Liquid Biopsy in Breast Cancer: A Focused Review

A liquid biopsy refers to a sample of body fluid that contains a variety of tumor-derived materials such as tumor DNA, RNA, intact circulating tumor cells (CTCs), tumor-educated platelets, or extracellular vesicles. Although they may be obtained in a similar manner, such as by blood draw or urine collection, isolation of the different tumor-derived materials, such as CTCs and circulating tumor DNA (ctDNA), from the sample of body fluid requires different technologies (see Figures 1 and 2). The 2 different components of a liquid biopsy also provide information that is independent, yet complementary.⁸  The idea of a liquid biopsy probably has its origin as far back as 1948, when cell-free DNA (cfDNA) was first detected in the bloodstream.⁹  Unlike ctDNA, cfDNA is present even in healthy individuals¹⁰  and can be found in increased quantities during trauma, myocardial infarction, or stroke or in autoimmune conditions.¹¹  Circulating tumor DNA is the fraction of cfDNA that is released by tumor cells via necrosis or apoptosis either at the tumor site, or from CTCs in the bloodstream. Circulating tumor cells are tumor cells that actively enter the body circulation or are passively shed from primary or metastatic tumors.¹² 

Other tumor-derived materials include circulating RNA such as cell-free messenger RNA and microRNA, which are noncoding RNA molecules of 19 to 24 nucleotides in length.¹³  They may be found in extracellular vesicles in the circulation, which are membrane-bound vesicles that are shed either by plasma membrane shedding from cells such as tumor-educated platelets or by tumor cells via exocytosis (the resulting extracellular vesicles being termed exosomes). These vesicles can contain tumor DNA and RNA as well as tumor-associated proteins and lipids,¹⁴  conferring protection and preventing the degradation of their contents from enzymes such as plasma nucleases. Tumor-educated platelets are platelets that have taken up tumor RNA from the circulation, similarly protecting them from degradation, and may have enhanced functions promoting tumor metastasis.¹⁵ 

Of the various tumor-derived materials in a liquid biopsy, CTCs and ctDNA are arguably the most well studied. Recently, the US Food and Drug Administration approved the Therascreen PIK3CA RGQ polymerase chain reaction (PCR) assay as a companion diagnostic assay to detect PIK3CA mutations after a clinical trial found patients with hormone receptor–positive, human epidermal growth factor receptor 2 (HER2)–negative, PIK3CA-mutated, advanced, or metastatic breast cancer to have significantly prolonged progression-free survival (PFS) when treated with alpelisib in combination with endocrine therapy fulvestrant than fulvestrant alone.¹⁶  The companion diagnostic assay was approved for use on tissue as well as liquid biopsies, thus bringing liquid biopsies in breast cancer another step closer to the bedside. In this review, we will focus mainly on the studies on CTCs and ctDNA in breast cancer that have been performed thus far (with selected studies summarized in the Table).

Although breast cancer is one of the few cancers with a well-established screening test using mammography, many authors have endeavored to investigate if tumor-derived materials in the blood of breast cancer patients can be used for early detection. Kruspe et al¹⁷  were able to demonstrate the utility of probes activated by nucleases derived from CTCs to discriminate metastatic breast cancer patients from healthy controls (area under the curve ranged from 0.851 [95% CI, 0.76–0.94] to 0.903 [95% CI, 0.80–1.00], depending on the probe used). The CTCs were isolated using the CTC capture technology ScreenCell (ScreenCell), which enables the capture of both epithelial and mesenchymal malignant cells from blood based on cell size. Although the assay appears to effectively discriminate metastatic breast cancer patients from healthy controls, its utility in detecting early breast cancer is uncertain, given that CTCs usually occur in lower levels in lower-stage disease.

Some studies also showed that the quantity of cfDNA in breast cancer patients were higher than in normal healthy controls and could have discriminatory value.^18,19  A study by Schwarzenbach et al,¹⁸  however, highlighted that patients with benign breast conditions also have elevated cfDNA. It is thus difficult to distinguish them from patients with cancer. Another study, by Agostini et al,¹⁹  looked at cfDNA integrity index. This is based on the observation that cfDNA released by tumor cells are usually larger fragments as a result of tumor necrosis, compared with the smaller fragments of cfDNA released during apoptosis in normal individuals. Agostini et al¹⁹  found that the baseline levels of ALU247 (tumoral cfDNA fragments of 247 base pairs using ALU247 primers) were significantly higher in cancer patients and were accurate in discriminating them from noncancer subjects (area under the curve = 1.00; 95% CI, 0.96–1.00). The levels of ALU247 were also significantly higher in 11 patients with lymph node metastasis compared with 22 patients without nodal disease. Apart from ALU247, loss of heterozygosity at 4 polymorphic markers in cfDNA (D13S159, D13S280, D13S282 at region 13q31-33, and D10S1765 at PTEN region 10q23.31) analyzed by PCR-based fluorescence microsatellite analyses also correlated significantly with lymph node status (P = .03).¹⁸  In addition to cfDNA quantity, integrity, and loss of heterozygosity, single-nucleotide polymorphisms in cfDNA also differ significantly from normal leukocytes in breast cancer patients compared with normal controls and may have potential in breast cancer diagnosis.²⁰ 

Compared with cfDNA, ctDNA is present in tiny fractions in a patient's blood, especially in patients with localized cancers. Board et al,²¹  using amplification-refractory mutation system allele-specific PCR and Scorpion probes, were able to detect the presence of PIK3CA mutations in ctDNA in 80% of patients with metastatic breast cancer harboring PIK3CA mutations. They were, however, not able to detect the presence of PIK3CA mutations in ctDNA for patients with localized breast cancer. In a study conducted by Beaver et al,²²  14 of 15 patients (93%) with nonmetastatic breast cancers harboring PIK3CA mutations had the same mutation detected in their ctDNA before surgery using droplet digital PCR (ddPCR). Bettegowda et al⁸  managed to detect the presence of ctDNA in 50% of patients with localized breast cancer using highly sensitive technologies such as BEAMing (beads, emulsification, amplification, and magnetics), PCR ligation, and massive parallel sequencing. The different results of these studies reflect the varying sensitivities of different ctDNA assays and highlight the need for highly sensitive assays to detect the presence of ctDNA if it is to be incorporated into breast cancer screening. In a meta-analysis of 69 studies on 5736 patients with breast cancer, Lee et al²³  found ctDNA mutation rates of TP53, PIK3CA, and ESR1 to be approximately 38%, 27%, and 32%, respectively, and concluded that they were too low to be used for breast cancer screening. In addition, there is also uncertainty as to how to manage patients with extremely early-stage disease where there is detectable ctDNA but without a clinically or radiologically evident lesion. The same genetic mutation, such as a TP53 mutation, can also be found in more than one cancer type and may not be specific to breast cancers. In order to overcome this limitation, Cohen et al²⁴  designed a test that takes into account not just the somatic mutations found in ctDNA but also the protein biomarkers found in plasma in order to accurately localize the primary site of the cancer. For the detection of ctDNA mutations, the test uses massive parallel sequencing to interrogate mutations across 16 genes via a 61-amplicon panel. Unfortunately, although the test has a specificity of greater than 99%, it has a sensitivity of only 33% in detecting breast cancers. Looking at the aforementioned studies, it appears that the search for a liquid biopsy test that can surpass the sensitivity of current breast cancer screening algorithms is not yet over.

Numerous studies have found liquid biopsies to have value in predicting survival outcomes. These studies looked at the levels of CTCs and ctDNA in both the nonmetastatic and metastatic settings. In a landmark study more than a decade ago, Cristofanilli et al²⁵  found that in 177 patients with metastatic breast cancer, those with 5 or more CTCs per 7.5 mL of whole blood at baseline before initiation of treatment had a shorter median PFS (2.7 versus 7.0 months, P < .001) and shorter overall survival (OS) (10.1 versus >18 months, P < .001) compared with those with fewer than 5 CTCs. Two recent studies showed similar results, one involving 40 metastatic breast cancer patients in which patients with 5 or more CTCs per 7.5 mL of blood had shorter OS (P = .04) and an almost statistically significant shorter time to progression (P = .06),²⁶  and another study by Rossi et al²⁷  on 91 patients who showed a difference in PFS and OS for baseline 5 or more versus fewer than 5 CTCs (P = .02 and P < .001, respectively). In a different study²⁸  on patients with nonmetastatic disease, patients with stage 1 through 3 breast cancer with 1 or more CTCs were found to have significantly lower PFS and OS; this was also found in a pooled analysis²⁹  of a few other studies. The CTC level in the blood also appears to provide additional prognostic information on top of imaging results. In a study of 138 patients with imaging studies done before and a median of 10 weeks after the initiation of therapy with CTC counts determined approximately 4 weeks after initiation of therapy, the median OS of patients with radiologic nonprogression and 5 or more CTCs was significantly shorter than that of the patients with radiologic nonprogression and fewer than 5 CTCs (15.3 versus 26.9 months; P = .04). The median OS of the patients with radiologic progression and fewer than 5 CTCs was significantly longer than that of the patients with 5 or more CTCs who showed progression by radiology (19.9 versus 6.4 months; P = .004).³⁰ 

To investigate the relationship of dynamic changes in CTC levels with clinical outcome, the patients in the study by Cristofanilli et al²⁵  were subsequently followed up and had their blood drawn at another 4 time points, leading up to 20 weeks after initiation of treatment.³¹  Patients with 5 or more CTCs had a significantly shorter OS at every time point up to 20 weeks and a significantly shorter PFS up to 14 weeks. Patients who had 5 or more CTCs at baseline but had a decrease to fewer than 5 CTCs during the course of 20 weeks also had better OS and PFS compared with those with persistently 5 or more CTCs. In another study, Pierga et al³²  found that patients with a drop in CTC count to fewer than 5 per 7.5 mL of blood postchemotherapy had better PFS (P < .001) and OS (P < .001). On the other hand, having 5 or more CTCs per 7.5 mL of blood is associated with radiographic progression even when measured 7 to 9 weeks before imaging.³³  Such a correlation between the level of CTCs in the blood and survival suggests that CTCs could be used to monitor treatment response and prompt changes in the therapeutic regime if necessary.

However, not all assays used to determine CTC positivity show the same results. The studies mentioned above used the CellSearch system (Menarini Silicon Biosystems), which captures CTCs from peripheral blood by anti–EpCAM-antibody-bearing ferrofluid nanoparticles and identification of the CTCs by their cytokeratin-positive/CD45-negative immunophenotype. In a study of 221 metastatic breast cancer patients, the results of another assay, the AdnaTest Breast, which determines CTC positivity by reverse transcription PCR of CTC messenger RNA to detect 3 tumor-associated transcripts (GA733-2, MUC-1, and HER2), did not show any association with OS.³⁴ 

At a static time point, ctDNA levels, similar to CTCs, correlate well with survival outcomes. Higher levels of ctDNA in the blood are associated with poorer recurrence-free survival and OS independent of other clinicopathologic factors.^35–38  One study,²⁶  however, found that ctDNA levels had no prognostic impact on time to progression or OS, which contradicts the findings of a similar study using the same sequencing technology.³⁹  The authors²⁶  postulated that one reason could be that their patient population comprised only triple-negative breast cancer patients rather than a mixture of different breast cancer types, unlike the other study. In addition to ctDNA levels, the number of genetic alterations in ctDNA (<2 versus ≥2) was also found to have an impact on PFS and OS.²⁷  When looking at cfDNA alone, the meta-analysis by Lee et al²³  found that cfDNA levels correlated with axillary lymph node metastasis (odds ratio, 2.148; P = .03).

When ctDNA is serially monitored, changes in ctDNA levels or mutant allele frequencies after treatment correlate with parameters such as tumor size change and survival outcomes.^40,41  This can be used to assess and monitor responses to treatment. In a study by Dawson et al,³⁹  ctDNA levels showed a greater correlation with changes in tumor burden when compared with plasma CA 15-3 levels or CTCs. This study also provided the earliest measure of treatment response, with an average lead time of 5 months before imaging.

In the nonmetastatic setting, ctDNA monitoring after neoadjuvant treatment and surgery can provide a means to assess the presence of minimal residual disease and the risk of relapse. In a retrospective study of 20 patients diagnosed with breast cancer with long-term follow-up, using a combined approach of low-coverage whole-genome sequencing of the primary tumor and quantification of tumor-specific rearrangements in the patient's plasma by ddPCR, the authors⁴²  found that ctDNA monitoring postsurgery was able to accurately discriminate patients who ended up with metastatic disease from those without (sensitivity 93%, specificity 100%, receiver operator curve area 0.98, P = .001). In addition, the detection of ctDNA preceded clinical detection of metastasis in 86% of patients with an average lead time of 11 months (range, 0–37 months), although patients without metastasis/disease recurrence had undetectable ctDNA postoperatively. In another study that looked at methylation of RASSF1 gene in ctDNA, RASSF1 methylated ctDNA significantly decreased in patients whose disease responded to neoadjuvant therapy (P = .006) but not in patients with no response to therapy. Among 47 patients with negative methylated ctDNA 1 year after neoadjuvant chemotherapy and surgery, none had recurrence after a median follow-up period of 23 months (range, 3–33 months).⁴³  The absence of ctDNA RASSF1 gene methylation after neoadjuvant treatment also predicted pathologic complete response more accurately than mammogram and ultrasound.⁴⁴  Conversely, detection of ctDNA in plasma after completion of neoadjuvant chemotherapy and surgery had predictive value for metastatic relapse with a median lead time of 7.9 months.⁴⁵  Sequencing the ctDNA also predicted the molecular profile of the subsequent relapsed tumor more accurately than sequencing of the primary cancer. This provides more valuable information that can better guide the treatment of the relapsed cancer.

Mutations in the ESR1 gene, which encode for estrogen receptor α (ER-α), have been found to appear in breast cancers of patients who received endocrine therapy. Primary breast cancers usually have few or no ESR1 mutations, although metastatic breast cancers are found to have higher rates of ESR1 mutation. Studies have shown that these mutations may be acquired under the pressure of endocrine therapy.^46–48  In a study by Allouchery et al⁴⁹  that looked at patients with early-stage breast cancer with relapse after adjuvant aromatase inhibitor (AI) therapy, no ctDNA ESR1 mutation was detectable at the end of AI adjuvant therapy. At first relapse, 5.3% (2 of 38) of the patients had a detectable circulating ESR1 mutation. At the time of progression on first-line metastatic treatment, 33% (7 of 21) of the patients under AI had a detectable circulating ESR1 mutation, compared with none of the 10 patients under chemotherapy. The prevalence of ESR1 mutations appear to be higher for patients treated with AI in the metastatic setting than in the adjuvant setting. Schiavon et al⁵⁰  found that ctDNA ESR1 mutation prevalence differed between patients exposed to AI as part of adjuvant treatment and patients given AI for metastatic disease (5.8% [3 of 52] versus 36.4% [16 of 44], P < .001). None of the patients in the study without previous AI exposure had detectable ESR1 mutations. Metastatic breast cancer patients with ctDNA ESR1 mutations also had a shorter PFS on subsequent AI-based therapy (hazard ratio, 3.1; 95% CI, 1.9–23.1; P = .004), linking ESR1 mutations to resistance to AI therapy. As mentioned above, profiling ctDNA for ESR1 mutations may be superior to sequencing the primary tumor, as they may mirror the molecular profile of the relapsed/metastatic tumor more accurately. Spoerke et al⁵¹  found that the concordance between ESR1 status in ctDNA and metastatic tissue was 47% whereas concordance between ctDNA and primary tissue was 5%. Concordance was 57% and 23%, respectively, for patients whose tumor tissue was collected after and before the administration of an AI. ESR1 mutations in tumor tissue that was collected before AI therapy were rare (3 of 81 patients) compared with tissue collected after progression on prior AI therapy (12 of 21 patients).

Patients with ESR1 mutations are found to have poorer survival outcomes when treated with AI such as exemestane. Using ddPCR, Wang et al⁵²  found that in 3 of their 4 patients who had serial blood draws for longitudinal monitoring of ctDNA, the allelic frequency of ESR1 mutations in ctDNA correlated with resistance to endocrine therapy and disease progression. In a review of 2 CTC studies comprising ER-positive metastatic breast cancer patients on endocrine therapy, the authors⁵⁸  looked at the ESR1 mutation status of ctDNA as well as CTCs. The patients were divided into a baseline cohort of 43 patients started on first-line endocrine therapy and a progressing cohort of 40 patients who progressed under any line of endocrine therapy. They found that ESR1 mutations in ctDNA were more prevalent in the progressing cohort (42%) than in the baseline cohort (11%) (P = .04), strongly suggesting that they played a role in conferring endocrine resistance in metastatic breast cancer. Interestingly, ESR1 mutations in CTCs were not enriched in the progressing cohort (8%) when compared with the baseline cohort (5%) (P = .66). Serial monitoring of ctDNA for ESR1 mutations is therefore useful in guiding therapy. This has been demonstrated in studies where patients with ESR1 mutations were found to have improved PFS after changing their drug regime from exemestane to fulvestrant (an estrogen receptor down-regulator) or adding everolimus (an mTOR inhibitor) to exemestane.^53,54 ESR1 mutations in ctDNA can also be detected 6.7 months before clinical progression, providing ample opportunity for early therapeutic intervention.⁵⁵ 

Although the tissue biopsy is an invaluable diagnostic tool that provides histologic information and allows immunohistochemical profiling of hormone receptor and HER2 status, it has inherent limitations. It provides only a single snapshot of the tumor at one point in time in one body site and reflects only one portion of the tumor. Cancers, however, are often heterogenous tumors in both time and space and can undergo clonal evolution in response to treatment pressures. This is demonstrated by a study by Gerlinger et al,⁵⁶  who showed that portions taken from different parts of a primary tumor and its metastases showed intertumoral and intratumoral differences. Serial longitudinal tissue biopsies, which theoretically can map the phylogenetic progress of the tumor and overcome the problem of tumor heterogeneity, are, however, impractical in the clinical setting, as tissue biopsy is invasive with potential complications, and therefore not amenable to repetition. Tissue biopsy is also not possible in cases where the tumor is in a difficult-to-access site or if the patient is too ill for an invasive procedure. Tissue sampling of every metastatic deposit in a patient with multiple metastases is also not feasible. Tumor heterogeneity poses a challenge in determining the best course of therapy based on a single tissue biopsy, as it is likely to underestimate the complexity of the genomic landscape of the tumor.

Liquid biopsy, on the other hand, has the potential to reflect the overall genomic landscape of the tumor, both spatially across all metastatic sites and longitudinally across time. However, there are also challenges in translating this technology to the clinical setting. Circulating tumor DNA, for example, is often poor in quality, being highly fragmented and often present in low concentrations in the blood. There is a need to further optimize the yield of ctDNA by improving on ctDNA extraction methods or to improve the sensitivity of current DNA analysis assays to detect ctDNA at very low allele frequencies. At present, tissue biopsy remains the reference standard for molecular analysis of tumors in clinical guidelines and recommendations. For example, the National Comprehensive Cancer Network treatment guidelines⁵⁷  for non–small cell lung cancer do not recommend liquid biopsy to be used in lieu of tissue diagnosis, quoting a false-negative rate of 30% for ctDNA testing. It is therefore recommended that a negative liquid biopsy for EGFR mutations prompt a tissue biopsy for repeat molecular testing. Although PCR-based methods such as ddPCR, BEAMing, and amplification-refractory mutation system are highly sensitive and can detect ctDNA at very low allele frequencies of less than 1%, they can test only for specific, predetermined genetic alterations. Targeted, massive parallel sequencing or next-generation sequencing of multiple genes offers a wider scope of molecular interrogation of the tumor, but it requires dedicated bioinformatics support and can suffer from specificity issues when trying to distinguish low-allele-frequency variants (which is usually the case in liquid biopsy) from the intrinsic background noise due to DNA polymerase errors. There is also a lack of standardization of techniques that extends to every step of the liquid biopsy analysis, from the analyte (CTC or ctDNA), the sample in which the analyte is extracted (serum or plasma for ctDNA testing), and quantification of the analyte (spectrophotometry or fluorescence-based techniques for quantifying ctDNA) to the assay itself (ddPCR, amplification-refractory mutation system, BEAMing, or next-generation sequencing).

Potential applications of liquid biopsy in breast cancer span the entire course of the disease from (early) diagnosis to treatment for metastases. In the area of diagnostics, CTC and cfDNA levels and integrity have been shown to be able to discriminate patients with advanced and metastatic breast cancer from healthy subjects. However, their ability to pick up patients with early-stage disease or even in situ carcinoma remains uncertain. Such qualities must be evident before liquid biopsy can replace the current use of mammography to detect early disease, including in situ ductal cancers. Although highly sensitive assays can detect trace amounts of ctDNA, not all breast cancers will present with the same genetic mutations, and the same genetic mutation may be seen in many different cancer types, which may pose challenges to localizing the cancer.

Despite current limitations, liquid biopsy may play a greater role in the following areas:

1. Monitoring response to different treatment regimens and providing an estimate of the risk of progression. Studies have shown that CTCs and ctDNA are able to predict disease progression months before imaging manifestations.

2. Determining whether a patient has minimal residual disease after completing neoadjuvant chemotherapy or surgery. Such information will be useful in deciding whether the patient requires further adjuvant treatment and may potentially provide better stratification than current clinicopathologic criteria. The sequencing of ctDNA may also provide a better guide in determining subsequent treatment, as the molecular profile of ctDNA has been found to mirror relapsed/metastatic disease more closely than the primary tumor.

3. Highlighting the acquisition of mutations that confer resistance to endocrine therapies such as AI. This can help prompt an early change in treatment regimen before disease progression.

4. Lastly, liquid biopsy may be an alternative to tissue biopsy in patients where tissue biopsy is contraindicated.

More work is needed to further improve the technologies in isolating and analyzing tumor-derived materials in the blood. Among the myriad of technologies and testing procedures that characterize liquid biopsy currently, there is a need to determine the most optimal testing strategy and to establish standardized protocols for its use in the clinical setting. At present, liquid biopsy is at best an ancillary investigation that complements and builds on results from conventional tissue biopsies. Liquid biopsies in breast cancer have yielded cautiously promising results thus far and the outlook remains optimistic. With time, it is certainly possible that liquid biopsies may play an even greater role in the breast cancer clinic.

The authors would like to thank the Singapore General Hospital Translational Pathology Centre and Circulating Tumor Cell Centre of Research Excellence (CTC CoRE) (Singapore) for their provision of the figures.