]. Just as for miR-107, the role of β-TrCP (encode by
BTRC) in cancer development and progression is convoluted.
BTRC has been proposed to be a DRG, having oncogenic properties in one context and anti-tumor functions in another [
159]. More recent literature on β-TrCP, however, suggests that this protein indeed suppressed tumor progression, as one study showed that β-TrCP regulates the degradation of CDK1, high levels of which promote certain aspects of tumor malignancy [
160]. Another study on β-TrCP in glioma found that a low level of this protein was associated with a poor prognosis [
161]. Our results agree with these studies; we see a downregulation of BTRC in high-grade tumor tissues and an upregulation of miR-10a, miR-10b, and miR-107 in matched interstitial fluids of these tumors. Importantly, the interaction between miR-10a and the BTRC transcript has been experimentally validated (luciferase reporter experiment) [
162]. Although we could not find any experimental validation for the BTRC-miR-107 interaction, a study by Yang et al. [
163] found that a combination of miR-107-BTRC-UBR3-miR-16 expression could distinguish between different BC subtypes, specifically between basal-like tumors and luminal types [
163].
Another gene of interest in relation to tumor grade was
CHST1 (
carbohydrate sulfotransferase 1), which was paired with miRNAs miR-301a/b and miR-454. Analysis revealed miR-301a/b and miRNA-454 to be co-abundant in the same module as miR-10a/b and miR-107 (module 1,
5), supporting the notion that these miRNAs might be associated with tumor grade and progression. The literature on
CHST1 and cancer is very limited; however, studies on other members of the carbohydrate sulfotransferase (CS) family show that while some CS members may be oncogenic, others could have tumor suppressor functions. Overexpression of
CHST3 and
CHST11 have been linked to BC aggressiveness, relapse, and development of metastasis [
164]; in contrast, downregulation of
CHST10 and
CHST14 has been linked to invasive melanoma and to late stages of colon cancer progression, respectively [
165,
166]. In the current study, we found
CHST1 to be downregulated in grade 3 vs grade 1|2 tumors. The miRNAs predicted to interact with
CHST1 are more well-studied then their target. MiR-301 is thought to be a breast cancer oncomiR, which promotes tumor invasion and nodal or distant relapses via direct interaction with
FOXF2,
PTEN,
BBC3iso-2, and
COL2A1 [
167]. This microRNA has also been shown to help regulate cancer-related immunity in solid tumors [
64]. In accordance with this, we found miR-301a and miR-301b to be upregulated both in the contrast of IF from high grade to medium/low grade and between high TILs and low TILs. High expression of miR-454 has been associated with a poor overall and disease-free survival in patients with TNBC [
168]. These findings were supported by a meta-study by Lu et al. [
169], although this review also highlighted the fact that miR-454 might have a dual role, exerting oncogenic effects in some cancer types, such as breast cancer, and tumor-suppressor functions in other types of cancer [
169].
Tumor-infiltrating lymphocytes
Analyses revealed a set of TIF miRNAs and co-expression gene targets, which were associated with tumor-infiltrating lymphocyte scores. This network included miRNAs; miR-103a, miR-136, miR-146a, miR-299, miR-301, miR-346, miR-369, and miR-494 predicted to interact with gene transcripts:
ASB2,
BCL11B,
BTLA,
CD3D,
CD3G,
CXCL13,
CXCR5,
FAM65B,
IKZF3,
IL7R,
KCNA3,
KLRC4,
LAMP3, and
LTB. This set of genes, which were all upregulated in high-TIL vs low-TIL samples, has all been linked to immune system processes [
170‐
173]. Genes such as
CD3D and
CD3G encode T cell surface glycoproteins and are well-known players in anti-tumor immunity [
174]. High levels of these two antigens have been linked to an overall better prognosis of patients with breast cancer [
175,
176]. The same is true for
BCL11B,
IKZF3, and
KLRC4, which have very recently been linked to a prognostic immunogenic signature of triple-negative breast cancers [
173,
177]. Liu et al. [
177] found that almost all populations of immune cells, immune system pathways, and their genes were enriched in TNBC compared to both normal samples and other breast cancer subtypes. This is in accordance with our findings; we see TNBC having not only overall higher grade but also infiltrating lymphocyte scores.
Of particular interest was the co-expression of genes:
BTLA,
CXCR5,
CXCL13,
IL7R,
LAMP3, and
LTB. The protein products encoded by genes have been linked to the presence or absence of high endothelial venules and tertiary lymphoid structures in multiple cancer types [
178‐
181]. TLS, which are lymphoid formations, have been found within tumors where they are thought to participate in anti-tumor responses. A high number of TLS is generally associated with an overall better patient’s survival in a range of different types of cancer [
182], and their presence correlates with the level of both TILs and HEV [
179,
180]. These observations are supported by the fact that high endothelial venules, which are specialized vessels normally found in the lymph nodes, are proposed to act as gateways for the infiltration of lymphocytes within tumors [
183]. The abundance of lymphoid chemokines such as CXCR5 and
CXCL13 has been linked to both the presence of TLS and HEV in breast cancer stroma [
181]. Tertiary lymphoid structures are modulated by a network of cytokines, and the central players in this network are lymphotoxin LT-β-related cytokines [
179]. One study [
178] found that lymphotoxin LT-β was overexpressed in breast tumors and that overexpression of LT-β was correlated with a high density of HEVs and dendritic cells. Dendritic cells are thought responsible for the production of LT-β in tumor tissues in general and in tertiary lymphoid structures. These findings might indicate that a high level of LT-β should be predictive of a better patient outcome. However, another study on the LT-β network in mice has shown that high levels of lymphotoxin LT-β promote a tumor-permissive microenvironment resulting in tumor progression [
184]. The results of our analysis support those from the aforementioned studies, with this set of genes found to be upregulated in samples with high levels of lymphocyte infiltration. For a more in-depth description of the relationship between TLS, TILs, and HEV, as well as the roles of
BTLA,
IL7R, and
LAMP3 in relation to these, we refer to the original publications [
178‐
180,
184].
The set of co-expressed immune genes discussed above was mainly predicted to be the targets of miR-146a and miR-494. We found these miRNAs, along with miR-206, miR-369, and miR-376a, to be co-abundant (module 2, Fig.
5). Both miR-146a and miR-494 have been linked to immune system response in connection with tumor development [
185‐
188]. miR-146a is a central player within the innate immune system, where it functions as a fine-tuning mechanism, modulating the scale of immunity vs tolerance [
189]. Generally, this miRNA is considered a negative regulator of immune response. This is supported by mouse knock-down experiments, in which loss of miR-146a was shown to result in autoimmunity and development of myeloid malignancies [
189,
190]. Re-establishing miR-146a expression within breast cancer has been shown to decrease the levels of immunostimulatory genes and to antagonize NF-kB signaling, reducing cancer cell migration and metastatic mechanisms [
185,
187]. The role of miR-494 in cancer immunity is not straightforward. One study found that this miRNA might help prevent anti-tumor immunity through the accumulation of myeloid-derived cells in the microenvironment, promoting tumor growth [
186], while another study showed that miR-494 suppresses the progression of breast cancer, through downregulation of CXCR4-mediated oncogenic communication [
188]. Although miR-146a and miR-494 had the most gene targets within the co-expressed immune gene cluster, other miRNAs were also of interest here among miR-103a, miR-301, and miR-369 all of which have been linked to tumor immunity [
64,
191]. A search thought the CMEP database revealed all of these to be DE in the blood of BC patients. miR-103a, miR-301a, miR-494, and miR-369 were all downregulated in TNBC compared to other subtypes. This is in full accordance with our results, as high immune score tumors were mainly TNBCs.
Conclusion
We identified genes that were differentially co-expressed between tumors with high and low infiltrating lymphocyte scores—most of these had already been associated with cancer immunity through other studies [
170‐
173]. Of particular interest were
CXCL13,
BTLA,
IL7R,
LAMP3, and
LTB as these genes have been linked to the presence of tertiary lymphoid structures (TLS) and high endothelial venules (HEV) within tumors. TIF miR-146a and miR-494, the most interconnected and co-abundant miRNAs in this cluster, were both previously annotated as negative regulators of immune-stimulatory genes and were DE in the plasma from patients with BC [
158,
192]. As tumor immune cell infiltration is highly related to patient prognosis [
2], we propose genes and miRNA from this module to be candidate markers of tumor immune status, prognosis, and potentially patient response to immunotherapy.
Another co-expression module encompassed genes, which were DE between luminal B tumors and TNBC. A subset of these was related to sphingolipid metabolism and predicted to be co-regulated by miR-23a. miR-23a has been found to be differentially abundant in the serum of healthy individuals and breast cancer, as well as between BC patients with different subtypes [
140]. As such, this miRNA is a candidate marker for BC subtype and potentially a new therapeutic target. TIF miRNAs DE between subtypes were all identified in contrasts of TNBC vs another subtype. Many miRNAs identified in these contrasts were generally related to BC progression and metastasis, such as members of the OncomiR clusters and miRNA families miR-30 and let-7. This observation is supported by other studies on secreted miRNAs, and we therefore propose that levels of secreted miRNAs do not reflect gene-based subtyping, but rather tumor aggressiveness, i.e., TNBC patients often have higher-grade tumors and a poor prognosis.
A small set of genes and TIF miRNAs were more specifically associated with tumor grade, here among miR-10a/b and gene target
BTRC. The interaction of miR-10-
BTRC has been experimentally validated [
162], and miR-10b was found to be delivered via exosomes to recipient cells, resulting in the downregulation of target genes [
24].
BTRC is proposed to have tumor-suppressive functions [
160,
161], while miR-10b is oncogenic; as such, it should be of interest to study this pair in relation to tumor invasiveness and metastasis.
Collectively, integration of expression data from interstitial fluid miRNAs and paired solid tissue mRNAs resulted in sets of miRNA-mRNA pairs, associated with underlying molecular mechanisms and clinical features of breast cancer.
Whether TIF miRNAs highlighted in our study are indeed transferred between cells in the tumor microenvironment, or whether these merely reflect that level of miRNAs within the tumor donor cells themselves, is unknown. However, as the uptake of miRNAs from the extracellular space is a well-known phenomenon, communication and transcriptome regulation via interstitial fluid miRNAs are an attractive therapeutic angle for cancer treatment.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.