Introduction
Female breast cancer has become the most commonly diagnosed malignancy worldwide according to global cancer statistics, with nearly 2.3 million new cases in 2020 [
1]. Compared with other malignancies, such as lung cancer, patients with breast cancer have relatively favourable treatment response rates and therefore longer 5-year survival rates (89.7%, by SEER dataset) [
2]. However, a significant percentage of breast cancer patients eventually develop incurable metastatic disease, which is characterized by an increasingly complex genomic landscape [
3]. Traditional Sanger sequencing cannot fulfil the requirements of rapidly generating comprehensive correlations between genomic variants and treatment response in a large number of clinical samples [
4]. Therefore, robust genomic profiling technologies should be applied to the characterization of advanced breast cancer (ABC) with the goal of defining genomic variation, evaluating prognosis and developing biomarkers to guide treatment selection.
Next generation sequencing (NGS) and molecular testing assays have evolved rapidly in recent years and are being successfully adopted in various clinical applications. When high coverage is achieved, targeted NGS enables the detection of low-frequency somatic mutations in heterogeneous tumour populations and circulating cell-free DNA (cfDNA) [
5]. A number of gene mutations, such as those in ESR1, PIK3CA, and TP53, have been found to be associated with drug resistance and worse prognosis in breast cancer [
6]. Given the strong detection capacity of NGS, it is now possible for clinicians to identify certain resistance-related mutations in a timely manner and make new therapeutic choices [
7]. Despite remaining challenges and limitations, efficient high-throughput sequencing technologies are leading to real and unprecedented benefits for the medical care of cancer patients [
8].
Circulating tumour DNA (ctDNA) is the small proportion of cfDNA that is released into the circulation from tumour cells and is detectable in a wide range of malignancies [
9]. Several studies have demonstrated high concordance between alterations detected in ctDNA and tumour tissue biopsy in ABC [
10‐
14]. Moreover, ctDNA can serve as a noninvasive biomarker for disease monitoring, predicting drug efficacy, and determining prognosis [
15]. The ctDNA fraction is defined as the proportion of ctDNA in cfDNA. In other types of cancer, such as metastatic prostate cancer, the ctDNA fraction has been shown to be a prognostic factor [
16]. Blood-based tumour mutation burden (bTMB) is a metric that represents the total number of mutations per coding area of a tumour genome, and is a metric that is frequently generated by NGS analysis [
17]. The magnitude and prognostic impact of bTMB is significantly different across solid tumour types [
18]. In triple-negative breast cancer (TNBC), bTMB is an emerging prognostic biomarker for immunotherapy [
19,
20]. However, it remains unclear in ABC patients whether the ctDNA fraction and bTMB correlate with different subtypes, treatment response or prognosis.
Herein, we investigated the molecular profiles of samples from Chinese ABC patient with a targeted 152-gene NGS panel. A comparison of the mutation profiles across matched tumour tissue and plasma ctDNA samples was conducted. Then, several genomic aberrations related to the treatment response were identified. Moreover, we explored whether the ctDNA fraction and bTMB could be used to evaluate treatment response and prognosis, respectively. Finally, we evaluated the clinical validity of ctDNA sequencing for longitudinally monitoring disease progression.
Discussion
In recent years, efforts have been made to investigate the genomic complexity of breast cancer [
2,
35‐
38]. However, these studies were either retrospective or lacked ctDNA-based dynamic disease monitoring and clinical management. Breast cancer is a heterogeneous disease that varies across different clinical subtypes, populations and geographical regions, all of which exhibit variable therapeutic responses [
39,
40]. With the increasing incidence rate of breast cancer worldwide, it is critical to identify region-specific genomic variants and relevant therapeutic targets. In this prospective study, a novel 152-gene targeted NGS panel was used to evaluate the mutation landscape of ctDNA collected from 141 Chinese ABC patients. In addition, we profiled drug response-related genomic variants, evaluated the prognostic value of the ctDNA fraction and bTMB, and assessed the ability of ctDNA to monitor disease progression.
In this study, molecular profiling of ABC revealed genomic variants consistent with several previous reports, and TP53 (44.0%) and PIK3CA (28.4%) were the two most commonly altered genes [
2,
35,
38,
41,
42]. TP53 is a well-characterized tumour suppressor gene that acts as a regulator of key cellular processes involved in controlling cell proliferation and maintaining genomic stability [
43,
44]. Almost all hallmark features of cancer are impacted by the functions of the TP53 protein and correlate with genomic alterations in TP53 pathways [
45]. In other words, the occurrence of many tumours requires TP53 mutations as a prerequisite. Indeed, TP53 was shown to be frequently altered in previous sequencing studies, with frequencies ranging from 38.24 to 74.11% [
2,
35,
38,
41,
42,
46]. In particular, the mutation rates of TP53 were 38.24%, 41.67%, 64.1%, 43.27%, and 47.0% in Chinese populations [
2,
35,
41,
42,
46]. Therefore, it is not surprising to find a high mutation rate of TP53 in this study. However, in regard to individual situations, more efforts are needed to reveal the specific role of TP53. HER2 + patients showed a significantly higher TP53 mutation rate than HR + patients. TNBC patients harbouring TP53 and PIK3CA mutations showed a significantly longer OS than those without these alterations. There is clear evidence that TP53 is associated with poor prognosis of HR + disease, while its clinical significance in HER2 + and TNBC patients remains controversial [
47]. Considering the high degree of heterogeneity of TP53 mutations, further large-scale sequencing studies are needed to address its clinical impact and association with response to therapy in distinct subtypes. PIK3CA is the most frequently mutated oncogene in human cancers [
48,
49]. The continuous activation of AKT that is induced by PIK3CA mutations can promote the growth and transformation of mammary epithelial cells and may explain the high mutation rate of PIK3CA [
50]. The approval of alpelisib has rendered the detection of PIK3CA mutations clinically actionable, as everolimus, an mTOR inhibitor, does not directly target PIK3CA mutations. Thus, HR + /HER2- ABC patients harbouring PIK3CA mutations can be offered another targeted therapy choice after endocrine resistance occurs [
51]. For TNBC, the prognostic impact of PIK3CA mutations remains debatable, and evidence for the clinical application of PIK3CA inhibitors is still lacking [
52,
53]. The BELLE-4 study tested the efficacy of buparlisib (a pan-PIK3CA inhibitor) in combination with placebo or paclitaxel in HER2- ABC patients [
54]. Approximately 25% of the patients were TNBC patients who showed a worse outcome with the addition of burpalisib than placebo (5.2 vs. 9.3 months; hazard ratio 1.86, 95% CI 0.91–3.79). Nonetheless, combination approaches targeting PIK3CA with other targeted drugs, such as androgen receptor and CDK4/6 inhibitors, may provide a potential therapeutic direction for specific subsets of TNBC patients [
53]. Several ESR1 hotspot mutations within the ligand-binding domain, including D538G and Y537S/N/C, were also detected at a lower frequency (4%). Given established treatment-associated patterns of acquired mutations in ESR1, it is not surprising that the mutation rate was not high in the setting of first-line treatment [
55].
ERBB2 was the gene with the most frequent occurrence of CNVs and had an amplification rate of 22.7% (32/141) in the entire cohort and 70.5% (31/44) in the HER2 + population. Compared with the study by Davis et al., the amplification rate in ERBB2 was higher in our cohort (70.5% vs. 44.0%) [
38]. This difference may be explained by two factors. The first factor is the patient population. We enrolled Chinese patients in this study, while the study by Davis et al. was conducted in the United States. This explanation is supported by a previous study by Xiao et al. which also revealed a high amplification rate of ERBB2 (90.3%, 158/175) in HER2 + Chinese patients [
46]. We (PredicineCARE, developed by Huidu Shanghai Medical Sciences Ltd.) and Davis et al. (Guardant360 assay, Guardant Health, Inc., Redwood City, CA) used different NGS panels to detect variants. The higher detection rate may indicate the improved ability of copy number detection from our NGS assay. Notably, one TNBC patient presented with an ERBB2 CNV, indicating tumour heterogeneity and the potential need to reassess HER2 status by ctDNA during the clinical management of TNBC [
56]. These findings underscore the feasibility of tracing ERBB2 CNVs with liquid biopsy during disease progression and introduce the possibility for applying anti-HER2 therapy. In the analysis of drug response-related variants, the detection of individual SNVs or CNVs showed no correlation with drug response. However, the combination analysis of SNVs with CNVs revealed potential associations between the PIK3CA/TP53 and FGFR1/2/3 variants and drug resistance in HR + and HER2 + patients, respectively. These findings suggest that integrating multiple genetic alterations could improve the identification of treatment resistance-related mechanisms compared to measuring a single alteration [
57].
CtDNA levels can dynamically reflect the tumour burden of a patient and predict disease progression prior to imaging [
58]. Moreover, several studies have revealed the prognostic role of the ctDNA fraction in ABC patients [
58‐
60]. Stover et al. found that a ctDNA fraction of ≥ 10% correlated with a worse metastatic OS (6.4 vs. 15.9 months) [
59], while a cut-off of 0.5% ctDNA (MAF) was regarded as prognostic for both PFS and OS in another report [
58]. In the present study, a cut-off of 0.174 was used to discriminate patients with high or low ctDNA fractions. Based on this grouping, we found a significant difference in PFS between TNBC patients with high vs. low ctDNA fractions (2.9 vs. 7.3 months,
P = 0.005). However, no significant differences in PFS were observed in other subtypes. This could result from subject variations in disease onset, diagnosis, or intervention since this study contained all subtypes of patients and multiple treatment regimens. Moreover, this heterogeneity could result in opposite associations between the ctDNA fraction and prognosis among different subtypes. In addition to its prognostic role, the ctDNA fraction can reflect the panel sensitivity, as the panel sensitivity decreased if samples displayed low tumour fractions. Therefore, we evaluated the correlation between the ctDNA fraction and the number of mutations in all baseline plasma samples using the Pearson correlation coefficient, and the results suggested a significant positive correlation (R = 0.56,
P < 0.05; Additional file
4: Fig. S4A). The number of mutations in samples with ctDNA fractions of 0–1% was significantly lower than that in samples with ctDNA fractions of 1–5% and > 5% (Additional file
4: Fig. S4B). These findings also illustrate the importance of improving sequencing sensitivity.
TMB is a measure of the somatic mutation frequency and mutation accumulation process of a tumour. In tumorigenesis, some nondriver mutations will lead to the activation of an antitumour response by generating neoantigens that are recognized by the immune system [
61]. Consequently, TMB is regarded as a biomarker that is associated with the response to immune checkpoint inhibitors in several types of cancer [
17]. Despite being a promising pancaner tool, the use of TMB is limited by several key remaining issues, including variable clinical impacts across cancer types and the lack of a standardized cut-off value [
17]. In this study, a cut-off of 43.3 was adopted to define high vs. low bTMB, and patients with high bTMB showed a shorter PFS (5 vs. 10 months,
P = 0.05) and OS (40.6 vs. 70 months,
P = 0.02) than those with low bTMB. These findings are consistent with previous reports that suggested that TMB was a prognostic factor for poor outcome [
19,
20]. Even so, the association between bTMB and prognosis should be interpreted with caution because of differences at the individual level, which may reduce its usability in practical situations. Indeed, in our further analyses based on subtypes, these associations were significant only in HER2 + (PFS 5 vs. 20 months,
P = 0.009) and TNBC patients (PFS 3 vs. 7.3 months,
P = 0.05). Hence, more research investigating the clinical utility of ctDNA-derived biomarkers in ABC is needed to address this issue. Interestingly, HER2 + samples displayed significantly elevated bTMB. Since recent evidence has indicated elevated programmed cell death-ligand 1 and tumour infiltrating lymphocyte expression in HER2 + disease [
62], further studies exploring the possibility of implementing immunotherapy in such patients are warranted.
Previous sequencing studies have revealed the concordant detection of DNA variants across matched sets of ctDNA and tumour tissue [
10‐
14]. Davis et al. described the genomic landscape of ctDNA in 255 ABC patients. They found an agreement of 79–91% between 105 pairs of ctDNA and tissue, with alterations in PIK3CA and TP53 showing moderate concordance (kappa = 0.5513 and 0.5809, respectively) [
38]. In this study, we also found moderate concordance for the SNVs in PIK3CA and TP53 (kappa = 0.61 and 0.41, respectively). In addition, a high concordance was revealed for ERBB2 CNVs (kappa = 0.77), which was consistent with a report by Zhou et al. [
35]. These findings suggest the equivalent utility of ctDNA with tissue sequencing in developing targeted therapeutic approaches for HER2 + patients.
We also performed longitudinal monitoring of disease progression to observe the dynamic changes in gene mutations and amplifications in ctDNA. The genomic variants detected in ctDNA at baseline and at disease progression were compared. Overall, the plasma ctDNA from samples collected at the time of clinical progression in HR + and TNBC patients had acquired genomic variants, indicating clonal and subclonal responses to treatment. In contrast, genomic variants detected at baseline in HER2 + patients were not detected at progression. In HR + patients, newly acquired ESR1 mutations at the time of disease progression were observed in four patients, which is consistent with previous findings that the emergence of ESR1 mutations is associated with the development of endocrine resistance [
55]. Other emerging genomic variants included FGFR amplification and CDKN2A deletion events. Aberrant FGFR signalling has been identified as a mechanism that drives tumour growth and promotes angiogenesis [
63]. The amplification of FGFR, which occurs in approximately 10–16% of HR + patients [
60], has been shown to mediate endocrine resistance [
64]. CDKN2A (cyclin-dependent kinase inhibitor 2A) is a tumour suppressor gene that was first reported in 1993, and it is negatively regulated by the CDK4/6/RB pathway [
65]. Despite rare alterations of this gene in breast cancer (approximately 5.8%) [
66], CDKN2A deletions were reported to be associated with poor outcomes in luminal B ER + patients [
67]. Importantly, the deletion of this gene implicates CDK4/6 as a therapeutic target to some extent [
68]. Taken together, the acquired FGFR amplification and CDKN2A deletion variants in progression samples may explain the disease progression in these HR + patients.
There are several limitations that should be mentioned. First, the sample size was relatively small, and only a subset of patients had matched blood and tissue samples. These factors may decrease the concordance of detected genomic variants. Second, the samples were collected in a single centre and sequenced by a single NGS assay. Hence, the results from this study need to be validated in other research centres using distinct sequencing assays. Third, since the NGS panel applied in this study only included 152 genes, its ability to detect tumour burden is theoretically inferior to genome-wide tests. Thus, the association between bTMB and prognosis needs more verification. Fourth, the sensitivity of this panel may decrease if samples display a lower ctDNA fraction, and the ctDNA fraction of all samples in this study varied greatly (Additional file
5). Finally, although treatment response-related variants were analysed, the effects of drug interventions guided by ctDNA profiling were not studied. Further studies will be required to evaluate the impact of ctDNA profiling-guided drug interventions on patient outcomes to establish the clinical utility of NGS liquid biopsy in the management of ABC patients.
In conclusion, this prospective study profiled the mutation landscape of Chinese ABC patients who underwent first-line standard treatment and demonstrated the clinical validity of ctDNA-based genomic analysis. Further studies are warranted to investigate the relationship between drug interventions and genomic changes.