Unraveling complexity and leveraging opportunities in uncommon breast cancer subtypes

The World Health Organization (WHO) recognizes about 20 special type breast carcinomas which are distinct from invasive breast carcinoma of no special type (IBC-NST; also known as invasive ductal carcinoma), the most common histologic form of breast cancer¹. In aggregate, these special types account for ~20% of all breast cancers¹. Invasive lobular carcinoma (ILC) is the most common special type, accounting for 10–15% of all breast cancers. Most other special type cancers are rare, each accounting for <1% to 5% of breast cancers. These uncommon types of breast cancer are distinct at the histologic, biomarker, transcriptomic, and genomic levels, and some have distinct clinical behavior. Some, including ILC, adenoid cystic, secretory, mucoepidermoid, and tall cell carcinoma with reversed polarity (TCCRP) exhibit disease-defining genetic alterations^2,3,4, whereas others do not, or they have yet to be identified. In addition, some special type cancers have prognostic implications that may inform therapy. For example, tubular, invasive cribriform, and mucinous carcinomas have an excellent prognosis, whereas some metaplastic carcinoma subtypes display an aggressive clinical course and therapeutic resistance.

Given their rarity, much of what is known about special type cancers is based on small, single-institution case series and case reports. Studies employing databases such as the National Cancer Institute’s Surveillance, Epidemiology and End Results (SEER) and the National Cancer Database (NCDB) provide larger numbers of cases but may be limited by the lack of central pathology review^5,6. Larger studies, including multi-institutional collaborations of these less common types of breast cancer are crucial, since some may exhibit molecular alterations also relevant to subsets of the more common types of breast cancer. In addition, uncovering previously unidentified genomic, transcriptomic or proteomic alterations/features may provide potential therapeutic targets for precision medicine approaches for patients affected by these special cancer types.

In this article, we will review the spectrum of uncommon types of breast cancer including defining features, germline predisposition, emerging diagnostic tools and biomarkers, the role of artificial intelligence (AI) in diagnosis, and new therapeutic opportunities. In addition, we will discuss rare clinical presentations of breast cancer.

Triple negative breast cancer (TNBC) encompasses an array of histologic entities ranging from low- to high-grade carcinomas with diverse clinical outcomes⁷. All are defined by the lack of expression of estrogen receptor (ER) and progesterone receptor (PR) and the lack of HER2 overexpression or gene amplification. A significant proportion of TNBCs, however, fall under the recently recognized category of HER2-low, i.e., expression of HER2 with an immunohistochemistry (IHC) score of 1+ or 2+ without ERBB2 gene amplification⁸. A subset of low-grade special type TNBCs, including most within the salivary gland-like tumor group, are characterized by pathognomonic genetic alterations and represent genotypic-phenotypic correlations. In contrast, other TNBC subtypes lack disease-defining genetic alterations but exhibit distinctive genomic features (Table 1). The spectrum of TNBC special tumor types includes metaplastic carcinoma, salivary gland-like tumors, secretory carcinoma, apocrine carcinoma and TCCRP though some of these tumors may rarely show HER2-positivity or ER-positivity.

Metaplastic carcinomas constitute a heterogenous group of breast cancers that show differentiation of neoplastic epithelium into squamous and/or mesenchymal elements (Fig. 1A)⁹. They represent <1% of all breast cancers. Clinical features are similar to other types of breast cancer, especially ER-negative subtypes, with a lower frequency of nodal metastasis compared to IBC-NST⁹. Heterogeneity is reflected in morphology, with tumors exhibiting a single line of metaplastic differentiation, multiple metaplastic elements, or metaplastic elements combined with areas of IBC-NST⁹. Metaplastic carcinomas also have variable histologic grade. Examples range from low-grade adenosquamous carcinomas and low-grade fibromatosis-like metaplastic carcinomas at the indolent end of the spectrum, to high-grade tumors with heterologous mesenchymal elements at the aggressive end of the spectrum. Full classification is available in the 5^th Ed WHO classification¹. Most metaplastic carcinomas are triple negative and express high molecular weight cytokeratins and EGFR¹⁰. They have complex genomic landscapes with alterations in TP53, and genes of the PI3K, MAPK and WNT pathways¹¹ (Table 1). Different subtypes have differing genomic profiles, e.g., TP53 is less frequent in low grade variants¹¹. Notably, TERT somatic genetic alterations have been reported in a subset of metaplastic carcinomas¹² (Table 1). Interestingly, gene expression analyses revealed that metaplastic squamous carcinomas from The Cancer Genome Atlas (TCGA) clustered with squamous carcinomas stemming from the head and neck, cervix, lung and esophagus. In contrast, metaplastic carcinomas with a predominant mesenchymal component clustered with non-mammary sarcomas, highlighting the molecular heterogeneity of these tumors¹³. Because of the rarity of metaplastic carcinomas, data from clinical trials is scarce. Furthermore, there is no robust prognostic or predictive data for therapeutic response. Ten-year survival is ~50%¹⁰.

Representative micrographs of A metaplastic squamous cell carcinoma, B apocrine carcinoma, C classic adenoid cystic carcinoma D and corresponding MYB RNA in situ hybridization, E solid basaloid adenoid cystic carcinoma, F acinic cell carcinoma G and corresponding alpha-1 antichymotrypsin expression, H secretory carcinoma, I tall cell carcinoma with reversed polarity, J and corresponding IDH2 R172 expression, K adenomyoepithelioma, L classic invasive lobular carcinoma (ILC), M and corresponding E-cadherin loss of expression, N solid ILC, O alveolar ILC, P pleomorphic ILC, Q mucinous carcinoma, R neuroendocrine tumor, S tubular carcinoma and T invasive micropapillary carcinoma.

Full size image

Apocrine carcinomas represent 0.3% to 4% of all breast cancers. The range in reported prevalence reflects the lack of uniform diagnostic criteria¹⁴. Apocrine carcinomas are composed of large cells with abundant granular eosinophilic cytoplasm, and enlarged nuclei with prominent nucleoli, recapitulating apocrine sweat glands¹ (Fig. 1B). The cytoplasm can occasionally be foamy to pale (resembling histiocytes). The cells often grow in solid sheets, but a glandular pattern with prominent apocrine snouts may also be encountered. Notably, apocrine morphology may be encountered in other breast cancers, including ILC. Apocrine carcinomas are usually triple negative or HER2-positive (30-57%) and show strong androgen receptor (AR) expression. Notably, ~40% of apocrine carcinomas are HER2-low, and most triple-negative apocrine carcinomas (~70%) fall into this category^15,16. Apocrine carcinomas frequently show alterations in genes of the PI3K pathway (Table 1) and overlap genetically with the luminal androgen receptor (LAR) and molecular apocrine subgroups¹⁴. HER2-positive apocrine carcinomas have been associated with a poor prognosis, whereas triple negative cases tend to have a more favorable prognosis relative to other TNBCs, despite lower rates of pathologic complete response (pCR) following neoadjuvant chemotherapy (NAC)^15,17,18.

Adenoid cystic carcinomas (AdCC) comprise only 0.1% of all invasive breast cancers. These tumors are rare salivary gland-like cancers composed of two cell types, luminal epithelium and myoepithelium¹. In its classic form, AdCC is associated with basophilic matrix, lacking significant atypia and mitotic activity. AdCC often shows a mixture of growth patterns, including cribriform, tubular, and solid (Fig. 1C). The solid pattern should constitute only a minority of the tumor, otherwise consideration of a solid/basaloid variant is warranted. Immunohistochemically, the tumor cells are positive for epithelial and myoepithelial markers and CD117 (CKIT)¹⁹. Classic AdCCs commonly harbor MYB-NFIB gene fusions, MYBL1 rearrangements, or MYB amplification²⁰ (Fig. 1D, and Table 1), constituting a convergent phenotype (Fig. 2A) Despite being of triple negative phenotype, classic AdCC is a slow growing tumor with an excellent prognosis.

A MYB-NFIB fusion gene, MYBL1 rearrangement or MYB amplification underpinning classic adenoid cystic carcinoma. B CDH1 mutations, promoter methylation, structural variants and homozygous deletions, as well as genetic alterations in other cell-cell adhesion, such as CTNNA1, CTNND1 or AXIN2 underpinning invasive lobular carcinoma.

Full size image

The solid basaloid variant of adenoid cystic carcinoma (SB-AdCC) is a rare neoplasm that represents <0.1% of all breast cancers. SB-AdCC is characterized by basaloid cells growing in nests, often with marked cytologic atypia, high mitotic count and necrosis¹ (Fig. 1E). While the original definition of this tumor requires >90% solid/basaloid histology²¹, tumors with areas of classic AdCC and/or high-grade transformation have been described. Basaloid cells often grow in solid nodules or sheets with occasional intercalated duct formation. The characteristic myxohyaline stroma may be a useful morphologic clue to distinguish SB-AdCC from other similar-appearing basaloid TNBCs²². Unlike classic AdCC, the frequency of the canonical MYB-NFIB rearrangement is low, despite frequent expression of MYB protein by IHC^22,23,24. Next-generation sequencing of SB-AdCC has revealed recurrent alterations in NOTCH family genes, CREBBP and chromatin remodelers, such as KMT2C/D^22,24,25 (Table 1). SB-AdCC can show an aggressive clinical course, with distant metastasis^{22,23,24,25,26}. When compared to classic AdCC, these tumors have an overall lower recurrence-free survival (46.5 months compared to 151.8 months)²².

Acinic cell carcinoma is a very rare breast cancer subtype, comprising ~0.01% of invasive breast cancers and ~3% of breast salivary gland-like neoplasms^27,28. These tumors are characterized by haphazard infiltration of fibrous stroma and adipose tissue by small glands with eosinophilic intraluminal secretions and solid cell nests of varying size. The tumor cell cytoplasm can be clear or amphophilic or have basophilic or eosinophilic granules. The eosinophilic granules may be fine or coarse (Paneth cell-like; Fig. 1F). Acinic cell carcinomas are typically triple negative and the tumor cells show expression of S100 protein, epithelial membrane antigen, lysozyme and alpha-1 antichymotrypsin²⁹ (Fig. 1G). Unlike other salivary gland-like breast cancers, acinic cell carcinomas are genetically distinct from their salivary gland counterparts. In particular, salivary gland acinic cell carcinomas show a recurrent t(4;9)(q13;q31) rearrangement that results in constitutive upregulation of the nuclear transcription factor NR4A3 which can be detected by IHC. In contrast, breast acinic cell carcinomas are consistently negative for NR4A3 protein expression and rearrangements³⁰. In addition, unlike salivary gland acinic cell carcinomas that show overexpression of DOG1, breast acinic cell carcinomas are DOG1-negative³⁰. Further, breast acinic cell carcinomas share molecular features with conventional TNBCs including frequent TP53 mutations and high levels of genomic instability³¹ (Table 1), whereas salivary gland acinic cell carcinomas do not. Finally, breast acinic cell carcinomas have morphologic and genetic overlap with microglandular adenosis (MGA) and atypical MGA. Thus, unlike other salivary gland-like carcinomas of the breast, the molecular underpinnings of salivary gland and breast acinic cell carcinomas are different and these likely represent distinct entities. In addition, the similarities between breast acinic cell carcinomas and MGA suggests a relationship between these two entities³⁰.

Secretory carcinomas are rare, representing <0.05% of all breast cancers¹. While these tumors may be seen in children and adolescents (and were formerly known as “juvenile carcinomas”), they are more common in adults. Secretory carcinoma is an invasive breast cancer characterized by extracellular eosinophilic secretions and intracytoplasmic secretory vacuoles (Fig. 1H). They can show microcystic, tubular and/or papillary architectural patterns¹. The tumor cells exhibit a faintly eosinophilic to foamy cytoplasm and low to intermediate nuclear grade. Prominent Alcian blue or periodic acid-Schiff-positive secretions are present both within the extracellular compartment and tumor cell cytoplasm. An in situ component with similar morphologic features can also be observed. Secretory carcinomas belong to the TRK-fusion cancer family, and are characterized by recurrent ETV6-NTRK3 fusion gene^32,33 (Table 1). While in situ hybridization for ETV6 gene rearrangement has been traditionally used to detect this abnormality, IHC analysis using a pan-TRK antibody can also be used as a reliable surrogate for this alteration, with secretory carcinomas showing strong nuclear staining³⁴. Despite the occasional occurrence of lymph node and/or distant metastases, these tumors typically have indolent clinical behavior^35,36,37,38. However, rare late recurrences may be seen.

Tall cell carcinoma with reversed polarity (TCCRP) is included by the WHO among rare and salivary-gland type breast cancers¹, with less than 100 cases reported. They usually occur in older women and have an indolent clinical course. These tumors are characterized by rounded nests of tall columnar epithelial cells arranged in solid and solid papillary patterns and present in a fibrotic stroma (Fig. 1I). Some of the nests with a solid papillary pattern contain foamy histiocytes within the fibrovascular cores. The hallmark feature of these tumors is the presence of nuclei at the apical rather than the basal poles of the columnar epithelial cells (Fig. 1I). TCCRPs express both low and high-molecular weight cytokeratins as well as calretinin. IDH2 R172 hotspot mutations have been identified in about 85% of cases studied and represent the defining genetic abnormality of these tumors^4,39. Concomitant PIK3CA or PIK3R1 mutations have been identified in about 70% of cases^4,39 (Table 1). Immunostaining for the mutant IDH2 R172 protein can be used in lieu of sequencing to confirm the diagnosis⁴⁰ (Fig. 1J). Striated duct adenoma of the salivary glands has some of the morphologic features and the same IDH2 R172 hotspot mutations as TCCRP, and likely represents its salivary gland counterpart⁴¹.

Adenomyoepithelioma (AME) is a rare neoplasm that comprises <1% of breast tumors. AME predominantly affects older women, and usually presents as a solitary palpable or mammographically-detected mass, occasionally associated with serous nipple discharge. AME is a biphasic neoplasm characterized by small glandular spaces lined by luminal ductal cells surrounded by a proliferation of variably enlarged abluminal myoepithelial cells^1,42 (Fig. 1K). AME comprises a spectrum of lesions ranging from benign to malignant tumors that overlap morphologically and immunophenotypically with high-grade metaplastic carcinoma. Based on the histological features, Rakha et al. ⁴² proposed a classification of AME into benign, atypical and malignant tumors, and the malignant AMEs (M-AME) were further classified into M-AME in situ, M-AME invasive and AME with invasive carcinoma. Benign or classical AMEs are associated with increased risk of local recurrence but no metastatic potential. Complete excision is the treatment. Atypical AME is recognized as a lesion of uncertain malignant potential with limited metastatic capability; it displays a clinical behavior intermediate between benign and malignant AME^43,44. Complete excision with follow-up is recommended for these lesions. Metastases have been reported in 16–32% of M-AME but the metastases appear to be restricted to tumors >2 cm⁴². While ER-positive AMEs frequently harbor mutations affecting PIK3CA or AKT1, ER-negative AMEs are underpinned by mutations affecting the HRAS Q61 hotspot, which frequently coexist with PIK3CA or PIK3R1 mutations⁴⁵ (Table 1).

Other exceedingly rare TNBCs are mucoepidermoid carcinoma (MEC) and polymorphous adenocarcinoma (PAC)⁴⁶. Although most breast MECs are triple negative, a subset of them are ER-low or HER2-positive⁴⁷. They are composed of different cell populations, including mucinous, intermediate and squamous cells, and are histologically similar to MECs arising in other anatomic locations, such as salivary glands, lung and cervix⁴⁶. MAML2 rearrangements, the hallmark genetic feature of MECs arising in other organs, have been reported in breast MECs^48,49. In contrast to common forms of TNBC, breast MECs display a low mutation burden, and a lack of copy number alterations and of TP53 mutations^47,49. Notably, the CRTC1–MAML2 fusion gene has been shown to induce signaling via the amphiregulin-EGFR axis⁵⁰, and breast MECs lacking MAML2 rearrangements, display overexpression of EGFR and its ligand amphiregulin⁴⁷, suggesting a convergent phenotype. Breast PACs are composed of monotonous tumor cells arranged in a spectrum of architectural patterns, such as solid nests, cords and single cells, morphologically similar to their counterparts in the salivary glands⁴⁶. PACs of the salivary glands are characterized by recurrent PRKD1 pGlu710Asp hotspot mutations, as well as rearrangements in PRDK1, PRKD2 or PRKD3^51,52. Whether breast polymorphous adenocarcinomas harbor the same genetic alterations, remains to be determined.

An array of special histologic subtypes of breast cancer exists under the umbrella of ER-positive breast cancer. Despite their distinctive phenotype, aside from ILC which has a strong genotypic-phenotypic correlation, entities within this group typically lack known pathognomonic genetic alterations. Nevertheless, they exhibit genetic characteristics that set them apart from ER-positive IBC-NST.

Invasive lobular carcinoma (ILC) is the most frequent special histologic subtype of breast cancer and represents 10–15% of all breast cancers¹ (reviewed in refs. ^53,54,55). These tumors are characterized by a non-cohesive growth pattern (Fig. 1L), a consequence of deregulation of cell adhesion, primarily due to loss of E-cadherin (CDH1), the hallmark molecular alteration of ILC⁵⁶ (Table 1 and Fig. 1M). Although not required for diagnosis according to the WHO¹, supporting IHC testing for loss of E-cadherin is used by many pathologists and it has been shown to increase inter-pathologists’ concordance for ILC diagnosis⁵⁷. Although E-cadherin loss of expression in ILC is due to loss-of-function CDH1 mutations in most cases, it can also be the result of promoter methylation, homozygous deletions or structural variants^55,58,59 (Fig. 2B). Moreover, inactivating genetic alterations affecting other cell-cell adhesion genes have also been reported in ILC^60,61 (Fig. 2B). In a subset of ILC cases, especially those containing tubular elements there is a rescue of the adherens junction as a result of E-cadherin switching, and specifically upregulation of P-cadherin⁶².

Lobular neoplasia refers to intra-acinar proliferation of dyscohesive epithelial cells in mammary lobules and includes classical lobular carcinoma in situ (LCIS), as well as LCIS histologic variants, such as pleomorphic LCIS and florid LCIS (reviewed in refs. ^63,64). These lesions are considered non-obligate precursors of ILC and already exhibit loss of CDH1, highlighting this as an early genetic event in the development of ILC. Classic LCIS is associated with a 7- to 10-fold increase in breast cancer risk compared to women in the reference population, with a long-term absolute risk of 1–2% per year. Of note, pleomorphic LCIS is more genetically advanced and shows a higher rate of association with invasive carcinoma. While classical LCIS is managed with surveillance, surgical excision is often recommended for its more aggressive subtypes.

ILC includes not only classic, but also several non-classic variants which make up a large subset of cases, and are characterized by distinctive architectural patterns, such as solid (Fig. 1N), alveolar (Fig. 1O), and trabecular, as well as by cytologic features, such as pleomorphic (Fig. 1P), histiocytoid and signet ring cell morphology. The majority of classic ILCs are of histologic grades 1 and 2 while the majority of grade 3 ILC comprise solid and pleomorphic variants⁶⁵. ILC with extracellular mucin (ILCEM) also present as higher-grade tumors, with worse outcomes⁶⁶. While outcome for patients with pleomorphic ILC is not worse after adjusting for other clinical variables^67,68, specific molecular features such as enrichment of ERBB2 mutations point towards differences in biology which deserve additional attention. ERBB2/ERBB3 mutations in pleomorphic LCIS were clonally related to synchronous invasive lesions providing further evidence that pleomorphic ILC can be direct precursor lesions for ILC⁶⁹.

The majority (95%) of ILC are ER-positive and lack ERBB2 amplification. Triple-negative ILCs are rare, constituting ~1–2% of all ILCs. Frequent expression of AR⁷⁰, as well as response to checkpoint blockade⁷¹ has been reported in these tumors. The PI3K/AKT pathway is frequently hyperactivated in ILC^72,73, and differences in the genomic landscape have been uncovered between ER-positive/HER2-negative ILC versus ER-positive/HER2-negative IBC-NST, including an enrichment of ERBB2 mutations, especially in metastatic ILC^55,74.

Despite favorable clinicopathological features, including lower proliferation rates, the long-term outcome of ILC is worse than IBC-NST, in part due to increased rates of late recurrences^75,76,77. As in ER-positive IBC-NST, the most frequent site of metastasis is bone. However, ILC demonstrates increased rates of metastases to unusual sites including the urogenital and gastrointestinal tracts, peritoneum, meninges and the orbit, while rates of metastases to liver and lungs are lower than in ER-positive IBC-NST⁵⁵. Furthermore, there is increasing evidence that ILC show lower response rates to chemotherapy^55,78. Currently, patients with ILC are often managed similarly to patients with IBC-NST. This is evolving, however, and new diagnostic and therapeutic approaches tailored specifically for ILC are emerging, such as various ongoing ILC-specific clinical trials⁵⁵.

Mucinous carcinoma accounts for ~2% of all invasive breast carcinomas, often occur in older women, and usually present as a mass on imaging¹. Mucinous carcinomas are characterized by clusters of tumor cells, with predominantly low to intermediate-grade nuclei, floating in pools of extracellular mucin (Fig. 1Q), and can be subclassified into type A (paucicellular) and type B (hypercellular)¹. These tumors are generally ER-positive/HER2-negative and of luminal A molecular subtype¹. The hypercellular type is typically strongly ER-positive and often expresses neuroendocrine markers. Pure and mixed mucinous carcinomas display >90% and 10–90% of mucinous component, respectively¹. Pure mucinous carcinomas harbor recurrent GATA3, KMT2C and MAP3K1 mutations and display low levels of genetic instability^79,80. They less frequently harbor molecular alterations commonly found in low-grade ER-positive IBC-NSTs, including concurrent chromosome 1q gains and 16q losses and somatic PIK3CA and TP53 mutations⁷⁹, and display aberrant DNA methylation of MUC2⁸⁰ (Table 1). Compared to primary tumors, metastatic mucinous carcinomas have a higher tumor mutational burden and more frequent ESR1 genetic alterations⁷⁴. Pure mucinous carcinomas are associated with a favorable clinical outcome, whereas mixed ones tend to have worse outcomes¹. Notably, breast mucinous carcinomas were found to be transcriptomically similar to mucinous carcinomas arising in other anatomic locations in the TCGA dataset, including gastric, rectal, colon, pancreatic and cervical mucinous adenocarcinomas. Both breast and non-mammary mucinous carcinomas showed an enrichment of genes involved in O-glycan processing and O-linked glycosylation¹³. Furthermore, breast, colorectal and gastric cancers all exhibit MUC2 upregulation controlled by DNA methylation, and the paucity of TP53 and PIK3CA mutations observed in breast mucinous carcinomas, was also found in mucinous carcinomas of other origins⁸¹. These findings underscore the conserved molecular features of these tumors.

Neuroendocrine neoplasms of the breast are rare lesions, accounting for less than 1% of all breast carcinomas. The 2019 WHO classification¹ has adopted the subdivision into neuroendocrine tumors (NET) and neuroendocrine carcinomas (NEC), following a similar, albeit not entirely overlapping, approach to how neuroendocrine tumors are classified in other organs. Both entities must demonstrate morphological and immunohistochemical features of neuroendocrine differentiation. Some special histologic types, such as hypercellular mucinous, lobular and solid-papillary carcinomas as well as IBC-NST may exhibit expression of neuroendocrine markers, but those tumors should be classified according to their histologic type and not as neuroendocrine neoplasms. NETs typically feature densely cellular, solid nests and trabeculae of cells that vary from spindle cells to plasmacytoid cells and polygonal cells separated by delicate fibrovascular stroma (Fig. 1R). NECs include large cell and small cell carcinomas with high mitotic count and a higher degree of nuclear atypia¹. NETs and NECs are graded according to the Nottingham grading system and NETs can be either histologic grade 1 or grade 2, whereas NECs are of histologic grade 3 by definition¹. Expression of neuroendocrine markers such as chromogranin, synaptophysin, and INSM1 is the characteristic feature of neuroendocrine neoplasms¹.

There are no notable or specific differences in presentation compared to other tumor types. Given the neuroendocrine differentiation, however, the clinical history of patients should be carefully considered to rule out possible metastatic lesions in the breast stemming from neuroendocrine carcinomas from other anatomic locations. NETs are typically ER-positive/HER2-negative, whereas NECs can also be of triple-negative phenotype. At the transcriptomic level they are typically luminal A/B^82,83. Previous studies have described a mutational landscape that slightly differs from common forms of luminal carcinomas, with a lower rate of PIK3CA and TP53 mutations, and genetic alterations in genes found to be mutated in neuroendocrine neoplasms of other organs, such as ARID1A, TBX3 and FOXA1, and chromatin remodeling genes^82,83 (Table 1). Indeed, breast NETs show a similar frequency of genetic alterations in chromatin remodelers as lung carcinoids and pancreatic NETs⁸⁴. Nonetheless, the mutational profiles of breast NETs more closely resemble those of IBC-NST than NETs arising in other organs, such as pancreatic NETs⁸⁵.

There are conflicting data on the prognosis of neuroendocrine carcinomas. A recent study assessing prognosis based on the new WHO classification suggests that aside from high-grade small- and large-cell NECs, breast neuroendocrine neoplasms behave like non-neuroendocrine breast carcinomas and should be managed similarly⁸⁶. Due to differences of neuroendocrine neoplasms arising in the breast and those originating in other anatomic location, the classification of breast neuroendocrine neoplasms continues to evolve.

Tubular carcinoma is a low-grade invasive breast carcinoma composed of well-formed tubules with open lumina lined by a single layer of neoplastic epithelial cells¹. Tubular carcinomas constitute ~1.6% of invasive breast carcinomas but are reported to have a higher prevalence in screened populations. These tumors are more likely to occur in older patients, and to be detected incidentally by mammographic screening^87,88. The diagnosis of tubular carcinoma requires >90% of the tumor to be composed of opened tubules, with the exception of an invasive cribriform component. The tubules are small, round to ovoid or angulated and set within a fibrous or fibro-elastotic desmoplastic stroma (Fig. 1S). Most of these tumors are small, with a mean diameter of 12 mm. Large-size tumors are likely to have more solid clusters of cells (IBC-NST component) that are associated with less favorable prognosis⁸⁸. Tubular carcinoma typically exhibits a luminal A phenotype⁸⁸ and is characterized by a low frequency of genomic alterations, with chromosomes 16q loss and 1q gain as the most frequent event⁸⁹. Furthermore, these tumors exhibit a high frequency of PIK3CA mutations (Table 1). The majority of cases present at a relatively early stage (T1 tumors with N0)^88,90. Tubular carcinoma has an excellent long-term outcome, significantly better than stage-matched grade 1 IBC-NST⁸⁸, and almost similar to that of age-matched women without breast cancer^88,90. Therefore, systemic chemotherapy or other targeted therapy, apart from endocrine therapy, can be avoided in patients diagnosed with pure tubular carcinoma.

Invasive micropapillary carcinoma (IMPC) of the breast is composed in >90% of its area of small, hollow, or morula-like clusters of malignant cells with an inside-out growth pattern, surrounded by clear spaces (Fig. 1T). Pure IMPC represents 0.9–2% of all breast cancers, whereas mixed forms are more frequent, and areas with a micropapillary pattern are found in about 7% of cases⁹¹. Micropapillae are morular aggregates, occasionally hollow, of cuboidal-columnar cells with an inverted or abnormal orientation of the nuclei, highlighted by an inside-out staining pattern with MUC1/EMA. These cell clusters are surrounded by clear spaces in peculiar stroma composed of a delicate framework of fibroblasts and connective tissue, with very rare TILs⁹². IMPC cells typically have an intermediate to high nuclear grade and are frequently bi-nucleated. Although IMPCs lack a pathognomonic genetic alteration, compared to IBC-NSTs, they more frequently harbor MYC amplification⁹³, as well as an enrichment in genetic alterations in genes playing pivotal roles in cellular polarity and shape, such as DNAH9 and FMN2⁹⁴ (Table 1).

IMPCs show higher frequency of lymphovascular invasion compared to IBC-NST (69% vs 26%), and lymph node metastasis are present in more than 2/3 of cases at presentation, with cN3 stage in about 33% of cases (vs 16% in IBC-NST)⁹¹.The majority of IMPCs are Luminal A or B carcinomas^13,93,94,95, but some can be ER-negative/HER2-negative (5–20%) or HER2-positive (5–95%). The HER2 staining pattern is unique, featuring incomplete membrane staining with a basolateral pattern, which needs to be reflexed to in situ hybridization in cases of moderate intensity. Similar 5-year disease-free survival and 5-year overall survival to IBC-NSTs has been described when cohorts are matched for stage and, in particular, nodal status. Nevertheless, it is important to note that IMPCs show higher stage at diagnosis because of the lymphovascular tropism and their special stroma^91,92.

A rare subset of breast cancers display hybrid mucinous and micropapillary features. Notably, whole exome sequencing analysis revealed that a subset of these cases resembled mucinous carcinomas at the genetic level, whereas others were more similar to IMPCs⁹⁶, suggesting that they might originate from mucinous carcinomas or IMPCs.

Male breast cancer is an infrequent presentation of this disease, representing 1% of all cancers in men⁹⁷. In general, men are diagnosed with breast cancer at an older age than women. It is recommended that all men with a diagnosis of breast cancer undergo genetic counseling and testing, as ~10–15% will have pathogenic mutations in BRCA1 or BRCA2. Most men (>90%) present with non-metastatic breast cancer, and their tumors are most often ER-positive (>95%), HER2-negative and of luminal-B phenotype (48.6%)⁹⁸. However, HER2 positivity is observed in about 10% of cases⁹⁸. There have been numerous efforts to better characterize the biology of breast cancer in men and differences compared to female breast cancer; however, no findings have been deemed ready for clinical practice implementation. The prognostic value of genomic signatures (i.e., Oncotype Dx, MammaPrint) seems to be similar to that in female breast cancer, but there are currently no data regarding their clinical utility for chemotherapy decision-making for early breast cancer in men^99,100.

Despite recent advances in breast cancer management, the survival rate for men has not improved in the past three decades¹⁰¹. In general, treatment for early breast cancer in males involves surgery, with either mastectomy (including skin-sparing) or breast-conserving surgery as options. Attention to the aesthetic outcomes are also important for men^99,102. Adjuvant radiation and chemotherapy options are administered with similar protocols as in women^99,102. For adjuvant endocrine therapy, tamoxifen is the treatment of choice; if an aromatase inhibitor is chosen, gonadal suppression is mandatory^99,102. Treatment of metastatic breast cancer in men should follow the same guidelines as for women, and male patients should have access to the same therapeutic options^102,103. In this setting, the choice between tamoxifen and aromatase inhibitor should be based on previous adjuvant treatment, disease-free interval, and patient preferences; preferably, an aromatase inhibitor with gonadal suppression should be used^102,103. Clinical trials are warranted to identify better treatment options for men with breast cancer. Education and awareness are crucial to avoid delays in diagnosis, which regrettably remain frequent.

Inflammatory breast cancer is a rare form of breast cancer defined by its presentation within a span of <6 months with a variable combination of redness or skin discoloration, edema, and dimpling (termed peau d’ orange) of the skin of the breast; a proposed scoring of clinical features is undergoing validity testing¹⁰⁴. Controlling for stage and subtypes, inflammatory breast cancer has a worse prognosis than non-inflammatory breast cancer. In the United States, the incidence of inflammatory breast cancer for white and black women is 2.6/100,000 and 4.5/100,000 and the 5-year mortality is 43% and 30%, respectively¹⁰⁵. Although inflammatory breast cancer survival has increased across all racial groups over the past four decades due to increased awareness and implementation of multidisciplinary care, a survival disparity between white and black patients persists¹⁰⁵. The presence of any clinical signs suggestive of inflammatory breast cancer warrant prompt breast imaging and skin biopsy that should contain the full thickness of the skin to examine dermal lymphatics for the presence of tumor emboli. However, dermal lymphovascular invasion is not a requirement for the diagnosis of inflammatory breast cancer in the appropriate clinical setting, and dermal lymphovascular invasion can be seen in patients without the clinical findings of inflammatory breast cancer. For optimal survival outcomes, inflammatory breast cancer requires a multimodal therapeutic approach, including neoadjuvant multi-drug regimens commonly used for non-inflammatory stage III breast cancer, followed by modified radical mastectomy and radiotherapy. Sentinel lymph node biopsy and immediate reconstruction are not recommended. Definitive biomarkers for inflammatory breast cancer remain elusive. Mechanistic studies have, however, revealed essential roles for the metastases-enabling gene RHOC¹⁰⁶, EGFR overexpression, as well as alterations of the immune microenvironment, such as infiltration of M2 macrophages and regulatory T cells and an immunosuppressive chemokine milieu promoted by the EGFR/EGR1 signaling axis¹⁰⁷.

Young onset breast cancer (YOBC) is variously defined as breast cancers that arise in women aged ≤50, ≤45 or ≤40 years^108,109. Women diagnosed with YOBC have a poorer prognosis in part due to a modest enrichment in TNBC cases in these age groups^108,109. However, the interplay between breast cancer and a woman’s childbearing years appears to impact outcomes more significantly than the ER status of the primary tumor¹¹⁰. This observation implicates reproductive biology rather than tumor-intrinsic biology as the dominant mediator of metastasis in YOBC, an observation supported by recent data from YOBC patients with germline BRCA mutations¹¹⁰. There are two distinct subtypes of YOBC impacted by childbearing, pregnancy-breast cancer (PrBC) and postpartum breast cancer (PPBC)¹¹¹.

PrBC, diagnosed while the women is pregnant, is the most common cancer affecting pregnancy and represents 1 in 3000 cases of breast cancers annually, fitting the criteria for a rare cancer^111,112. PrBC is on the rise, likely due to delays in maternal age at childbirth globally and to increased detection through the advanced cell-free DNA prenatal testing, which incidentally detects occult tumor DNA¹¹³. PrBC is characterized by higher stage at diagnosis, higher grade and proliferation index, and increased frequency of TNBC or HER2-positive subsets, yet outcomes for PrBC are equal to those of non-pregnant women diagnosed with similar tumors^114,115. PPBC are cases diagnosed postpartum, up to 10 years after last childbirth. PPBCs are also increasing globally due to advancing maternal age, as they represent up to 50% of YOBC, and ~5% of total cases, and represent an unmet need for better treatments. PPBCs are notable for increased distant metastasis and death compared to nulliparous women, women diagnosed more than ten years after their last childbirth, and women aged 45–65. Poor outcomes persist after controlling for clinical prognostic markers, patient age, and year of diagnosis, identifying a postpartum diagnosis as an independent predictor of metastasis. Notably, the majority of PPBC cases are ER-positive, and a recent report finds that three times as many young women diagnosed with ER-positive PPBC succumb to metastasis than those diagnosed with ER-negative PPBC¹⁰⁹. Furthermore, liver metastases are preferentially increased in PPBC patients, consistent with a poorly understood interaction between ER-positive PPBC and liver metastasis^116,117. Moreover, PPBCs often recur early, highlighting resistance to treatment¹¹⁸. To date, no hallmark biomarker or genetic alteration has been described for PrBC or PPBC, and the relationship to most recent childbirth remains the most important and under-captured biomarker for risk of metastasis. Several candidate pathways and markers are being explored, including dysregulated ER signaling and immune infiltrate in PPBC¹¹⁹. Such studies promise for better future understanding of the complex interactions between childbirth and breast cancer in young women¹²⁰.

A number of genes are now known to predispose to breast cancer, the most common being BRCA1 and 2, PALB2, TP53, PTEN, ATM, CHEK2, RAD51D, CDH1, STK11 and NF1. For most of the predisposition genes, there is little data on the association with particular morphological subtypes but good evidence that some have stronger associations with ER-positive or ER-negative IBC-NST¹.

While what had been called medullary and atypical medullary carcinoma has an association with BRCA1 predisposition, the WHO no longer recognizes this tumor type as a separate entity, and it is now considered a variant of IBC-NST¹. The strongest morphological association with uncommon subtypes is with ILC. Whilst most tumors arising in the setting of BRCA2 predisposition are high-grade IBC-NST, an association with lobular carcinoma in situ (LCIS) and ILC, including the pleomorphic variant, has been reported¹²¹. ILC is also associated with germline mutations in CHEK2, PALB2 and ATM¹²¹. Germline inactivation of the CDH1 gene, which encodes the cell-cell adhesion protein E-cadherin, also predisposes to ILC, alone or in the setting of hereditary diffuse gastric carcinoma¹²². Although data are limited, two other rare morphological associations have been described, the association of apocrine carcinoma with Cowden’s syndrome (PTEN)¹²³, and rarely, ILC with NF1 germline mutations.

Uncommon breast cancer subtypes have also been reported with germline predisposition, the best example being the association of triple positivity (ER, PR, and HER2) with TP53 mutations (Li–Fraumeni syndrome)¹²⁴. Of the many common low risk predisposition loci that have been linked to breast cancer, most predispose to ER-positive breast cancer in general but lobular-specific polymorphisms have been identified¹²⁵.

Various breast cancer subtypes discussed in this review show distinctive tumor microenvironment (TME) features. A recent study¹²⁶ analyzed immune infiltration in ILC and, while confirming the previously described “cold” TME in the majority of cases, it identified a subset infiltrated with immune cells, especially M2-like macrophages, which were transcriptionally distinct from those in IBC-NST. Differences in gene expression can be mined further in the accompanying single-cell dataset. Of note, there were subtype-specific associations between neighborhoods of T-cells and macrophages and patient outcomes, highlighting the importance of spatial approaches when studying immunobiology in ILC. A machine learning based characterization of tissue-specific cell states revealed that, compared to common forms of TNBC, metaplastic breast cancers exhibit increased infiltration of Tregs, M2-like macrophages, myeloid-derived suppressor cells and cancer associated fibroblasts, creating an immunosuppressive niche enriched in epithelial-mesenchymal transition (EMT) and hypoxia¹²⁷. Single cell (sc)RNA-seq and scTCR sequencing analysis¹²⁸ showed that male breast cancer, compared to female breast cancer, shows lower T cell infiltration and a dysfunctional T cell state, with activation of p38 MAPK and lipid oxidation pathways. Pregnancy and germline BRCA1/2 alterations also shape the immune environment. scRNAseq of benign breast¹²⁹ revealed an association of parity with increased proportions of plasma, CD8 T_EM and CD4 T cell types, which could reflect changes during pregnancy¹²⁹. Immune expansion with increased expression of canonical immune checkpoint/exhaustion receptors in normal breast of BRCA1 and BRCA2 germline carriers was also observed¹²⁹. Notably, Samstein et al. ¹³⁰ demonstrated that BRCA1 and BRCA2 deficiency differentially affects breast cancer TME. Compared to BRCA1-deficient cancers, BRCA2-deficient cases showed an enrichment in adaptive and innate immune gene expression programs and improved survival following immune checkpoint blockade¹³⁰.

With the significant advancement in the application of AI in the field of medical imaging including whole slide image (WSI) technology, it is now possible to use AI algorithms to help pathologists diagnose rare and uncommon breast cancers more accurately and reproducibly. With the availability of datasets containing tens and hundreds of thousands of annotated breast cancer cases, AI will play a crucial role in predicting the behavior and response to therapies for these uncommon breast tumor types. Recently, a diagnostic breast algorithm underwent rigorous training with 2000 annotated slides specifically selected from a series of 115,000 breast biopsies to ensure the representation of uncommon morphologies¹³¹. An expert team of 18 pathologists annotated the slides. The external validation series included 34 rare subtypes of breast cancer, such as tubular, apocrine, mucinous, micropapillary, metaplastic, and acinic cell carcinomas. The algorithm demonstrated strong performance in detecting breast cancer across all histological subtypes (area under the curve (AUC): 0.998, in the training set; AUC: 0.990, in the validation cohort). The differentiation between ILC and IBC-NST also displayed high accuracy (AUC: 0.932 in the training and AUC: 0.973 in the validation series)¹³¹. AI models can identify genomic, gene expression, and epigenetic data that are reflected on the morphology of the tumors through analysis of myriads of features from WSIs¹³². A recent AI-powered method was developed to diagnose ILC and predict the presence of CDH1 biallelic mutations⁵⁹. The model yielded an AUC of 0.959 for CDH1 mutational status prediction and 0.983 for ILC diagnosis.

The scarcity of cases has significantly hindered the development of customized detection models for other uncommon breast cancer subtypes. The emergence of foundation models that enable generalization from large datasets and is revolutionizing AI-based cancer detection¹³³, is particularly valuable for uncommon entities, as they address the limitations posed by their rarity. A pan-cancer detector built on the largest foundation model in computational pathology to date, demonstrated superior performance for detection of rare breast cancer subtypes compared to traditional tissue-specific clinical grade models¹³⁴. For instance, it detected breast adenoid cystic carcinoma and secretory carcinoma with AUCs of 0.974 and 0.998, compared to AUCs of 0.899 and 0.794 obtained with breast-specific clinical models¹³⁴.

With ongoing global digitalization in pathology laboratories and the promising development of AI tools, achieving improved reproducibility in diagnosis and prognostic stratification of rare tumor types seems likely in the near future. The success of these algorithms, however, depends on the quality of the input data, which is key to ensure robust training.

Novel therapeutic approaches that target the specific ‘omics’ (genomics/transcriptomics/proteomics/metabolomics) or immunophenotypes of uncommon breast cancers might be the route towards better outcomes for patients with these subtypes. Across various subtypes, targeting HER2-low with Trastuzumab deruxtecan may be effective^135,136. PI3-kinase and AKT inhibitors may also be beneficial in subtypes showing an enrichment in PIK3CA mutations and other genetic alterations in the PI3-kinase pathway. Metaplastic breast cancers often possess targetable molecular alterations, including (i) PIK3CA/PTEN mutations targetable via PI3-kinase and/or AKT inhibitors, (ii) alterations in the nitric oxide signaling gene RPL39, targeted via NOS inhibitors¹³⁷, and (iii) mutations in the Wnt pathway driver genes β-catenin (CTNNB1) or APC¹¹ targetable via tankyrase, porcupine or frizzled blockade. The epithelial-mesenchymal transition (EMT) phenotype in metaplastic breast cancer¹³⁸ could also, in principle, be targeted by ferroptosis-inducing small molecules such as GPX4 inhibitors¹³⁹. The immune milieu of metaplastic breast cancer could also be targeted via checkpoint blockade, and indeed, a prospective phase II trial assessing the anti-CTLA4 antibody ipilimumab, combined with the anti-PD1 antibody nivolumab, elicited a 18% objective response rate (ORR) in patients with advanced metaplastic breast cancer¹⁴⁰. Pembrolizumab was found to be active in metaplastic breast cancer in a phase I study in synergy with chemotherapy¹⁴¹. In other uncommon breast cancer subtypes, synthetic lethal approaches have been proposed. Loss or mutation of E-cadherin (CDH1) in pre-clinical models of ILC can be synthetic lethal targeted using small molecule MET/ALK/ROS1 inhibitors¹⁴², an effect being currently tested clinically in two trials⁵⁵. In ILC, other therapeutic approaches have also been proposed, including: (i) targeting of mTOR, PI3K/AKT, FGFR1 or IGF1; (ii) targeting of ERBB2 and ERBB3 mutations via therapeutic antibodies/antibody-drug conjugates; (iii) immune checkpoint blockade in ILCs with elevated tumor mutational burden; or (iv) AR inhibition in AR-positive ILCs⁵⁵. AR inhibition might provide therapeutic benefits in other subtypes, such as apocrine carcinomas, due to their high prevalence of AR positivity¹⁴. Nonetheless, given the enrichment of AR splice variant-7 (AR-V7) observed in apocrine carcinoma¹⁴³, testing for its presence would be warranted. In ILC, and other uncommon breast cancer subtypes, the assessment and ultimate application of novel therapeutic approaches could also be enhanced by methods that aid diagnosis, staging and monitoring of disease burden. For instance, methods such as positron emission tomography (PET) using 16a-18F-fluoro-17b-estradiol (¹⁸F-FES) are non-invasive approaches that exploit the ER ligand-binding of breast cancer lesions as a way of improving tumor staging and assessment of ER expression across the whole body tumor burden¹⁴⁴; applying such approaches to the early detection of metastatic disease in ILC could also improve overall outcomes¹⁴⁴.

Emerging technologies have facilitated the recognition of a subset of special histologic subtypes of breast cancer that represent molecular entities with distinctive phenotypes¹⁴⁵. Ongoing studies are contributing to the further refinement of breast cancer taxonomy and unveiling convergent phenotypes, in the form of molecular alterations targeting the same signaling pathway, in these uncommon breast neoplasms. For instance, alterations in genes playing key roles in cell-to-cell adhesion other than CDH1, such as AXIN2 or CTNND1 alterations, have been recently described in ILC⁶¹ (Fig. 2B). Furthermore, novel entities, such as tubulopapillary/serous-like carcinoma¹⁴⁶, although not yet in included in the WHO classification, are being described.

Despite substantial advances made in the study of these uncommon breast cancer subtypes, significant gaps persist, such as the need for advanced diagnostic approaches, deeper insights into their unique biology, elucidation of their immune and other TME features, unveiling the molecular mechanisms of progression and identification of novel therapeutic targets. Given the rarity of these entities, their accurate histologic diagnosis is challenging, underscoring the need for efforts to refine diagnostic criteria and to optimize the use of ancillary tools, as conducted for ILC^57,147. Furthermore, genotypic-phenotypic correlations that characterize a subset of these uncommon breast cancer subtypes could be used for the development of AI-based algorithms for their diagnosis, by training AI-models using a genetic, rather than a histologic ground truth. Moreover, foundation models, that allow the development of robust AI tools with smaller cohorts¹³³, could be leveraged to improve the diagnosis of these rare entities. Emerging post-mortem research programs, which allow the collection of high-volume samples, have the potential to provide key insights into progression, therapeutic resistance, tumor heterogeneity, dormancy, and interactions with TME in these uncommon breast cancer types. Finally, subtype-specific breast cancer clinical trials, such as those investigating synthetic lethal approaches in ILC, or those focusing on young women’s breast cancer with attention to parity status for PPBC identification, hold promise for advancing precision medicine in these uncommon and unique entities.

The limited number of cases available for these uncommon entities pose significant challenges, however. Hence, multi-institutional efforts are essential for the accrual of these cases. Sharing of biospecimens and/or the development of centralized repositories for clinicopathologic and molecular data would facilitate research efforts in these uncommon breast cancer subtypes. Standardization of data collection and handling of biomaterials across institutions is crucial for these efforts. Multi-institutional collaboration would not only facilitate the identification of molecular drivers, which could serve as therapeutic targets and be potentially leveraged in consortium trials, but also allow the development and implementation of AI tools for the diagnosis and study of these entities. Diligent endeavors to understand biology and molecular underpinnings, alongside efforts to advance diagnostic tools and therapeutic approaches for these special breast cancer subtypes, hold promise for improving outcomes in affected patients.