Tumor burden monitoring using cell-free tumor DNA could be limited by tumor heterogeneity in advanced breast cancer and should be evaluated together with radiographic imaging

A total of 49 patients with estrogen receptor (ER) positive breast cancer were prospectively enrolled in the study after signing the appropriate informed consent. Patients agree to the publication of anonymized data. The study protocol was approved by the Hospital Clinico San Carlos Ethics Committee (internal code 13/383-E). Eligibility criteria comprised the following: women aged between 18 and 80 years with a pathologically confirmed diagnosis invasive breast carcinoma, clinical stage IIB-IIIB or IV and availability of the archived tissue for genotyping at the Hospital Clinico San Carlos Biobank. Blood samples were collected at the beginning of a treatment that the patient was going to initiate when included in the study. In 43 cases, the tumor specimen collected was the primary breast tumor; a biopsy of a FFPE metastatic breast cancer lesion was available in the remaining 6 patients. Information concerning demographics, clinicopathological features (stage, grade, estrogen and progesterone receptor status and HER2 status), courses of treatment(s), and vital status was obtained from the clinical and pathology reports (Additional file 1).

Eighty six blood samples were collected in BD Vacutainer® CPT™ Cell Preparation Tube with Sodium Heparin (containing a gel barrier). Hemolyzed samples were discarded for further analyses. Plasma was derived from blood samples by centrifugation within 3 hours after blood extraction. cfDNA was isolated from 3 ml of plasma using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Valencia, CA), according to manufacturer’s instructions. cfDNA was eluted in 35 μL of the supplied elution buffer. The eluate was reloaded in the column and eluted by maximum speed centrifugation in order to increase the amount of cfDNA. Samples were then analyzed by digital PCR, using Rare Mutation Assays¹ for E542K (AHD2BSD), E545K (AHABHHX) and H1047R (AHPAVCD) on QuantStudio® 3D Digital PCR System (Applied Biosystems, South San Francisco, CA). For the dPCR reaction, 8 ul of cfDNA was mixed with 0.5 ul of the aforementioned 40X TaqMan assays and 10 ul of 2x QuantStudio 3D Master Mix, in a 20ul reaction volume. Subsequently, 15 ul were loaded into QuantStudio 3D Digital PCR 20 K chips. The cycling conditions were as follows: initial denaturation at 96 °C for 10 min, followed by 40 cycles at 56 °C for 2 min, and 98 °C for 30s, a step of 72 °C for 10 min and finally samples were kept at 22 °C for at least 30 min. Chip fluorescence was read twice. Results were analyzed with QuantStudio® 3D Analysis Suite™ Cloud Software. The automatic call assignments for each data cluster where manually adjusted when needed. The analysis of dPCR data was performed blinded to the clinical and tumor information by two independent investigators (AR and ML). Double positive data points (green data points) where only considered when FAM data points where visualized. The result of the assay is reported as mutant allele frequency (MAF) which is defined as the ratio of mutant DNA molecules vs the sum of wild-type (wt) DNA molecules and mutant DNA molecules. Samples were considered as positive when MAF was greater than 0.15%. A wt control DNA was included in every run. All assays were validated using plasmid DNAs harboring the aforementioned the PIK3CA mutations (GeneArt, Thermo Fisher Scientific). Germline wt DNA from healthy donors was mixed with different mutant allele concentrations (i.e. 1%, 0.5%, 0.1%, 0.05%) in order to estimate the Limit of quantitation (LOQ) and limit of detection (LOD). LOD and LOQ were calculated based on the standard deviation of the response and the slope according to ICH Q2(R1) guideline (http://​www.​ich.​org/​products/​guidelines/​quality/​article/​quality-guidelines.​html). The standard deviation of the response was calculated based on standard error of the y-intercept. Additionally, LOD and LOQ were estimated based on blank measurements. In this case, LOD was expressed as the average of MAF corresponding to the wt samples plus three standard deviation and LOQ was the average of MAF corresponding to the wt samples plus ten standard deviations. Two replicates were performed for all patients.

For the analyses of FFPE tumor and metastasis samples, a hemotoxylin and eosin-stained slide was reviewed by a pathologist to confirm the presence of tumor cells within the section. Overall, the percentage of tumor cells in the analyzed samples ranged from 80 to 90%. Subsequently, DNA was purified from four 5-mm-thick unstained FFPE sections according to the QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA) protocol.

The identification of PIK3CA mutations on FFPE samples was performed using COBAS® PIK3CA Mutation Test (Roche Molecular Systems, Branchburg, NJ). The test is intended to detect R88Q in exon 1, N345K in exon 4, C420R in exon 7, E542K, E545X (E545A, E545D,E545G, and E545K), Q546X (Q546E, Q546K, Q546L, and Q546R) in exon 9, and M1043I, H1047X (H1047L, H1047R, and H1047Y), and G1049R in exon 20 when the percent mutation is equal or greater than 5%. PIK3CA mutation status was further confirmed by the aforementioned TaqMan assays on the QuantStudio® 3D Digital PCR System. Conflicting samples were further amplified and sequenced. Primers and conditions used are available upon request. PCR products were then sequenced on the ABI 3130 genetic analyzer (Applied Biosystems, Foster City, CA), using the BigDye Terminator v1.1 Cycle Sequencing Reaction kit (Applied Biosystems, Foster City, CA) following the manufacturer’s protocol kit.

Serum samples were used to measure the levels of CA 15–3 antigen and CEA (Carcinoembryonic Antigen) using the high throughput immunoassay analyzer UniCelDxI 800 (Beckman Coulter), according to manufacturer’s instructions. In brief, the assay is a two-step sandwich immunoassay using direct, chemiluminescent technology.

Mammogram, ultrasound-based, magnetic resonance imaging (MRI), computed tomography (CT) and Positron emission tomography (PET)-CT tumor measurements were obtained, as clinically indicated. The clinical response was evaluated according to RECIST v1.1 criteria comparing radiological assessments. Complete response (CR) was defined as resolution of all palpable, visible or radiology abnormalities in the breast and regional lymph nodes. Partial response (PR) was defined as a decrease of ≥30% in the sum of the longest diameter in the breast and regional lymph nodes. Stable disease (SD) was assigned to patients who did not meet the criteria for CR, PR or PD. Clinical Progressive disease (PD) was defined as an increase of at least 20% in the sum of the longest diameter in the breast and regional lymph nodes or progression of other clinical manifestations of disease.

We report data from 49 patients with ER positive breast cancer undergoing therapy that were prospectively enrolled in the study. The pathological characteristics of study population are summarized in Table 1 and Additional file 1. The source of archival tissue sample was the primary tumor for 43 (87.8%) patients and biopsy of a distant metastasis for 6 (12.2%). DNA isolated from archival-tumor tissue samples was analyzed to identify PIK3CA somatic mutations. According to COBAS, among the analyzed tumors, 11 (22.4%) harbored the mutation H1047R, 5 (10.2%) harbored the E545K and 4 (8.1%) harbored the E542K. These are expected frequencies according to previously published data. Interestingly, we did not encounter any other less frequent PIK3CA mutation that could be detected by COBAS methodology. Subsequently, the three PIK3CA hot spot mutations were screened in FFPE derived samples using QuantStudio 3D system. The proportion of observed agreement between COBAS and QuantStudio3D system when assessing the PIK3CA mutation status in the archival tumor was 100% (K = 1) (Additional file 1).

Next, we evaluated the detection sensitivity of TaqMan assays. For sensitivity assays plasmid carrying E542K, E545K and H1047R mutations were mixed at different allele concentrations with wt DNA extracted from peripheral blood cells form healthy donors. Mutant allele frequencies correlated with the expected mutant allele frequencies in E542K assay, E545K assay and H1047R assay (Pearson’s correlation coefficient, 0.9985, 0.9966 and 0.9991 respectively). LOD for E542K, E545K and H1047R assays were 0.04, 0.05 and 0.03% respectively. LOQ for E542K, E545K and H1047R assays were 0.13, 0.15 and 0.08% respectively (Additional file 2). Additionally, wt DNA from healthy donors were used to evaluate the false positive signals. The mean of mutant allele fraction of wt DNA form healthy individuals was 0.01% for E542K assay, 0.00% for E545K assay and 0.01% for H1047R.

Finally, we evaluated the agreement between PIK3CA mutation status of FFPE samples and the matched plasma sample. The median time between blood collection at study entry and time that the tumor biopsy was obtained from the patient was 7 years (range = 0–15 years). Importantly, cfDNA from all plasma samples was successfully amplified. Representative plots for the three assays, performed on cfDNA, are displayed in Additional file 3. The ratio of mutant DNA molecules vs total DNA molecules ranged from 0.19 to 53.4% in the positive sample cases. The median of PIK3CA mutated copy number in positive samples was 1496 copies/ml (range: 63.3–92742).

Overall, we found a moderate agreement between PIK3CA mutation status in plasma and FFPE tumor samples (K = 0.591; 95% IC = 0.371–0.811). Specifically, we found that in 9 samples, the assay failed to detect on cfDNA the mutation identified in the tumor sample (data obtained from Additional file 1). However, we did not find any mutation in the plasma that was not present in the FFPE sample (sensitivity 55%, specificity 100%, positive predictive value 100%, and negative predictive value 76%). Restricting the analysis to the metastatic patients, we found a good agreement between PIK3CA mutation status assessed in liquid and solid biopsy (K = 0.798 95%; IC = 0.586-1) (data obtained Additional file 1). In this case, the cfDNA assay failed to detect the tumor mutation in 3 samples (sensitivity 77%, specificity 100%, positive predictive value 100%and negative predictive value 86%).

In 8 breast cancer patients with stage IV disease in which PIK3CA somatic mutation was identified in both FFPE and plasma samples, mutant cfDNA and tumor markers CA15-3 and CEA were quantified in a total of 27 serial plasma samples (Table 2). Circulating tumor DNA (ctDNA) was identified in 21 of the 27 plasma samples analyzed. Tumor markers CA15-3 and CEA levels were evaluated at all-time points in all 8 patients. Circulating tumor DNA showed serial changes with fluctuations correlating with tumor markers in 7 cases. In one case (from now on patient 3) changes in PIK3CA mutation quantification were in the opposite direction than tumor markers. In patients in which tumor markers CA15-3 and CEA and PIK3CA mutant DNA fluctuated in the same way, Pearson correlation coefficients ranged from 0.99 to 0.46 and from 0.99 to 0.38 respectively (mean 0.82 for CA 15–3 and 0.79 for CEA).

As an exploratory analysis, in the aforementioned 8 patients we compared the performance of PIK3CA assays with measurable disease (as defined by RECIST criteria). The median follow-up was 7.5 months. Overall, changes in ctDNA were in concordance with treatment responses observed in imaging. Partial responses were recognized in two patients (patient 25 and patient 44) during the follow-up period. In both cases, a reduction of mutant allele fraction in plasma samples was evidenced (Additional file 4: Figure S1 and Fig. 1b respectively). Similarly, progressive disease (PD) was documented in 4 women (patient 44, patient 2, patient 31, patient 3) during the study. An Increase in PIK3CA mutant allele fraction reflected progressive disease in 2 patients (patient 2 represented in Fig. 1a and patient 44 represented in Fig. 1b). Noteworthy, in one case (patient 31), although H1047R mutation quantification went down to 0% over the course of treatment (from 0.19 to 0%; Table 2) correlating with tumor markers CA15.3 and CEA variations and with the decrease in size of the pulmonary metastases and the axillary lymphadenopathy, the contrast-enhanced CT scan from the same period of time demonstrated that the lytic bone metastases increased in size (Fig. 1c). Similarly, in case of Patient 3, the hiliar and mediastinal adenopathies decreased in size from 13 to 8 mm to 7 and 4 mm respectively, correlating with E545K mutation quantification that significantly decreased from 3.07% (CI: 2.71% – 3.465%) to 1.97% (CI: 1.343% – 2.88%) (Table 2) in the corresponding plasma samples. However, the pulmonary metastases increased in number and size during the same period of time correlating with tumor markers quantification (Fig. 2). Finally, stable disease was documented in two patients (patient 15 and patient 6) in which tumor circulating free DNA fluctuated in the same way that tumor markers fluctuated (Additional file 4: Figure S3 and Figure S4).