Over the past 20 years, breast cancer has been recognized to be a family of diseases with distinct pathological, molecular, and clinical features. Several classification systems have been proposed to represent the biology underpinning different manifestations of breast cancer that influence prognosis and treatment. Because effective, well-established targeted therapies exist for estrogen/progesterone receptor positive (luminal) and Her2 receptor positive subtypes, research efforts are being focused on gaining a better understanding of the biology of triple negative breast cancers (TNBC). These account for approximately 15% of cases, afflict younger women [1], are clinically aggressive and associated with early recurrences [2, 3].

TNBC is a clinical definition of convenience based on negative staining for three standard immunohistochemical (IHC) biomarkers, and their associated lack of obvious systemic targeted (endocrine or anti-Her2) treatments. Thus, it is not surprising that TNBC delineates a group with heterogeneous biology [4,5,6,7]. Indeed, the field of translational ‘omics has seen advances over the past decade that recognize at least four biologically and clinically-distinct TNBC subtypes. Many challenges impede the facile conversion of omics-based classifiers into assays with clinical utility [8], and TNBC subtyping tests have yet to become a routine part of clinical care. Cytotoxic chemotherapy remains the mainstay of systemic treatment for TNBC [9], supplemented by a still-limited set of targeted therapeutics that can improve overall survival (OS) for metastatic TNBC patients beyond a median of 15 months [10,11,12]. This is partly attributable to current clinical trial designs that indiscriminately lump together all TNBCs, ignoring subtypes that have critical biological and immunological differences, but done to hasten accrual and obtain results applicable to a larger target market should the trial read out as positive.

Nonetheless, lessons learned over the past two years including well-designed exploratory analyses of two major immunotherapy trials [13, 14] have highlighted the need to account for distinct molecular TNBC subtypes and find better ways to match TNBC patients to their best treatment.

In this review, we detail recent advancements in genome-wide DNA, RNA, and protein-based classifications of TNBC, discuss their utility in translational oncology in the context of recent developments from major clinical trials assessing chemotherapy and targeted therapy, and highlight the importance of integrating the tumor microenvironment in current TNBC subtyping to uncover relational features of tumor, stromal and immune cells as revealed by recent single cell profiling studies.

Building on the cDNA microarray-based work first described by Perou et al.[15], the existence of luminal A, luminal B, Her2-Enriched and basal-like breast cancer intrinsic subtypes has been recapitulated using different gene expression platforms and independent datasets [16,17,18,19,20,21,22]. Of these, the basal-like gene expression subtype is distinctly different from others in biology and clinical outcomes [15, 23,24,25,26]. In current clinical practice, this subtype is still approximated by TNBC IHC status, which identifies a heterogeneous biology [4,5,6,7]. Between 50–86% of TNBC have been shown to be basal-like by gene expression [11, 17, 26,27,28], meaning the terms should not be considered interchangeable.

While the genomic landscape of breast cancer in large-scale studies including The Cancer Genome Atlas (TCGA) [26] and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) [28] has been mainly characterized in the context of breast cancer gene expression subtypes, the genomic landscape of TNBC has been specifically investigated in studies using whole genome, whole exome or targeted sequencing [25, 29,30,31,32,33]. Several somatic mutation patterns have been described in TNBC, which has the highest overall mutation rate when compared to other forms of breast cancer [26]. This higher mutational burden is nevertheless characterized by very few consistently mutated genes other than TP53, highlighting the complex repertoire of somatic mutations that underpin the heterogeneity of TNBC [25, 26, 34]. Thus, developing clinically-useful biomarkers targeting the numerous low-frequency genetic events of TNBC is a considerable challenge. TNBC as a group is characterized by a complex pattern of copy number alterations implying higher chromosomal instability when compared to other subtypes [28, 35]. These patterns have been described as a sawtooth, with many small and narrow segments of deletions and duplications, but not involving high copy number amplifications [36]. EGFR, FGFR2 and MYC amplifications as well as PTEN losses are more frequent in TNBC [25, 29, 35] but appear to be more clinically-relevant when analyzed in the context of TNBC transcriptomic subtypes [6, 35].

Mutational analyses of the somatic alteration repertoire of breast cancer have contributed to the understanding of the mutational processes that characterize TNBC [31, 32, 37]. Catalogue of Somatic Mutations in Cancer (COSMIC) substitution-based mutational signatures [38, 39] of homologous recombination DNA repair deficiency (HRD) signatures (COSMIC 3 and 8) are known to be more characteristic of TNBC when compared to clock-like (1 and 5) and APOBEC signatures that dominate among luminal breast cancers [31, 32, 37]. The mutational processes underlying the high genomic instability and HRD in TNBC can be further captured as patterns of indels and structural variants known as rearrangement signatures (RS) [31]. Of these RS-1, RS-3 and RS-5 have been associated with HRD in TNBC with the advantage of subclassifying tumors according to their BRCA1/2 mutation status [31]. Small tandem duplications (< 10 kb, RS-3) are associated with BRCA1-mutation vs. large tandem duplications (> 100 kb, RS-1) which are more characteristic of BRCA wild-type TP53-mutated [31,32,33]. Sizes of tandem duplications further define biological classes as part of a distinct genomic scar enriched in TNBC known as tandem duplicator phenotype (TDP) [40]. The TDP-1 biological class (< 10 kb) is associated with BRCA1 mutated TNBC, but not BRCA2 [40]. High TDP scores have been linked with response to platinum-based chemotherapy in TNBC preclinical models; thus, TDP may serve as a predictive genomic biomarker [40].

Several tests have been developed to characterize HRD in TNBC given that BRCA1 germline mutations are found in approximately 15% of these tumors and most BRCA1-mutated tumors are of the TNBC subtype [41,42,43,44,45,46,47]. TNBC is known to display several BRCAness features through somatic BRCA1/2 mutations or germline/somatic alterations in homologous repair-related genes (e.g., PALB2, ATM, CHEK2) [48, 49], characterized by high scores using the copy number based HRD algorithm [50, 51] or the genomic-signature based HRDetect [32, 52, 53]. Epigenetic silencing of BRCA1 and RAD51C is also associated with BRCAness [54] with BRCA1 promoter methylation being exclusive to TNBC [55]. The HRDetect algorithm covers several patterns of BRCAness combining features of HRD index (loss of heterozygosity, telomeric allelic imbalance, and large-scale transitions); COSMIC substitution signatures 3 and 8; RS-3 and RS-5 signatures; and microhomology-mediated deletions to classify TNBC into HRDetect-high, HRDetect-intermediate, and HRDetect-low with the HRDetect-high subgroup associated with a better response to standard chemotherapy [32, 52]. When compared to HRD index, HRDetect has demonstrated superior performance to detect BRCA1/2 deficiency [52], a rationale that informed the evaluation of HRDectect as a predictive biomarker for carboplatin response in the phase III TNT trial of advanced TNBC, where results using the HRD index scores were negative [56].

Overall, the complexity of TNBC at both the copy number and mutational levels with many and variable low-frequency genomic events that contribute to diverse sub-clonal populations explains in part the lack of current genomic-based clinical tests to guide therapeutic choices. However, the integration of these data, especially HRD-based signatures in the context of transcriptomic TNBC subtypes, could aid in developing genomic aberration-based biomarkers in a TNBC subtype-specific manner.

Applying RNA expression profiling to TNBC cases, several molecular systems have been suggested to identify subtypes and associated biomarkers that guide treatment decisions.

In 2011, Lehmann and colleagues performed gene expression profiling of 587 TNBCs and postulated the existence of six main transcriptional subtypes: basal-like 1, basal-like 2, immunomodulatory, luminal androgen receptor (LAR), mesenchymal and mesenchymal stem-like (MSL) [4]. A follow-up study in 2016 refined these into four subtypes [57] because after a careful histological evaluation and microdissection prior to gene expression analysis, the transcriptional profiles of immunomodulatory and MSL were found to reflect tumor infiltrating lymphocytes and stromal content, rather gene expression by carcinoma cells. The revised Lehmann TNBC-type4 classification includes basal-like 1, basal-like 2, mesenchymal and LAR categories, among which basal-like 1 had a higher pathologic complete response rate (pCR) after neoadjuvant chemotherapy and a better OS [57, 58]. However, a study by Bareche and colleagues [35] which analyzed 550 TNBC samples from both TCGA and METABRIC using the original Lehmann molecular subtyping [4] still described five molecular subtypes. After exclusion and reclassification of “unstable” samples through bioinformatics methods applied to the online TNBCtype tool used in Lehmann’s original study, the authors reported that TNBC can be classified as basal-like 1, immunomodulatory, mesenchymal, LAR, and MSL [35]. However, this revised 5-subtype classification requires assessment and validation on independent datasets.

Another TNBC molecular classification was first presented by Burstein and colleagues in 2015 who performed mRNA expression and DNA profiling on 198 TNBC tumors, identifying four transcriptomic subtypes termed basal-like immune activated (BLIA), basal-like immune suppressed (BLIS), LAR, and mesenchymal [7]. These subtypes were identified using a reduced 80-gene RNA-based classifier, were shown to differ in survival and to exhibit distinctive patterns of copy number variations, gene expression and pathway enrichment [7]. Among these four categories, survival outcomes were most favorable for BLIA and worst for BLIS.

The 4 TNBC subtypes in Burstein et al.were recapitulated on an independent larger cohort of 504 Chinese TNBC cases [6]. Compared to the TCGA dataset, this study highlighted distinct patterns in the Chinese TNBC population including higher frequencies of PIK3CA mutations and LAR subtype. Zhao et al.subsequently utilized the RNA-sequencing data from this study to develop an IHC-based classifier on AR, CD8, FOXC1 and DCLK1 [59], although this lacks external independent validation yet.

Altogether, based on independent DNA and RNA level analyses, the current consensus is that TNBC comprises of at least four major molecular subtypes categorized as BLIA (20–30%), BLIS (25–40%), LAR (15–25%), and mesenchymal (15–20%) [6, 7, 35]. Characteristic mutation profile, copy number, gene expression and pathway enrichment of the different TNBC subtypes are depicted in Fig. 1.

Overview of the characteristic mutation profile, copy number, gene expression and pathway enrichment for the different triple negative breast cancer subtypes Triple negative breast cancer subtypes of luminal androgen receptor (LAR), basal-like immune suppressed (BLIS), basal-like immune activated (BLIA), mesenchymal (MES) and the possible addition of mesenchymal stem-like (MSL) are shown at the core. Genes and pathways denoted in red (indicate high expression) while those denoted in green (indicate low expression). Data are aggregated from Lehmann et al.[57], Burstein et al.[7], Jiang et al.[6], Bareche et al.[35], Bareche et al.[64], Lehmann et al.[58], Gong et al.[76] and Asleh et al.[60]. Abbreviations: TNBC, triple negative breast cancer; HRD,

The mutational profile of BLIA demonstrates that > 80% of these tumors harbor TP53 mutations, with otherwise a high number of individually low-frequency mutations [6,