Breast Cancer Classification: Molecular Subtypes and Staging

Breast cancer is not a single disease. It is classified along several overlapping axes at once: what the cells look like under a microscope, which receptors they carry on their surface, how their genes behave, how far the disease has spread, and increasingly, how the tumor interacts with the immune system. Each classification layer answers a different clinical question, and together they guide nearly every treatment decision, from whether chemotherapy is worth pursuing to whether a targeted drug can be used. Understanding how these layers fit together helps make sense of a diagnosis that can feel overwhelmingly technical.

Histological Type

The oldest way to classify breast cancer is by looking at the tumor cells under a microscope and identifying their shape and growth pattern. The vast majority of breast cancers are invasive ductal carcinoma (now often called “invasive carcinoma of no special type”), which accounts for roughly 80% of cases. The second most common is invasive lobular carcinoma, making up about 10–15% of diagnoses.¹ These two types behave differently in ways that matter. Lobular cancers are harder to detect on mammography, tend to be diagnosed at a larger size, and are more likely to be estrogen-receptor positive.²

Beyond these two dominant types, there are more than a dozen special histological subtypes, collectively accounting for about 14% of invasive breast cancers.³ These include mucinous, tubular, papillary, medullary, metaplastic, micropapillary, and cribriform carcinomas, among others. Some carry notably better prognoses: mucinous breast cancer, for instance, tends to present at an earlier stage with less lymph node spread and has five- and ten-year survival rates well above those of standard ductal carcinoma.⁴ Metaplastic carcinoma, on the other hand, is associated with worse overall survival, and medullary carcinoma, despite its aggressive-looking features under the microscope, tends to have a somewhat better outcome than ductal carcinoma overall.⁵ These subtypes are not just academic footnotes. They influence whether hormone therapy, chemotherapy, or other treatments are appropriate.

Receptor Status

Histology tells you what the cancer looks like. Receptor status tells you what the cancer responds to. Every breast cancer biopsy is tested for three key markers: estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). These are measured using a staining technique called immunohistochemistry, sometimes supplemented with additional testing for HER2 when the initial result is borderline.⁶

If a tumor is ER-positive, it means the cancer cells have receptors that respond to estrogen, and hormone-blocking drugs can slow or stop its growth. If it is HER2-positive, the cancer cells overproduce the HER2 protein, and drugs that target HER2 specifically can be added to treatment. If a tumor lacks all three markers, it is called triple-negative breast cancer (TNBC), and treatment relies mainly on chemotherapy and newer approaches like immunotherapy. Receptor status is arguably the single most important classification for deciding treatment, because it determines whether an entire category of targeted therapy is on the table.

One practical wrinkle: receptor status is sometimes tested twice, once on the initial needle biopsy and again on the surgically removed tumor. These results agree most of the time, but not always. In one study, the agreement between needle biopsy and surgical specimen was about 90% for estrogen receptor, 81% for progesterone receptor, and 97% for HER2.⁷ That means progesterone receptor results are the most likely to shift between biopsy and surgery, which can occasionally change treatment planning.

The Rise of HER2-Low

For years, HER2 was treated as a yes-or-no question: either a tumor overexpresses HER2 (positive) or it does not (negative). But it turns out that nearly half of tumors previously labeled “HER2-negative” actually show some low-level HER2 expression.⁸ This in-between zone, now called HER2-low, is defined as a score of 1+ on immunohistochemistry or 2+ without gene amplification.⁹

HER2-low matters because newer antibody-drug conjugates can deliver chemotherapy directly to cells expressing even small amounts of HER2. This has turned a classification distinction that was previously irrelevant into one that opens up new treatment options. The challenge is that scoring HER2 at these low levels is not always reproducible between pathologists or between different tissue samples from the same patient, raising questions about how reliably this category can guide treatment decisions.¹⁰

Molecular Subtypes

Receptor status gives you three markers. Molecular subtyping, typically done through gene expression profiling, looks at the activity of dozens of genes at once and groups tumors into biologically distinct categories. The most widely used molecular classification identifies roughly five intrinsic subtypes: Luminal A, Luminal B, HER2-enriched, Basal-like, and Normal-like. In one large cohort study, these broke down to about 46% Luminal A, 18% Luminal B, 14% HER2-enriched, 15% Basal-like, and 8% Normal-like.¹¹

These subtypes carry real prognostic meaning. Luminal A tumors tend to be slower-growing, strongly hormone-receptor-positive, and generally have the best outcomes. Luminal B tumors are also hormone-receptor-positive but grow faster and sometimes benefit from chemotherapy in addition to hormone therapy. The Basal-like subtype overlaps heavily with triple-negative breast cancer and has a higher rate of relapse within the first five years.¹² HER2-enriched tumors are driven by HER2 overexpression and respond to HER2-targeted therapies. The Normal-like category is less clinically useful and sometimes considered a testing artifact.

Molecular subtyping does not perfectly align with receptor status measured by standard staining. When luminal subtypes are grouped together, agreement between gene expression profiling and standard receptor testing improves but remains imperfect.¹³ This means that two patients who look identical on a standard pathology report can belong to different molecular subtypes with different prognoses, which is one reason gene expression testing has become increasingly important.

Triple-Negative Breast Cancer Is Not One Disease

Triple-negative breast cancer deserves special attention because it is defined by what it lacks rather than by what it is. The absence of estrogen receptor, progesterone receptor, and HER2 overexpression leaves TNBC without the obvious drug targets that other breast cancers have. But gene expression analysis has shown that TNBC itself is a collection of biologically distinct diseases. An influential study identified six molecular subtypes within TNBC: two basal-like types, an immunomodulatory type, a mesenchymal type, a mesenchymal stem-like type, and a luminal androgen receptor subtype.¹⁴ Subsequent work refined this to four core tumor-specific subtypes, noting differences in age at diagnosis, grade, and patterns of disease progression.¹⁵

The luminal androgen receptor subtype is particularly interesting because, despite being triple-negative by standard testing, these tumors are driven by the androgen receptor and behave more like hormone-receptor-positive cancers in some respects. Clinical trials are exploring whether anti-androgen therapies could benefit this group. The basal-like subtypes, which make up the majority of TNBC, tend to be more aggressive and responsive to chemotherapy. This internal diversity is why researchers increasingly argue that treating all TNBC the same way misses opportunities for more tailored approaches.

Clinical Staging and the Shift Toward Biology

Staging describes how far a cancer has spread. The standard system, maintained by the American Joint Committee on Cancer, has historically relied on three anatomic factors: tumor size (T), whether cancer has reached nearby lymph nodes (N), and whether it has spread to distant organs (M). Combining these produces a stage from I through IV. But clinicians noticed that patients with identical anatomic stages sometimes had wildly different outcomes, and the reason turned out to be biology.¹⁶

The eighth edition of the staging system, adopted in 2018, addressed this by integrating tumor grade, hormone receptor status, HER2 expression, and gene expression profiling results alongside the traditional anatomic information.¹⁷ The result is that a small, node-negative, hormone-receptor-positive tumor with a favorable gene expression profile can now be staged lower than the same-sized tumor that is triple-negative with a high grade. This is a major conceptual shift: stage is no longer purely about anatomy. Low scores on commercially available gene-based assays can actually lower a patient’s prognostic stage.¹⁸

Genomic Tests That Guide Chemotherapy Decisions

Two gene expression tests come up frequently in early-stage breast cancer: Oncotype DX and MammaPrint. Both analyze tumor tissue to predict how likely the cancer is to return and whether chemotherapy would meaningfully reduce that risk. They address different but overlapping clinical questions and are used primarily in hormone-receptor-positive, HER2-negative early breast cancers, precisely the group where the chemotherapy decision is most uncertain.

Oncotype DX analyzes 21 genes and produces a recurrence score. The test is included in treatment guidelines for estimating both distant recurrence risk and likely chemotherapy benefit in patients with early-stage, hormone-receptor-positive, HER2-negative disease, including those with one to three positive lymph nodes.¹⁹ Patients with low recurrence scores generally derive little benefit from chemotherapy, sparing them its side effects.

MammaPrint is a 70-gene signature that classifies tumors as either genomically low-risk or high-risk. In the landmark MINDACT trial, patients who were classified as high clinical risk but low genomic risk had a five-year rate of survival without distant metastasis above 94% even without chemotherapy, and adding chemotherapy improved that rate by only about 1.5 percentage points.²⁰ Longer-term data have confirmed that the 70-gene signature separates outcomes over 25 years of follow-up as well.²¹ More recently, MammaPrint added an “ultralow” risk category. In a study of patients 70 and older, those classified as ultralow had a ten-year distant recurrence rate of just 2%, compared with 17% for high-risk patients.²²

Immune Cells as a Classification Layer

A relatively recent addition to the classification toolkit is measuring how many immune cells, specifically tumor-infiltrating lymphocytes (TILs), are present in and around the tumor. The density of TILs turns out to carry significant prognostic information, but the effect depends on the tumor subtype. In triple-negative and HER2-positive breast cancers, higher TIL levels are associated with better outcomes. A pooled analysis of nearly 3,800 patients who received chemotherapy before surgery found that each 10% increase in TILs was linked to improved disease-free survival in TNBC and HER2-positive tumors.²³

In TNBC specifically, the evidence is strong enough that TILs may eventually factor into clinical decisions. A large analysis found that each 10% increase in TILs was associated with roughly an 8% lower risk of disease recurrence and a 12% lower risk of death after adjusting for other factors like tumor size and lymph node status.²⁴ This finding is particularly important because TNBC has few reliable prognostic markers beyond stage and grade. TILs are one of the few biological features that consistently predict who does better within a subtype that is otherwise defined by the absence of actionable markers.

The picture is more complicated in luminal, HER2-negative breast cancers, where higher TIL levels may actually be associated with slightly worse overall survival.²⁵ The reasons for this are not entirely settled, but it may reflect the immune system reacting to a more genetically unstable or proliferative subset of luminal tumors.

Classifying How Well a Tumor Responds to Treatment

When chemotherapy is given before surgery (neoadjuvant chemotherapy), the surgical specimen provides a direct measure of how well the treatment worked. If no cancer cells remain, that is called a pathologic complete response, or pCR, and it generally signals a very good prognosis. But the range between pCR and “nothing happened” is wide, so the residual cancer burden (RCB) index was developed to quantify what is left. It considers the dimensions of any remaining tumor, the density of cancer cells within it, and the burden of cancer in nearby lymph nodes, producing four categories: RCB-0 (complete response), RCB-I (minimal residual disease), RCB-II (moderate residual disease), and RCB-III (extensive residual disease).²⁶

The prognostic separation between these categories is dramatic. In one study, compared with patients who achieved a complete response, those with RCB-II had more than four times the risk of death and those with RCB-III had more than ten times the risk. The pattern held for recurrence as well, with risk rising steeply from RCB-0 through RCB-III across all breast cancer subtypes.²⁷ RCB gives clinicians a practical tool to decide whether additional therapy after surgery, such as a different chemotherapy regimen or targeted agents, is warranted.

Radiological Classification and BI-RADS

Before a tissue sample is ever taken, breast abnormalities are classified on imaging using the BI-RADS system (Breast Imaging Reporting and Data System). This standardized scale runs from 0 (needs more imaging) through 6 (known cancer) and helps radiologists communicate how suspicious a finding looks and whether a biopsy is needed. The categories most relevant to cancer detection are BI-RADS 4 (suspicious, with sub-categories 4A, 4B, and 4C reflecting increasing concern) and BI-RADS 5 (highly suggestive of malignancy).²⁸

For microcalcifications, which are tiny calcium deposits sometimes visible on mammography, the specific shape and distribution pattern help predict malignancy risk. Fine pleomorphic calcifications arranged in a linear or segmental pattern carry a very high positive predictive value for cancer, while amorphous calcifications in a regional pattern have a much lower one (around 4%).²⁹ BI-RADS does not classify the cancer itself; it classifies the imaging finding and the urgency of the next step. It is the first classification a patient encounters in the diagnostic process, often before any tissue diagnosis has been made.

Liquid Biopsy and Circulating Tumor DNA

An emerging classification frontier involves blood tests rather than tissue samples. Circulating tumor DNA (ctDNA), fragments of tumor genetic material shed into the bloodstream, can be used to detect minimal residual disease after treatment, sometimes months before a relapse becomes visible on imaging. In a prospective study of early breast cancer patients who had completed surgery and chemotherapy, ctDNA detection in blood predicted metastatic relapse with a median lead time of nearly eight months over clinical detection.³⁰

More recent work has explored “plasma-only” approaches that do not require a prior tumor biopsy as a reference, relying instead on detecting both cancer-specific mutations and methylation patterns in a blood sample. One study using this approach detected ctDNA at or before distant recurrence in about 79% of patients who relapsed.³¹ These tests are not yet standard practice, but they point toward a future where breast cancer classification extends beyond a one-time snapshot of tumor tissue and incorporates real-time monitoring of disease activity through a simple blood draw.

Racial Disparities in How Breast Cancer Presents

Breast cancer subtypes are not distributed equally across racial and ethnic groups, and this has tangible consequences for treatment and outcomes. In the United States, Black women have roughly twice the incidence of triple-negative breast cancer compared with White women. A national analysis found the TNBC incidence rate among Black women was about 25 per 100,000, compared with roughly 13 per 100,000 among White women. The disparity was statistically significant in every state with enough data to measure it, with the rate ratio ranging from about 1.4 in Colorado to 2.3 in Delaware. Asian and Pacific Islander women had the lowest rates, at about 9 per 100,000.³²

Because TNBC lacks the hormone receptor and HER2 targets that make other subtypes treatable with less toxic targeted therapies, the higher incidence among Black women contributes to worse overall breast cancer outcomes in that population. The reasons behind the disparity are an active area of research and likely involve a combination of genetic ancestry, differences in environmental exposures, and systemic inequities in access to screening and early treatment. From a classification standpoint, these patterns underscore that the distribution of molecular subtypes varies by population, and awareness of this variation matters for screening strategies and clinical trial enrollment.

AI and the Future of Pathology Classification

Pathologists have been classifying breast cancers by eye for well over a century, but computational tools are increasingly entering the picture. Deep learning algorithms trained on digitized whole-slide images of breast tissue can now classify tumors as benign, in situ, or invasive with reasonable accuracy.³³ Separate efforts are using AI to identify and quantify features like tumor-infiltrating lymphocytes directly from slide images, potentially standardizing a measurement that currently depends heavily on the individual pathologist’s judgment.³⁴

These tools are not replacing pathologists. They function more like a second set of eyes, flagging regions of interest, reducing interobserver variability, and potentially speeding up workflows. The promise is that as more classification layers are added to breast cancer diagnosis (molecular subtypes, TIL density, HER2-low status), AI could help integrate information that is currently spread across multiple tests into a more unified picture. The limitation, for now, is that most of these systems perform well on the specific datasets they were trained on but can falter when applied to tissue processed differently at a different hospital, a problem researchers are actively working to solve.