Concordance analysis of non-invasive techniques for PIK3CA and ESR1 mutations in advanced HR+/HER2- breast cancer: the CANIPE study

Abstract

Hormone receptor-positive (HR+) and HER2-negative (HER2−) breast cancer represents ~70% of breast tumors. Accurate detection of PIK3CA and ESR1 mutations is crucial for guiding targeted therapies, particularly in advanced disease. Liquid biopsy testing is increasingly recommended, especially for ESR1 mutation, but standardized genotyping methods remain limited. The CANIPE study, a prospective, multicenter Spanish trial, evaluated mutation prevalence in 106 baseline blood samples. PIK3CA mutations were identified in 41.9% by ddPCR droplet digital PCR (ddPCR) and 42.8% by targeted next-generation sequencing (NGS), while ESR1 mutations were detected in 7.8 and 9.5%, respectively. A concordance analysis between ddPCR and NGS was performed in 62 paired samples, demonstrating excellent agreement (Kappa >0.85) and strong correlation in variant allele frequency for PIK3CA, confirming the reliability of the methods. Baseline circulating free DNA levels were prognostic for survival. These results, derived from real-world patient samples, support ddPCR and NGS as robust, complementary tools for mutation detection and therapeutic decision-making.

Introduction

The combination of cyclin-dependent kinase 4/6 inhibitors (CDK4/6i) with endocrine therapy (ET) is the standard first-line treatment in patients with hormone receptor-positive (HR+), HER2− negative advanced breast cancer (BC), which represents ~70% of BC patients¹. Several clinical trials (PALOMA, MONARCH, and MONALEESA) have shown that combined therapies have improved progression and/or overall survival rates^2–6. However, 20% of patients do not respond to standard therapies, and those who initially respond eventually acquire therapy resistance^7,8. In this context, the choice of second-line treatment often depends on the availability of therapies targeting mutations in PI3KCA, ESR1, AKT, or PTEN.

In the latest revisions to clinical guidelines, partly due to the approval of therapies for PIK3CA inhibition or ESR1 blockade^9–12, the analysis of these mutations in tissue or liquid biopsy is recommended to improve the clinical management of these patients, and specifically for ESR1 testing, the study of cfDNA represents the preferred option^13,14. However, few studies directly compare the efficacy of different methods for detecting genetic variants, particularly those using liquid biopsy for identifying PIK3CA or ESR1 mutations. Furthermore, reliable data on PIK3CA or ESR1 status in ctDNA from real-life cohorts undergoing this combined therapy are currently scarce, as most studies have been conducted with selected patients and a single CDK4/6 inhibitor¹⁵. The CANIPE study addresses this gap as a prospective, multicenter Spanish cohort of patients with HR+/HER2- advanced BC who initiate first-line therapy. The study compares two approaches for evaluating the prevalence of PIK3CA and ESR1 mutations: digital droplet (dd)PCR and the broader NGS AVENIO panel. ddPCR is known for its high sensitivity and specificity in detecting hotspot mutations. AVENIO offers multiplexed mutation profiling across 77 genes, including PIK3CA, ESR1, and other relevant genes such as AKT, PTEN, MTOR, or BRCA.

The CANIPE study has both short- and long-term objectives. In the long term, it aims to evaluate the evolution of the most frequent PIK3CA and ESR1 mutations using liquid biopsy in patients with advanced luminal breast cancer treated with CDK4/6 inhibitors and hormonal therapy, using ddPCR. In the present work, the concordance between AVENIO and ddPCR technologies for detecting these mutations was assessed, and mutation detection rates in a real-world cohort were analyzed. Overall, it seeks to define best practices for mutation analysis and inform personalized treatment strategies.

Results

CANIPE cohort overview

From May 2023 to December 2024, 106 advanced HR+/HER2− BC patients were recruited to the CANIPE study in 11 Spanish Hospitals. The engaged patients comprised 105 females and one male, with a mean age of 63.10 years (SD = 12.78; range: 33.26–88.6). The most frequent site of metastatic involvement was the bone (35.85%), followed by visceral metastases (26.42%), mixed patterns (27.36%), and lymph node involvement (7.55%). Additionally, 51.89% showed multiple metastatic locations. Molecular subtyping by IHC and FISH revealed that most tumors were Luminal B-like (47.62%). Luminal A-like subtypes represented 33.33% of cases, while 19.05% were undetermined. Adjuvant regimens varied, with chemotherapy alone (5.66%), combined with ET (36.79%), or ET alone (13.21%), while 30.19% of patients had not received prior adjuvant therapy. In first-line endocrine therapy, Letrozole was the most commonly used (68.57%), followed by Fulvestrant (20.95%). Among CDK4/6 inhibitors, Ribociclib was the most frequently administered (39.05%), followed by Palbociclib (33.33%) and Abemaciclib (27.62%). Initiation of CDK4/6i therapy varies among patients. Although therapy was initiated at the time of metastasis diagnosis in 21 patients, the remaining 84 cases experienced a mean treatment delay of 27.78 days (SD = 29.65). Interestingly, 30.19% had the novo metastasis. At the time of metastatic diagnosis, 51.89% of patients presented two or more metastatic sites. The mean time of progression to CDK4/6i therapy was 10.7 months (SD = 4.8 months), while the mean follow-up was 12.8 months (SD = 13 months). At first CT scan, 70 patients evaluated showed stable disease (24.53%), partial response (16.98%), or complete response (6.60%), while 19 of 70 showed radiological progression (RECIST criteria). Twenty-three patients developed disease progression, and 11 patients died during follow-up. The clinical data were summarized in Table 1.

Table 1.

Clinical data

Category	n	(%)
Age at baseline (range 33–88 years)
<63	53	50.00
≥63	53	50.00
Subtype PT
Luminal A-like	35	33.02
Luminal B-like	50	47.17
Luminal-like(undetermined)	21	19.81
Hormonal status
ER+	105	99.06
PR+	76	71.70
ECOG
0	41	38.68
1	29	27.36
>2	13	12.27
Metastasis de novo
Yes	32	30.19
No	73	68.87
Metastatic location
Visceral	57	53.77
Others	46	43.40
Number of metastasis
<2	45	42.45
≥2	55	51.89
Therapy (CDK4/6i)
Abemaciclib	29	27.36
Palbociclib	35	33.02
Ribociclib	41	38.68
Therapy (CDK4/6i + ET)
CDK4/6i + Letrozole	72	67.92
CDK4/6i + Fulvestrant	22	20.75
CDK4/6i + Others	11	10.38

Any differences from 100% are due to unavailable data.

Association between cfDNA levels and tumor characteristics

Baseline cfDNA showed a mean concentration of 53.90 ± 159.05 ng/mL (range: 6.24–1534 ng/mL) (n = 106). High cfDNA (based on the median value) was not associated with ECOG performance status or the number of metastases at diagnosis, but was strongly associated with the site of metastasis. Thus, patients with visceral metastasis showed higher cfDNA values (p value = 0.0011, Chi-square test), particularly in those with both visceral and bone metastasis. Survival analysis was performed, including only patients with at least 12 months of follow-up (n = 48). Median PFS could not be estimated in either group because the 50% event rate was not reached. However, high baseline cfDNA levels may be associated with a higher risk of progression (p value = 0.04, by log-rank test, n = 48). Accordingly, high baseline cfDNA levels were associated with shorter OS (undefined vs 579 days, p value = 0.0008 by log-rank test, n = 48) (Fig. 1A, B). In the high-cfDNA group, 7 of 25 patients died, while in the low cfDNA group, no patients died. Next, a ROC analysis was performed, yielding an area under the curve (AUC) of 0.84 (95% CI: 0.71–0.93). The optimal threshold of cfDNA concentration, identified using the Youden index, is ~21 ng/mL, yielding 100% sensitivity and 71.4% specificity, indicating strong performance in minimizing false negatives. This supports its further investigation as a candidate prognostic biomarker (Fig. 1C).

Fig. 1. cfDNA levels association with clinical data.

A, B Kaplan–Meier plot stratified by cfDNA levels (based on median value): high (purple line) or low (blue line) for PFS (A) and OS (B) (n = 48). Statistics were performed by the log-rank test. C Receiver operating characteristic (ROC) curve for cfDNA concentration at baseline as a predictor of death. The threshold value shown reflects the exact Youden index output.

PIK3CA and ESR1 mutational profile in advanced HR+/HER2− breast cancer patients by ddPCR and AVENIO

To explore PIK3CA and ESR1 most frequent mutations in the CANIPE cohort, a multiplex ddPCR and AVENIO panel was applied. Genotyping of these genes by ddPCR was successful in 105 cases for PIK3CA and in 102 cases for ESR1. The analysis with the AVENIO panel was performed on a subset of the cohort, 63 samples (Fig. S2A), and all tests were successfully assessed.

The PIK3CA mutational status analysed by ddPCR was assessed using two multiplex assays that cover different variants (see M&M). 41.9% (44/105) of patients had at least one mutation in the analysed variants. Of the 44 patients with PIK3CA mutations, 43 were positive for Panel 1 (p.Glu542Lys, p.Glu545Lys, p.His1047Leu, p.His1047Arg, and p.His1047), and one patient showed mutations also in Panel 2 (p.Arg88Gln, p.Asn345Lys, p.Cys420Arg, p.Glu545Ala, p.Glu545Gly, p.Gln546Lys, and p.Glu545). In the case of the ESR1 gene, 7.8% (8 out of 102) of patients depicted a mutation, six patients in Panel 1 (p.Leu536Arg, p.Asp538Gly, p.Glu380Gln, and p.Tyr537Cys), five cases in Panel 2 (p.Tyr537Ser, p.Tyr537Asn, and p.Ser463Pro), respectively, and three patients showed positivity with both panels (Fig. 2A).

Fig. 2. Mutational profile of advanced HR+/HER− breast cancer patients.

A Detection frequency of PIK3CA and ESR1 mutations by ddPCR (n = 105 and 102, respectively) and AVENIO (n = 63). B, C Frequency of PIK3CA (B) or ESR1 (C) variants detected with AVENIO panel. D Frequency of mutated genes (SNVs) studied with the AVENIO panel. E Heatmap depicting co-mutations per patient observed by AVENIO.

After analysis with the AVENIO panel, 42.8% of patients had a PIK3CA mutation, while six out of 63 (9.52%) had an ESR1 mutation (Fig. 2A). Four patients had mutations in both PIK3CA and ESR1. For PIK3CA, the most frequent variants were p.His1047Arg and p.Glu542Lys, accounting for 32.14% each (Fig. 2B). For ESR1, the most frequently represented variants were p.Asp538Gly and p.Glu380Gln, with frequencies of 33 and 22%, respectively, as depicted in Fig. 2C.

AVENIO panel identified SNVs in other genes (Fig. 2D). Thus, 39 out of 77 genes included in the panel were found to be mutated in this cohort. PIK3CA was the most frequently altered gene, followed by TP53 (18.75%) and ESR1 genes. BRCA1, DPYD, or GNAS mutations were found in four patients, while CTNNB1 or ERBB2 mutations were detected in three patients. The other mutations identified were anecdotal (in three, two, or just one patient). Co-mutations were identified in 30 patients, while no mutation was found in 14 patients (Fig. 2E). Additionally, copy number variations (CNVs) were observed for EGFR (present in one patient) and MET and ERBB2 (present in two patients each).

The presence of PIK3CA or ESR1 mutations was not associated with metastasis localization or the number of affected sites. ESR1 mutations appear mostly in Luminal B-like subtype (p value = 0.0042, Fisher’s exact test) (Fig. 3A), while PIK3CA mutations were independent of the subtype (Fig. 3B). In the CANIPE cohort, PIK3CA mutation status was associated with the overall distribution of radiological response categories (complete response: CR; partial response (PR); stable disease (SD); and progression disease (PD)) at first CT evaluation (p = 0.022, Fisher’s exact test) (Fig. 3C). Notably, complete responses were observed exclusively in PIK3CA wild-type tumors. Accordingly, if response was considered as CR versus non-CR, the absence of PIK3CA mutations was significantly associated with CR (p = 0.017, Fisher’s Exact Test). ESR1 mutational status did not associate with response (p = 0.25, Fisher’s exact test). In addition, PIK3CA mutations were observed more frequently in patients with de novo metastasis and less frequently in those with recurrence ≥5 years after the initial diagnosis, suggesting a possible association between PIK3CA alterations and more aggressive or early-onset forms of the disease (p value = 0.045, Chi-square test). Besides, a trend for shorter outcome PFS and OS in PIK3CA-mutated tumors was observed (PFS: undefined vs. 506 days, p = 0.14 by log-rank test; OS: undefined, p value = 0.06 by log-rank test, n = 48) (Fig. 3D, E). Notably, contingency analysis reveals that patients with low cfDNA levels are associated with the absence of PIK3CA mutations (p value = 0.03), whereas no association was observed for ESR1. A moderate positive linear correlation between cfDNA and mutation VAF in cfDNA (cor = 0.28) (CI 95% = 0.10–0.45) (p value = 0.0027) was detected, suggesting that patients with higher cfDNA levels tend to have a greater tumor-derived DNA burden, which is reflected in higher mutation VAF values.

Fig. 3. Mutational status association with clinical data.

A, B Frequencies for the ESR1 (A) (n = 84), and PIK3CA (B) (n = 85). mutational profile and the molecular subtype. C Frequencies for the PIK3CA mutational status based on the clinical outcome at the first CT scan (CR complete response, PR partial response, SD stable disease, P progression disease). Statistical analysis was performed using Fisher’s exact test. D, E Kaplan–Meier plot stratified by patients showing PIK3CA mutated (MUT, in red) or wildtype (WT, in blue) for PFS (D) and OS (E) (n = 48). Statistics were performed by the log-rank test.

A multivariable Cox model was subsequently performed, including cfDNA levels and PIK3CA status. For overall survival, the effect of high cfDNA could not be reliably estimated due to the small number of events (8 of 48 cases) and the lack of separation in risks (all in the high-cfDNA group). PIK3CA mutation status was not independently associated with OS (HR 2.03; 95% CI: 0.39–10.46; p value = 0.40). Furthermore, including PIK3CA mutation status did not significantly improve the model based solely on cfDNA (LRT, p value = 0.37).

Concordance analysis between ddPCR and AVENIO for the detection of PI3KCA or ESR1 mutations

Paired mutational profile data for PIK3CA and ESR1 status, obtained by ddPCR and AVENIO, were available for 62 metastatic BC HR+/HER2− patients. The mutation frequencies observed in this subset differed slightly from those seen in the whole cohort. Using ddPCR, PIK3CA mutations were detected in 45.16% of patients, and ESR1 mutations in 6.67%. With AVENIO, PIK3CA, or ESR1 mutations were found in 43.54 and 8.3% of the analysed patients, respectively (Fig. S2B). The median VAF for PIK3CA was 3.36% ± 8.83 using the AVENIO assay and 2.93% ± 7.2 with ddPCR. For ESR1, the median VAFs were 0.08% ± 0.4 (AVENIO) and 0.056% ± 0.32 (ddPCR).

The concordance analysis revealed three discordant cases among the 62 patients for PIK3CA mutation, based on the established criteria for classifying each patient as mutated across the technologies (see M&M). PIK3CA mutations were detected by ddPCR in two patients, with VAFs of 0.05 and 0.06%, respectively. One patient was detected by AVENIO with a VAF of 0.18%.

The concordance analysis between the two technologies was first assessed using Cohen’s Kappa coefficient. The Kappa value obtained was 0.902, indicating excellent agreement between the raters, with a z-value of 7.11 (p value <0.001). To explore the similarity of both measures within the same subject, an Intraclass correlation (ICC) analysis was performed using the PIK3CA VAF% measurements, yielding an ICC value of 0.968 (95% CI: 0.948–0.981), indicating excellent agreement (n = 62) (Fig. 4A). The F-test (F = 64.1, p < 0.001) confirms significant evidence of agreement. To test for the presence of systematic bias and any differences observed between the two methods, a Bland–Altman plot was applied, showing that 96.77% of the cases fall within the limits of agreement (marked in green) (Fig. 4B). A slight trend is observed for VAF values greater than 20%, with the two highest points falling outside the corresponding limits. Additionally, an average positive bias (red line above zero) indicates that AVENIO tends to yield higher VAF values than ddPCR. We applied the Passing-Bablok Regression analysis to evaluate whether AVENIO and multiplex ddPCR exhibit a linear relationship or show a systematic or proportional bias. For PIK3CA VAF%, the analysis showed a slope of 0.956 (95% CI: 0.91–1) and an intercept of 0 (95% CI: 0–0). This indicates an excellent agreement and no constant bias between the two methods. A slight proportional bias was observed, with the slope slightly below 1, indicating that ddPCR tends to report slightly lower values than AVENIO (mean −0.43% ± 1.99), as previously detected. Additionally, Pearson´s r = 0.989 indicates a very strong positive linear correlation between multiplex ddPCR and AVENIO panel (Fig. 4C).

Fig. 4. Concordance analysis of PIK3CA and ESR1 mutations detected by AVENIO and multiplex ddPCR.

A VAF% measurements for AVENIO and multiplex ddPCR for PIK3CA (n = 62). B Bland–Altman Plot for PIK3CA analysed by AVENIO and multiplex ddPCR. Green lines delimit the mean of the differences ±1.96 standard deviation. The red-dashed line represents the mean difference (bias) between the two methods (0.43%). Red dots indicate VAF % = 0. C Passing-Bablok regression analysis for PIK3CA analysed by AVENIO and multiplex ddPCR. The blue line represents the slope. Orange circles represent paired VAF measurements from multiplex ddPCR (y-axis) and AVENIO (x-axis). D VAF% measurements for AVENIO and ddPCR for ESR1 (n = 60). E Bland–Altman plot for ESR1 analysed by AVENIO and multiplex ddPCR. Green lines delimit the mean of the differences ±1.96 standard deviation. The red-dashed line represents the mean difference (bias) between the two methods (0.03%). Red dots indicate VAF % = 0. F Passing-Bablok regression analysis for ESR1 analysed by AVENIO and multiplex ddPCR. The blue line represents the slope. Orange circles represent paired VAF measurements from multiplex ddPCR (y-axis) and AVENIO (x-axis).

One case showed discordant ESR1 status results between the two methods. In this case, AVENIO detected an ESR1 mutation with a VAF of 0.26% while ddPCR failed to detect any mutation. Cohen’s Kappa coefficient obtained was 0.88, indicating a high level of agreement between the two methods, with a z-value of 6.87 (p value <0.001). Following the same approach as with PIK3CA, different statistics were applied. Thus, the ICC value was 0.706 (CI 95%: 0.552–0.813), suggesting a moderate to good agreement between VAF% for ddPCR and AVENIO (n = 60) (Fig. 4D). F-test = 5.77, (p < 0.001) indicates that the observed correlation is statistically significant. Similarly, a generally good agreement is observed in the Bland–Altman plot, with only one value outside the limits of agreement. A slight bias (0.029) could be underestimating the VAF% values by ddPCR (Fig. 4E). The Passing-Bablok Regression test showed a slope of 0.928, suggesting that ddPCR tends to measure slightly lower than AVENIO (mean −0.029% ± 0.28). At the same time, no significant consistent bias was found (intercept = 0). Pearson’s r test (r = 0.72) indicates a moderate to strong positive linear correlation between the AVENIO and multiplex ddPCR methods (Fig. 4F).

According to the reference method (AVENIO), the multiplex ddPCR demonstrates excellent diagnostic performance for detecting PIK3CA mutations, with a sensitivity of 96.3%, indicating that it correctly identifies the majority of true mutation-positive cases. Additionally, its 100% specificity confirms the absence of false positives. The positive predictive value (PPV) was 100%, meaning all samples that tested positive truly harbored the mutation. The negative predictive value (NPV) of 97.2% indicates that nearly all negative results were accurate. For ESR1 mutation detection, ddPCR showed 80% sensitivity and 100% specificity, with no false positives, as supported by a PPV of 100%. The NPV of 98.2% suggests that the technique reliably identifies negative cases, though a small proportion of mutation-positive cases may go undetected.

Single ddPCR genotyping analysis refines concordance to AVENIO

All previous analyses for PIK3CA and ESR1 were performed using multiplex ddPCR data. To evaluate whether multiplexing accurately reflects the specific PIK3CA variant, we reanalysed using single-target genotyping in 18 samples with cfDNA available. In this case, only the mutation detected by AVENIO was analyzed in each sample by ddPCR. Single-genotyping was positive for the specific mutation in all cases, with VAF% values closer to those obtained by multiplexing (Fig. 5A).

Fig. 5. Concordance analysis of PIK3CA detected by AVENIO and single genotyping ddPCR.

A VAF% measurements for AVENIO and single genotyping ddPCR (n = 18). B, C Passing-Bablok regression analysis: Multiplex ddPCR and AVENIO (B) or single genotyping ddPCR and AVENIO (C) (n = 18). The blue line represents the slope. Orange circles represent paired VAF measurements from multiplex ddPCR (y-axis) and AVENIO (x-axis).

We performed concordance analysis on these samples (n = 18) in multiplex or single genotyping compared to AVENIO. For the multiplex ddPCR, Passing-Bablok regression yielded a slope of 0.923 (CI 95%: 0.738–0.995) with an intercept of 0.035 (CI 95%: −0.177 to 0.432), suggesting the absence of constant bias but with a significant proportional bias of ddPCR regarding the reference method (AVENIO) (mean −1.45). However, when single-genotyping ddPCR was considered, the slope was 1.021 (95% CI: 0.928–1.06), with an intercept of 0.0146 (95% CI: −0.166 to 0.075), indicating no significant bias in either the intercept or the slope. Pearson´s coefficient for both analyses was 0.987 and 0.997, respectively (Fig. 5B, C).

Discussion

PIK3CA and ESR1 gene mutations occur frequently in HR+/HER2- BC^16–18. PIK3CA mutations have been reported to activate the PI3K/AKT/mTOR pathway, which enhances cell growth and survival and also leads to endocrine resistance. Approximately 40% of HR+/HER2− breast cancers harbor a pathogenic PIK3CA mutation. The SOLAR-1 trial demonstrated that targeted therapy with alpelisib, a selective PI3Kα inhibitor, and fulvestrant improved PFS in PIK3CA-mutated cases¹⁹. Furthermore, the CAPItello-291 trial demonstrated that the AKT inhibitor capivasertib, when combined with fulvestrant, is particularly effective in patients with PI3K/AKT/PTEN pathway alterations. ESR1 mutations, which typically occur in the ligand-binding domain, result in ligand-independent constitutive activation of the estrogen receptor and are commonly observed in patients with metastatic disease following aromatase inhibitor therapy²⁰. Results from the EMERALD trial have led to the FDA approval of elacestrant, an oral selective estrogen receptor degrader (SERD), representing a significant advance in re-treating tumors with ESR1 mutations. The SERENA-6 trial results indicate that in patients with emerging ESR1 mutations, switching to camizestrant plus a CDK4/6 inhibitor significantly prolongs progression-free survival compared to continuing an aromatase inhibitor combination^18,21). Additionally, several studies are underway, including novel SERDs in combination with PIK3 inhibitors, such as the INAVO-121 (NCT05646862) trial. Since recent clinical guidelines emphasize the importance of obtaining a mutational profile of these genes to inform second-line therapies²², there is an emergent need for validated tools for mutation monitoring in these patients.

Liquid biopsy (ctDNA) is increasingly used to identify mutations to aid in therapeutic decisions in real time^23–26. The CANIPE study provides valuable real-world data on the prevalence of PIK3CA and ESR1 mutations in a Spanish HR+/HER2− BC cohort, analyzing plasma cfDNA by two different approaches. The frequencies of PIK3CA and ESR1 mutations detected by ddPCR and AVENIO at baseline in this multicentric study were similar to those previously reported, ~40%^27,28 and 6–10%^20,29, respectively. Although our data show a slightly higher percentage, these values are likely not directly comparable due to differences in technologies and/or sample sources. In this study, a VAF threshold of >0.1% was used to validate SNVs detected with AVENIO, based on the assay’s documented analytical performance and previous reports demonstrating high sensitivity at low VAF levels^30,31. In addition, the multiplex ddPCR assay developed to analyze PI3KCA and ESR1 mutations in cfDNA demonstrated high accuracy. As expected, PIK3CA and TP53 were the most frequently mutated genes, followed by ESR1, in line with the previous literature^27,32,33. Interestingly, the ERBB2 amplification (CNV) was detected in two patients, suggesting a potential molecular subtype switch. The CANIPE study aims to compare the AVENIO panel and a multiplex ddPCR strategy for detecting PIK3CA and ESR1 mutations in cfDNA. While multiplex ddPCR offers a rapid and cost-effective response, AVENIO enables the identification of specific variants and mutations that may serve as actionable therapeutic targets. However, for PIK3CA and ESR1 mutations, both technologies showed similar results. The high Kappa coefficient indicates strong agreement between AVENIO and ddPCR, for both mutated and non-mutated samples. Also, as observed in the Bland–Altman Plot, the two methods give overall similar VAF% measurements. For the PIK3CA mutation, two samples within the concordance limits have the highest VAF%. The outliers occurring at the highest VAF values suggest greater inter-method variability at higher allele fractions. Additionally, Passing-Bablok regression showed a high correlation and no constant bias for PIK3CA and ESR1. It was determined that ddPCR significantly underestimates VAF% compared to AVENIO, though the difference is negligible. The observed increase in VAF% is proportional: as values from the AVENIO method increase, those from ddPCR tend to increase similarly, following a nearly straight-line relationship. Indeed, if single genotyping is performed, there is perfect agreement between the two raters for PIK3CA. For ESR1, there is high imprecision in the estimate, which limits the ability to draw a conclusive interpretation of the results. This can be explained, in part, by the limited number of ESR1-mutated cases detected in the study, and by the fact that ESR1 mutations were assessed at baseline. At this time, these alterations are typically absent or present at low levels, as they are usually acquired and expanded under the selective pressure of endocrine therapy. This may have contributed to lower detection sensitivity and slightly lower agreement between methods than those for PIK3CA. Thus, while the results support the interchangeability of both methods for PIK3CA detection, the conclusions regarding ESR1 should be interpreted with caution. Indeed, single ddPCR genotyping for ESR1 was limited by sample availability and by the absence of identical DNA inputs, which impaired its validation.

In general, our data align with prior studies on PIK3CA and ESR1 mutation identification³⁰, which found high agreement when exploring different PCR technologies, such as BEAMing and ddPCR³¹, as well as NGS strategies and ddPCR. For example, the Tempus xF assay, a 105-gene ctDNA seq panel, is highly concordant with ddPCR, tumor tissue-based NGS, and plasma-based NGS assays³⁴. Therefore, our data reinforce the consideration of ddPCR and AVENIO as valid and reliable methods for detecting ESR1 and PIK3CA mutations on cfDNA from patients with advanced BC. However, the strengths and limitations of each method point to different clinical applications. ddPCR provides rapid, easily interpretable results without the need for bioinformatics support, but its focus on predefined mutations limits the scope of information it provides. In contrast, AVENIO offers a broader genomic perspective, potentially revealing additional therapeutic targets and providing a more complete view of the tumor landscape, despite at the cost of longer turnaround times and the need for bioinformatics expertise. These findings are particularly relevant in adaptive treatment settings, where the timely identification of emerging resistance mutations is crucial. Ultimately, the choice between the two methods should depend on clinical context, available resources, turnaround time requirements, and the desired breadth of genomic information. Preliminary clinical data from this study in HR+/HER2- patients indicate no association between tumoral mutational status and the number of affected sites. However, higher cfDNA levels were observed in patients with visceral metastases, suggesting that cfDNA concentration may reflect increased tumor activity or metastatic spread. Although PIK3CA mutations have been linked to poorer outcomes in chemotherapy-treated patients, their prognostic significance in HR+/HER2– BC patients receiving CDK4/6 inhibitors remains unclear. In both PALOMA-3 and MONALEESA-3 trials, a non-significant trend toward worse outcomes was observed in patients with PIK3CA mutations. However, a pilot study by de Re et al. indicated that these mutations may contribute to therapy resistance³⁵. In our study, PIK3CA-mutated patients tend to have shorter OS, but non-significant results have been obtained so far. However, cfDNA levels have demonstrated prognostic value, as reported in previous studies^31,36. In the multivariable Cox analyses, which included both cfDNA levels and PIK3CA mutational status, cfDNA remained the main prognostic factor for overall survival, while PIK3CA mutations did not show an independent prognostic effect. A major limitation of the analysis is the low number of deaths, given the short follow-up period. Indeed, median progression-free survival has not yet been reached in many patients, and only patients with follow-up longer than 12 months were included in the survival analysis. Thus, these findings should be interpreted with caution. The results of the extended CANIPE study will be crucial for deepening our understanding of this patient subgroup, confirming previous results, and addressing additional objectives, such as analyzing mutational status during follow-up and progression and investigating the impact of coexisting mutations on clinical outcomes.

The analysis of ctDNA by multiplex ddPCR or AVENIO methodologies in HR+/HER2- advanced BC patients represents a sensitive tool for identifying PIK3CA and ESR1 mutations. The proportion of patients with detectable mutations in the CANIPE cohort was comparable, albeit slightly higher, to that reported in previous studies. High concordance was observed between ddPCR and AVENIO for PIK3CA and ESR1 mutational status using cfDNA, although the ESR1 results should be interpreted with caution due to limited statistical power. This study offers a unique contribution, as few previous reports have applied liquid biopsy genotyping directly to real-world patient cohorts, providing new evidence on its clinical applicability.

Methods

Clinical samples and tubes

About 106 advanced HR+/HER2− BC patients were recruited from May 2023 to December 2024. Eleven Spanish Hospitals were involved: Complexo Hospitalario Universitario de Santiago de Compostela, Complejo Hospitalario Universitario de Ourense, Hospital Álvaro Cunqueiro, Hospital Universitario Lucus Augusti, Complexo Hospitalario Universitario de Pontevedra, Hospital Universitario Central de Asturias, Hospital de León, Hospital Ramón y Cajal, Hospital Universitari Sant Joan de Reus, Hospital Arnau de Vilanova, and Hospital Ribera Povisa. Clinical data were obtained from the patient’s medical history at each hospital. All procedures followed the guidelines of the Helsinki Declaration and were approved by the Ethics Committee of Santiago-Lugo, under approval reference number 2022/386. All patients provided written informed consent. No person’s data was included.

The inclusion criteria for participants were to be adult males or females, over 18 years of age, with a diagnosis of advanced BC HR+/HER2- by immunohistochemistry and FISH technologies at the Pathology service of each hospital. Participants must initiate first-line treatment with CDK4/6 inhibitors and ET. Exclusion criteria were current or past diagnosis of a malignant tumor within five years, uncontrolled known psychiatric illness, and a life expectancy of less than three months.

At baseline, four blood Streck tubes (40 mL) were obtained from each patient and shipped, when necessary, to the Liquid Biopsy and Epigenomics Platform at the Health Research Institute of Santiago de Compostela within two days of collection. Plasma was obtained by a two-step centrifugation (1600×g and 5500×g, respectively), always within 3 days of blood drawing. The samples workflow is depicted in Fig. S1.

cfDNA isolation and quantification for digital droplet PCR (ddPCR)

cfDNA isolation was performed using 10 ml of plasma with the QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany), following the manufacturer’s instructions. The concentration of cfDNA was measured using the Qubit™ 1X dsDNA High Sensitivity Assay Kit with the Qubit™ 4 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA), following the manufacturer’s instructions.

Multiplex droplet digital PCR (ddPCR)

ddPCR reactions were performed with 30 ng of cfDNA in most cases. In those patients with lower amounts of cfDNA, the maximum possible amount was used, but never less than 12 ng per assay (the input DNA range was 12–118 ng). cfDNA was added to a PCR reaction mixture containing 12 µL of ddPCR™ Supermix for Probes (No dUTP) (Bio-Rad Laboratories, Hercules, CA, USA), 1.2 µL of primer/probe multiplex to ESR1 or 2.4 µL of primer/probe multiplex to PIK3CA (Table S1), 0.5 µL of HindIII (Cat. No. 1060AH, Takara Bio, Shiga, Japan) and nuclease-free water. Droplets were generated by loading 22 µL of the mixture into a single-use DG8™ Cartridge for QX200™ (Bio-Rad Laboratories, Hercules, CA, USA) with 70 µL of QX200™ Droplet Generation Oil for Probes (Bio-Rad Laboratories, Hercules, CA, USA). Droplets (40 µL) were generated with a QX200™ Droplet Generator (Bio-Rad Laboratories, Hercules, CA, USA) and transferred to a ddPCR™ 96-Well Plates (Bio-Rad Laboratories, Hercules, CA, USA) for PCR amplification on the C1000 Touch™ Thermal Cycler (Bio-Rad Laboratories, Hercules, CA, USA), according to the manufacturer’s protocol. The thermal cycling conditions were as follows: 95 °C for 10 min, 40 cycles of 94 °C for 30 s and 55 °C for 1 min, and a final inactivation step at 98 °C for 10 min. Positive and negative template controls (NTCs) were included in every PCR plate. The plate was read on the QX200™ Droplet Reader (Bio-Rad Laboratories, Hercules, CA, USA). The fraction of PCR-positive droplets was quantified assuming a Poisson distribution. QX Manager v2.3 Standard Edition (Bio-Rad Laboratories, Hercules, CA, USA) was used to obtain the number of mutated (FAM fluorophore) and non-mutated (HEX fluorophore) events for each control and sample.

To consider a sample as mutated for PIK3CA or ESR1 by ddPCR, at least 6 mutant events must be detected on FAM channel. For the positive and the wild-type controls, the positivity threshold for the mutant assay must be ≥6 events on the FAM channel. For the no template control, the positivity threshold must be fewer than 6 events in either of the two channels (FAM or HEX). The assay is considered invalid if any of the controls do not meet these criteria.

Single genotyping was performed using the previously described protocol, employing individual probes for each mutation in place of the multiplex.

cfDNA isolation and quality assessment for the AVENIO panel

cfDNA isolation was performed using 4 mL of plasma with the AVENIO cfDNA Isolation Kit (Roche Diagnostics, Basel, Switzerland), following the manufacturer’s instructions. The concentration of cfDNA was measured using the Qubit™ 1X dsDNA High Sensitivity Assay Kit with the Qubit™ 4 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA). cfDNA quality and fragment size distribution were assessed using the High Sensitivity D5000 ScreenTape Assay with the 4200 TapeStation system (Agilent Technologies, Santa Clara, CA, USA).

Library preparation and sequencing

According to the manufacturer’s instructions, NGS libraries were prepared with 10 to 50 ng of input cfDNA using the AVENIO ctDNA Expanded Kit v2 (Roche Diagnostics). Final libraries were quantified using the Qubit™ 1X dsDNA High Sensitivity Assay Kit, and their quality was evaluated with the High Sensitivity D5000 ScreenTape Assay on the Agilent 4200 TapeStation system. Libraries were pooled in batches of ten samples and sequenced on an Illumina NextSeq 500 system using the NextSeq 500 High Output v2 kit (300 cycles) with a 2 × 150 bp paired-end run configuration. Each sequencing run included the Structural Multiplex cfDNA Reference Standard (HD786; Horizon Discovery, Cambridge, UK) as a positive control. This reference material contains known single-nucleotide variants (SNVs), insertions and deletions (indels), and copy number variations (CNVs), which were used to monitor sequencing performance and consistency.

Sequencing data processing and gene variant analysis

Binary Base Call (BCL) files obtained from the sequencer were demultiplexed, aligned to the human reference genome assembly hg38 and analyzed using the AVENIO ctDNA Analysis Software version 2.1.0 (Roche Diagnostics). This software pipeline performs variant calling for SNVs, insertions/deletions, gene fusions and CNVs.

A mutation was considered positive if it met the following criteria: single-nucleotide variants (SNVs) with an allele fraction ≥0.1%, insertions and deletions (indels) with an allele fraction ≥1% and copy number variations (CNVs) with a score ≥5.

Statistical analysis

The chi-square test or Fisher’s exact test was used to assess associations between gene mutation status, cfDNA values, and clinical variables. Progression-free survival (PFS) and overall survival (OS) were visualized using Kaplan‒Meier plots and tested by the log-rank test. Receiver operating characteristic (ROC) analysis was conducted using the pROC package³⁷. Multivariate analysis was performed using the Cox model, and model comparison was performed using the likelihood ratio test (LRT). For concordance analysis, the psych, BlandAltmanLeh, and descTools R packages were used to calculate the Kappa coefficient and the Intraclass Concordance Index (ICC), generate the Bland-Altman plot, and perform the Passing-Bablok regression test, respectively. The ggplot2 package was used for graphics. Statistical analyses were performed in R (version 4.4.3) within RStudio (version 2024.12.1).

Author contributions

C.C., L.M.-R, A.D.-L., and R.L.-L. were responsible for the conceptualization. Methodology was performed by R.L., C.A., A.A., A.R.-C., and C.C. Formal analysis, visualization, and writing-original draft preparation were carried out by C.C. Resources were provided by T.C., C.R. P.P., J.C., M.C., A.C., S.M., J.G., K.A., M.L.F., Y.F., I.F.P., I.F.S., M.G.C., C.P., J.G.-M., A.S.d.M. Writing-review & editing were undertaken by C.C., L.M.-R., A.D.-L, R.L.-L., A.R.-C., A.C, J.C., and R.L. Supervision and funding acquisition were managed by R.L.

Data availability

All data were included in the article and Supplementary Table S2.

Competing interests

Rafael López (R.L.). reports grants and personal fees from Roche, Merck, AstraZeneca, Bayer, Pharmamar, and Leo, and personal fees and nonfinancial support from Bristol-Myers Squibb and Novartis, outside of the submitted work. The remaining authors declare no competing financial or nonfinancial interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41698-026-01546-3.