Breast cancers can be classified into six intrinsic molecular subtypes by gene expression (GE) profiling (17, 24) as follows: Luminal A, Luminal B, Her2 enriched, normal-like, basal-like, and Claudin low. Each subtype is identified within the TNBC group as defined by immunohistochemistry. Basal-like tumours are most frequent (50–75%). However, they are not exclusive to the TNBC phenotype (24). The claudin-low subtype represents 25–40% of TNBC and was more recently introduced (25).

Basal-like tumours are characterised by the presence of cytokeratins typically expressed by the basal layer of the skin, widespread genomic instability, high proliferation markers, loss of function of BRCA1, and dysregulation of MYC and RB1 pathways (24). Claudin-low tumours have several features in common with basal-like tumours but are uniquely characterised by low levels of cell adhesion proteins, the enrichment of mesenchymal traits and stem cell features (26). Luminal tumours overexpress a ‘luminal signature’ containing ESR1, GATA3, FOXA1, XBP1, and MYB. Her2 amplification concomitantly with overexpression of HER2-amplicon-associated genes defines the Her2 enriched subtype (24).

Intrinsic subtypes provide independent predictive information regarding the response to neoadjuvant chemotherapy (NACT) when considering all subtypes of breast cancer, although not consistently for the TNBC cohort when viewed in isolation. Claudin-low tumours are associated with lower pathological complete response (pCR) rates compared to basal-like subtypes (27). In a subgroup analysis of the BrighTNess trial, pCR rates were higher for basal-like vs. non-basal tumours (52.3% vs 35.4%, p=0.003) (27). In contrast, no difference in pCR rate was observed with the addition of carboplatin for patients with basal-like TNBC vs non-basal TNBC in the CALGB40603 study (28). These results illustrate that the predictive value often linked to the basal-like subtypes has not always been reproduced in the early setting of TNBC, making intrinsic subtypes less reliable biomarkers of response within this group.

Lehmann et al. selected clustering analyses to identify six TNBC subtypes displaying unique GE patterns and ontologies. Each subtype is characterised by the activation of specific signalling pathways that lead to a selective response to targeted therapies in vivo (18). Additional histopathological quantification and laser-capture microdissection prompted a refined classification with only four tumour-specific subtypes (TNBCtype-4). The original immunomodulatory and mesenchymal stem-like subtypes were deemed to originate from infiltrating lymphocytes and tumour-associated stromal cells, therefore excluding the impact of these elements from the classification. The new approach demonstrated differences in clinical baseline characteristics and both local and distant disease progression (29). Basal-like 1 (BL1) revealed increased markers of proliferation and elevated expression of the DNA damage response (DDR) genes. Basal-like 2 (BL2) is characterised by features of basal/myoepithelial origin and activation of growth factor pathways such as EGF, NGF, MET, Wnt/β-catenin, and IGF1R. The mesenchymal (M) subtype displays activation of pathways involved in epithelial–mesenchymal transition (EMT), cellular differentiation, and growth pathways. Luminal androgen receptor (LAR) tumours are characterised by high expression of androgen receptor (AR) and downstream AR targets, and enrichment of pathways involved in steroid synthesis, porphyrin metabolism, and androgen/oestrogen metabolism (18).

A retrospective analysis from the validation cohort of the TNBC subtype classification presented by Masuda et al. showed that the likelihood of pCR with NACT was subtype dependent. BL1 had the highest pCR rate; BL2 and LAR had the lowest. TNBC subtypes demonstrated improved pCR predictions compared to intrinsic subtypes (basal-like vs. non-basal) (30). In a retrospective analysis of clinically annotated microarray datasets of BC patients, TNBC type-4 subtyping was not associated with significant differences in pCR in the TNBC subgroup. However, the overall incidence of pCR for the subtypes demonstrates trends similar to those observed in previous studies. BL1 displayed the greatest pCR rate (41%), and LAR and BL2 displayed the lowest (29 and 18%, respectively). BL1 patients had significantly higher pCR rates compared with other subtypes (49% vs. 31%, p = 0.04) (29). Santonja et al. explored the performance of Lehmann subtypes and their association with pCR in 125 TNBC patients treated with neoadjuvant anthracyclines and/or taxanes with and without carboplatin, and their results agreed with previous reports (31). The pCR rate for carboplatin containing regimens was highest for BL1 tumours (80% vs 23%, p = 0.027). LAR tumours had the lowest pCR rate for all treatments (14.3% vs 42.7%, p = 0.045).

Burstein and colleagues applied non-negative matrix factorisation clustering to identify four distinct TNBC subtypes characterised by key molecular features and prognosis: LAR, mesenchymal, basal-like immunosuppressed (BLIS), and basal-like immune-activated (BLIA). BLIS and BLIA had the best and worst clinical outcomes, respectively. LAR and mesenchymal subtypes revealed significant overlap with Lehmman’s classification. Burstein’s subtypes based on immune signalling (BLIA, BLIS) revealed a combination of BL1 and BL2 subtypes (19).

Liu et al. developed a classification system based on the transcriptome profiles of both messenger RNAs and long non-coding RNAs to divide TNBC into four distinct clusters. Cluster A: immunomodulatory subtype, Cluster B: luminal androgen receptor subtype (LAR), Cluster C: mesenchymal-like subtype, and Cluster D: basal-like and immune-suppressed (BLIS) subtype. No significant difference in prognosis was found between the four subtypes. Tumours classified as the BLIS subtype experienced poorer relapse-free survival (RFS) compared with all other subtypes (20, 32). Further classification of BLIS tumours based on their homologous recombination deficiency (HRD) status (33) showed that high-HRD BLIS TNBCs and low-HRD BLIS TNBCs exhibited distinctive genomic characteristics and prognoses. Patients with tumours defined as low-HRD had a worse prognosis than those in the high-HRD subgroup (5-year RFS of 73 and 95%, respectively, p = 0.002) (32).

Combining GE and DNA copy number analysis within the METABRIC dataset further expanded the taxonomy of breast cancer (22). Eleven Integrative Clusters (IntClust) with distinctive copy number profiles and clinical outcomes were identified. TNBCs are most frequently classified as IntClust 4ER- or IntClust 10. Rueda et al. showed that patients with tumours classified as IntClust 10 (n = 222) have a low probability of late relapse (five years after diagnosis), while those classified as IntClust4ER- (n = 73) show a persistent and increasing risk of relapse or cancer-related death after 5 years. Classification by immunohistochemistry or intrinsic subtypes did not show this risk (21). The predictive value of IntClust to define response to NACT is yet to be fully established.

Prado-Vasquez et al. developed a probabilistic graphical model to classify the cellular components of tumours into four groups based on the ‘stem cell hypothesis,’ defined based on the grade of development of the cells from which they derived luminal (LAR), basal, claudin-high (CLDN-high), and claudin-low (CLDN-low). The sparse k-means method was used to define high or low immune activity and to classify the tumour as immune metanode positive or negative. Immune metanode activity was prognostic overall, and particularly in the luminal group defined by the cellular classification and TNBC type4-LAR (23).

Combining molecular knowledge with patient management is an increasingly accepted technique across tumour types. In early TNBC, lack of reproducibility and the absence of a unified approach have led to the continuous use of unselected clinical strategies that remain insufficient. Stable commonalities among the classification methods of molecular subtyping in TNBC suggest the presence of clear biological groups suitable for personalised therapeutic interventions. For instance, luminal-like and mesenchymal tumours are consistently identified across the methods with decent overlap and reproducible outcome data. Moreover, most methods include a measurement of the interaction between tumours and the immune response, highlighting the importance of considering this element as a key component of the TNBC taxonomy. Overall, these efforts provide the basis for understand how the molecular complexity of TNBC influences outcomes. Considering the treatment response due to dynamic network interaction, rather than focusing on individual static components, is likely to have more predictive power. But even with reproducible and reliable classification, delivering this in a clinical timeframe suitable for neoadjuvant therapy decision-making remains a challenge.

Anthracyclines target cell proliferation pathways by interacting with DNA gyrase and leading to DNA double-strand breaks (DSBs). The ABC trials proved that the addition of an anthracycline to taxane and cyclophosphamide improved patient outcomes, with the greatest benefit in high-risk patients, those with lymph node involvement or hormone-negative disease (37). More recently, a large meta-analysis by Braybrooke et al. found an 18% reduction in the 10-year recurrence risk with the addition of anthracycline to taxane chemotherapy, as compared to taxane alone (38). There are multiple anthracycline-taxane-based regimens now in use, with evidence to support one “optimal” standard of care regimen for TNBC lacking (39).

Anthracycline-free chemotherapy regimens are considered when cardiotoxicity is a concern, and routine use of such regimens for treatment de-escalation is an area of increasing interest (40) (see Table 3 ). Evidence regarding efficacy as a standard treatment for TNBC is conflicting, although a recent meta-analysis has established anthracycline-free chemotherapy to be acceptable for lower risk, early-stage HER2-negative BC (39).

Taxanes inhibit cell division by stabilising microtubules, preventing depolymerisation, spindle formation, and progression through the cell cycle. Paclitaxel and docetaxel are regularly used to treat early-stage TNBC. An EBCTCG meta-analysis showed that the addition of taxane to anthracycline resulted in a proportional reduction in mortality rates of 15–20% (36). The European Cooperative Trial in Operable Breast Cancer (ECTO) also demonstrated significant improvements in RFS and distant RFS (45). Although this evidence is not unique to TNBC, these studies provide the strongest evidence to support taxane use in this cohort. BL1 and BL2 tumours appear to derive an increased benefit from this drug class (46).

There are several novel alternatives to traditional taxanes under investigation. Nab-paclitaxel is a solvent-free albumin-bound nanoparticle formulation of paclitaxel. It potentially enables higher intra-tumoural taxane concentrations, better efficacy, and improved tolerability. The GeparSepto (47) and ETNA trials (48) showed conflicting results with a significant difference in pCR rates seen only in GeparSepto ( Table S2 ), which may reflect the relative dose intensities used.

Epothilones are promising alternatives to taxanes in development. These novel potent microtubule stabilisers can bypass common resistance mechanisms seen with taxanes, such as drug efflux pumps and β-tubulin. In the early setting, the phase 3 TITAN trial has shown similar efficacy and reduced rates of peripheral neuropathy, dose modifications, and discontinuation with Ixabepilone comparised with paclitaxel (49).

The clinical activity of platinum agents has been significantly associated with a DDR vulnerability in both sporadic and gBRCA-associated TNBC. Carboplatin is increasingly used in neoadjuvant regimens, improving both pCR and long-term outcomes (50). Please see section DNA Damage Response (DDR).

Capecitabine is an oral prodrug of the antimetabolite 5-fluorouracil. Capecitabine is not currently recommended in clinical guidelines for the neoadjuvant or adjuvant treatment of TNBC, though it is selectively used as a post-neoadjuvant treatment for residual TNBC. Insights for use in the adjuvant setting are accumulating ( Table 4 ), but in most cases, studies have not incorporated the molecular features of the TNBC cohort into the planned analysis for response assessment. The recent phase 3 CBCSG-010 trial for unselected patients with TNBC with concomitant use of capecitabine 1,000 mg/m² and standard anthracycline-taxane adjuvant chemotherapy (ACT) established a significant 5-year disease free survival (DFS) benefit (53). This is supported by the FINXX trial (54) and the Ye et al. meta-analysis, which demonstrated improved DFS and OS with a tolerable increase in toxicity (56).

Chemotherapy can be delivered in an adjuvant or neoadjuvant setting with no significant difference in long-term outcomes, as illustrated by the NSABP B-18, EORTC 10902, and IBBGS trials (62–64). More recently, an EBCTCG meta-analysis demonstrated no significant difference in distant recurrence or death between NACT and ACT but a more frequent local recurrence rate (65). A TNBC-specific meta-analysis suggested that NACT is associated with a comparable DFS but worse OS than ACT (66), perhaps explained by patients with higher disease burden being more likely to receive NACT. In this meta-analysis, patients who achieved pCR had superior OS and DFS compared to those treated with ACT. This evidence does not support the suggestion that NACT promotes cancer cell dissemination (67).

Advantages of NACT include downstaging tumours, resulting in increased rates of breast-conserving surgery and associated improved cosmesis and reductions in postoperative lymphoedema. Additionally, it allows the assessment of treatment response, provides valuable prognostic information (68), guide choice of post-surgical treatment and allows for ineffective treatment to be ceased to avoid unnecessary toxicity. NACT also provides an ideal platform for translational research, assessment of biomarkers, and genetic testing (69).

The same anthracycline/taxane-based regimens are typically used in NACT and ACT. Whether the scheduling of these combinations has any effect on efficacy has been the subject of extensive research. It is been shown that using taxanes and anthracyclines sequentially increases efficacy and decreases toxicity (70). Some evidence suggest that administration of taxane chemotherapy before anthracyclines is associated with improved pCR rates (71).

There has been an increasing interest in personalising the treatment timetables to take patient and tumour characteristics into account. Dose-dense NACT is now a widely accepted treatment strategy for high-risk TNBC in order to prevent cancer cell repopulation (72). It has been consistently shown to improve the rates of pCR, breast-conserving surgery, and recurrence in hormone-low BC (73, 74). Although this regimen has not translated into a significant survival benefit (74), this approach should be considered in selected patients with a high disease burden. Dose-dense ACT improves DFS and OS rates in patients with low hormone receptor levels, although this is accompanied by increased toxicity and patients need to be selected carefully (75).

At the other end of the spectrum, metronomic chemotherapy is given at a minimum biologically effective dose either continuously or with minimal extended breaks from treatment to reduce severe toxicity. It is thought to have angiogenic, stroma-targeting, and immunostimulatory effects (76). It has been investigated as a single approach and is being used in combination to intensify standard chemotherapy. It may have a role as a maintenance therapy for high-risk patients or for use by patients who would not otherwise be able to tolerate the adverse effects of standard treatments. The SYSUCC-001 study showed significant improvement in 5-year DFS with 1 year of maintenance capecitabine (55). The IBCSG 22-00 trial confirmed a 7.9% reduction in the absolute risk of relapse in patients with node-positive TNBC (77) after 1 year of low-dose capecitabine and methotrexate maintenance treatment, although no improvement in DFS was observed.

Tumours evade detection and eradication by the immune system through the dysregulation of pathways controlled by immune checkpoints. Immunotherapy harnesses the immune system of the patient to target malignant cells using immune checkpoint inhibitors (ICI), chimeric antigen receptor T cells, or cancer vaccines. ICIs release the immune system from tumour-induced inhibitory signals, allowing an effective anti-tumour response. They include monoclonal antibodies (mAbs) against cytotoxic T lymphocyte-associated antigen-4 (CTLA-4), programmed cell death-1 (PD-1), and programmed cell death ligand-1 (PD-L1).

In unselected TNBC, response to PIK3CA inhibitors remains low. The BELLE-4 study evaluated the efficacy of buparlisib in the locally advanced setting for patients with HER2-negative BC along with paclitaxel versus placebo and observed no benefit from PIK3CA inhibition. Worse outcomes were observed in the TNBC cohort treated with the PIK3CA inhibitor, and the lack of benefit was independent of PIK3CA mutation or PTEN loss by immunohistochemistry (170). Shorter treatment duration in the buparlisib arm due to adverse events and longer progression-free survival (PFS) in the placebo arm than anticipated are possible explanations for the worse outcomes in this subgroup. The global lack of activity is possibly due to inadequate patient selection and the absence of an accurate biomarker. Parallel pathway activation could also explain a resistance mechanism that requires addressing.

Ipatasertib was reviewed in the neoadjuvant setting along with paclitaxel for TNBC patients in the FAIRLANE trial (171). Adding ipatasertib did not significantly increase the pCR rate compared with paclitaxel alone, and this effect was independent of PIK3CA/AKT1/PTEN or PTEN low status. A complete clinical response was absent in the placebo-treated group in patients with tumours defined as the LAR subtype, but was observed in 50% of those treated with ipatasertib. This difference was not evident in pCR rates. Elevated immune scores were more strongly associated with improved outcomes in paclitaxel-treated compared with ipatasertib-treated patients, highlighting the key interaction with the immune system. All ipatasertib-treated patients with low immune scores and complete clinical response had PIK3CA/AKT1/PTEN-altered tumours. MK2206 has been trialled in the neoadjuvant setting in the I-SPY2 trial for stage 2–3 BC of any subtype (172). Patients received paclitaxel chemotherapy with or without MK2206, then AC. pCR for the TNBC group was 40.2% with MK2206 vs. 22.4% without. Following assessment of biomarkers in the AKT pathway in the TNBC subgroup, higher levels of phosphorylated AKT and its substrates were paradoxically associated with a reduced response to MK-2206.

Everolimus has been reviewed in the neoadjuvant setting for patients with TNBC along with cisplatin and paclitaxel (173), and along with docetaxel, 5-fluorouracil, epirubicin, and cyclophosphamide (174). No improvement in the response rate has been demonstrated.

The exact contribution of drugs targeting the PIK3CA/AKT1/PTEN pathway in early TNBC has not yet been defined. The complexity of the immune microenvironment and parallel molecular alterations can obscure accurate estimation of clinical benefit if they are not in some way accounted for. It is important to try these therapies in a way that reduces these confounders and separates the TNBC subtypes to determine their individual responses. Current approaches include combining alpelisib with nab-paclitaxel in the neoadjuvant setting (NCT04216472) for anthracycline refractory TNBC with PIK3CA or PTEN alterations, with exploratory objectives to assess biomarkers of response and resistance to alpelisib and nab-paclitaxel combination.

Table S5 summarises ongoing trials that target this pathway in early TNBC.

Approximately 35% of TNBCs as defined by immunohistochemistry could be classified as HER2-low (181). Somatic ERBB2 mutations occur in approximately 3% of TNBC (6), and a subset of TNBC tumours are classified as HER2-enriched by gene expression. This biological heterogeneity has expanded therapeutic opportunities in this population of patients. In an exploratory analysis of a cohort of the I-SPY2 trial, activation of HER2-EGFR was identified as a positive predictor of pCR in 49 TNBC patients treated with a pan-HER inhibitor (182). A significant correlation between the response to HER2 inhibition and HER2 pathway activation has been demonstrated in TNBC cell lines (183).

Neratinib has been investigated in the neoadjuvant setting for high-risk clinical stage II or III BC. The pCR rate overall in the I-SPY 2 trial was 37.5% in the neratinib arm, and among patients demonstrating the phosphorylation of HER2 or EGFR (i.e., biomarker-positive for EGFR Y1173 or ERBB2 Y1248), it rose to 63% (184). Encouraging results in the HER2-low–expressing refractory BC setting with Trastuzumab Deruxtecan (OR 37%) (185) and Trastuzumab Duocarmazine (OR 40%) (186) now require translation into the early setting. These trials illustrate the importance of identifying patients categorised as TNBC who are more accurately defined as HER2 low ( Table S5 ).

VEGF promotes angiogenesis, invasion, and increases vascular permeability and is an essential element in TNBC formation, progression, and metastasis. VEGF-A expression is higher in TNBC compared with other BC subtypes (187), and enhanced angiogenic potential is associated with poor prognosis in BC (188). Targeting of VEGF has been extensively tested in TNBC, but no clear predictive biomarkers of treatment response have been identified.

Trials targeting VEGF in the neoadjuvant TNBC setting have shown disappointing results to date, with no difference in DFS or OS. The addition of bevacizumab significantly increased the rate of pCR among patients with Her2-negative disease in some studies (103, 189–191). The ARTemis and GeparQuinto trials reported increased benefits primarily in the TNBC subgroup. In the adjuvant setting, the BEATRICE trial added bevacizumab to anthracycline and/or taxane-based chemotherapy (192), and no difference in iDFS or OS between treatment groups was found. The underlying reason for the lack of treatment effects with these drugs is poorly understood. It is possible that a fundamental flaw in either the drug or the signalling pathway is being overlooked. Attempts to overcome drug resistance using novel agents and combinations are ongoing ( Table S5 ).

The fibroblast growth factor receptor family includes FGFR1–4. Signalling through this pathway regulates cell survival, proliferation, and differentiation. Genes that encode for these receptors are amplified in ~10% of BC (24). Although FGFR1 is the most frequent genomic alteration in all subtypes of BC, amplification and overexpression of FGFR2 are more frequently observed among TNBCs (~4%). Basal BC with elevated MET and FGFR1 signatures is associated with poor relapse-free survival (193). The interplay between MET and FGFR regulates cancer stem cells in mesenchymal subtypes (194).

Trial data in this setting is limited to a few studies that do not select for TNBC but in which some response to this target has been seen. It seems likely that the correct biomarker has not yet been identified. Ongoing trials for this target in the neoadjuvant setting include a window of opportunity trial combining lenvatinib and pembrolizumab (NCT04427293).

EGFR dysregulation is frequently reported in TNBC (195) and enrichment for this pathway signalling is predominantly observed in BL2 tumours (196). In contrast to EGFR mutations, EGFR amplification is a relatively frequent event (11% vs 23%, respectively) (24, 197) and is considered an independent prognostic factor for poor disease-free survival (198). Several attempts to target this pathway with tyrosine kinase inhibitors and mAbs in the context of mTNBC have been pursued without success. A limited number of trials have used these therapies in the early setting.

Trials are underway in the locallyCetuximab has been trialled along with neoadjuvant docetaxel in a pilot phase two study, including stage II–IIIA TNBC (199). The pCR rate was 24% [95% CI: 7.3–40.7] and the pre-therapy ratio between CD8+ and FOXP3+ TILs equal or higher than 2.75 was predictive of pCR (43% versus 0%). In addition, panitumumab and the EGFR/HER2 inhibitor lapatinib failed to demonstrate additional benefit in the advanced setting independent of EGFR activation (200, 201). The paucity of accurate biomarkers predictive of sensitive patients has led to unsatisfactory outcomes and limited clinical utility despite increasing evidence for EGFR as the driver of tumorigenesis in some TNBC.

Table S5 summarises ongoing trials that target these pathways in early TNBC.