PIK3CA hotspot mutations in circulating tumor cells and paired circulating tumor DNA in breast cancer: a direct comparison study

Liquid biopsy analysis, mainly based on circulating tumor cells (CTCs) and circulating tumor DNA (ctDNA), provides an extremely powerful tool for the molecular profiling of cancer patients in real time. In this study, we directly compared PIK3CA hotspot mutations (E545K, H1047R) in EpCAM‐positive CTCs and paired plasma‐ctDNA in breast cancer (BrCa). PIK3CA hotspot mutations in CTCs and ctDNA were analyzed using our previously developed highly sensitive (0.05%), specific, and validated assay in plasma‐ctDNA from 77 early and 73 metastatic BrCa patients and 40 healthy donors. We further analyzed and directly compared PIK3CA hotspot mutations in DNAs isolated from CellSearch® cartridges (CTCs) and paired plasma‐ctDNA, in 56 cases of early and 27 cases of metastatic breast cancer, and 16 corresponding primary tumors. In plasma‐ctDNA,PIK3CA hotspot mutations were identified in 30/77(39.0%) early and 35/73(47.9%) metastatic BrCa cases; none (0/40, 0%) of the healthy donors’ plasma‐ctDNA samples were positive. Our direct comparison study in DNAs isolated from CellSearch® cartridges (CTCs) and paired plasma‐ctDNA from the same blood draws has shown a lack of concordance in early BrCa (27/56, 48.2%), while the concordance in the metastatic setting was higher (18/27, 66.6%). Our results were validated by ddPCR methodology, and the concordance between our assay and ddPCR for PIK3CA E545K hotspot mutation was 30/37 (81.1%). In many cases, PIK3CA hotspot mutations were detected in samples found to be negative for CTCs in CellSearch®. Our data demonstrated for the first time that (a) PIK3CA hotspot mutations are present at high frequencies in CTCs isolated from CellSearch® cartridges and paired plasma‐ctDNA both in early and metastatic BrCa, (b) the detection and concordance of PIK3CA hotspot mutations between plasma‐ctDNA and CTCs are higher in the metastatic setting, (c) PIK3CA mutational status significantly changes after therapeutic intervention, and (d) PIK3CA mutation detection in CTCs and plasma‐ctDNA provides complementary information.


Introduction
Dynamic changes of tumor features over time, also known as tumor heterogeneity, are now recognized as one of the most significant issues in tumor biology (Bedard et al., 2014). A major limitation of the classic tissue biopsy approach is that genetically different subclones that are present at a minor frequency in the primary tumor and thus possibly not initially detected may be selected after treatment or differentially represented in the metastatic process, leading to treatment resistance (Bedard et al., 2014). The minimally invasive 'liquid biopsy' approach overcomes these limitations since by providing new ways to monitor tumor genetics and tumor dynamics in real time (Alix-Panabi eres and Pantel, 2016).
Liquid biopsy has a high potential to significantly change the therapeutic strategy in cancer patients but there are still many challenging questions to be answered (Bardelli and Pantel, 2017). Detection, enumeration, and molecular characterization of circulating tumor cells (CTCs) and analysis of circulating tumor DNA (ctDNA) provide useful information regarding the individual molecular profile of each patient in real time, before and after treatment (Lianidou and Hoon, 2017). The field of liquid biopsy applications is growing exponentially, including molecular target identification, prognosis assessment, diagnosis of recurrence, monitoring of response to treatment, and monitoring of tumor genomic profiles over time (Lianidou and Hoon, 2017). Most importantly, blood-based tests are very challenging and highly important in cases where tumor biopsies are not accessible (Ignatiadis et al., 2015;Lianidou et al., 2015;Wang et al., 2017). Recently, the US Food and Drug Administration (FDA) approved the first liquid biopsy-based companion diagnostic test for the administration of first-to third-generation TKI therapies in NSCLC patients, based on the presence of sensitive or resistant EGFR mutations in cell-free DNA (cfDNA) (Malapelle et al., 2017).
Several studies so far have shown that both CTCs and ctDNA can be detected in peripheral blood of cancer patients not only in advanced but even at early stages (Bettegowda et al., 2014;Stathopoulou et al., 2002). ctDNA analysis in plasma has been suggested as an alternative to the classic biopsy approach since it essentially comprises a subtype of total cfDNA that is derived from tumor cells thus providing their genomic signature. However, there is an essential difference between CTCs and ctDNA; CTCs as viable cells circulating in peripheral blood can provide real-time information on the metastatic spread and reveal active and possibly targetable signaling networks, while ctDNA can give specific information on the presence or absence of specific alterations deriving from the tumor, indicating therapy response or resistance (Bidard et al., 2018;Gold et al., 2015;Siravegna et al., 2017).
Circulating tumor cells enumeration using the Cell-Search Ò system has been approved by the FDA more than a decade ago as a prognostic marker in metastatic breast, colorectal, and prostate cancer . CTC detection and enumeration are associated with progression-free survival (PFS) and overall survival (OS) in metastatic (Bidard et al., 2014;Cristofanilli et al., 2004) and early BrCa (Lucci et al., 2012;Rack et al., 2014;Stathopoulou et al., 2002). Beyond enumeration, CTC molecular characterization is very important, since it can provide significant information at the gene expression, DNA methylation, and DNA mutation level (Aktas et al., 2016;Chimonidou et al., 2011Chimonidou et al., , 2013Markou et al., 2011Markou et al., , 2014Mastoraki et al., 2018;Mostert et al., 2015;Sieuwerts et al., 2018;Strati et al., 2011). However, CTCs are highly heterogeneous, and there is a lack of a unique marker for their isolation and identification, since there are not well-defined universal surface targets among all malignant cell types (Lianidou and Hoon, 2017). This issue is becoming more complicated when surface CTC-enrichment targets change during epithelial-tomesenchymal transition (EMT) (Gorges et al., 2012). Based on that, a variety of EpCAM-independent systems have been now developed to capture and identify CTCs (Lianidou and Hoon, 2017). CTC molecular characterization can offer very important information on patient prognosis and can also identify therapeutic targets or mechanisms of resistance, such as mutations in driver genes (Bingham et al., 2017;Kidess-Sigal et al., 2016;Liu et al., 2017;Pestrin et al., 2015;Shaw et al., 2017).
Plasma-ctDNA analysis is another important component of liquid biopsy. Increased concentrations of ctDNA fragments have been detected in plasma and serum of cancer patients with various tumor types, and their presence has been correlated with unfavorable outcome (Bettegowda et al., 2014;Dawson et al., 2013;Schwarzenbach et al., 2011;Wan et al., 2017). ctDNA analysis based on tumor-derived genetic alternations is now increasingly used for treatment purposes (Malapelle et al., 2017), and the large dynamic range of plasma-ctDNA presents a considerable correlation with changes in tumor burden (Bettegowda et al., 2014;Dawson et al., 2013).
The evaluation of mutational status of specific genes in liquid biopsy material is highly informative, but requires highly sensitive and robust methodologies. Recently, our group developed and validated an ultrasensitive and highly specific methodology for the detection of PIK3CA hotspot mutations (E545K, H1047R) in CTCs, based on the combination of allele-specific priming, competitive probe blocking of wild-type amplification, asymmetric PCR, and probe melting analysis (Markou et al., 2014). Using this highly sensitive assay (0.05%), we have shown that PIK3CA hotspot mutations are present at a relatively high frequency in the EpCAM-positive CTC fraction not only in patients with metastatic but with early BrCa as well. Our study has also shown that PIK3CA mutational status can change during disease recurrence or progression in BrCa patients (Markou et al., 2014). The aim of the present study was to use this highly sensitive methodology to directly compare the mutational status of PIK3CA (E545K, H1047R) in CTC-DNAs isolated from CellSearch Ò cartridges and paired plasma-ctDNA from the same blood draws in patients with early and metastatic BrCa.

Materials and methods
The outline of the workflow of our study is shown in Fig. 1.

Clinical samples
We initially analyzed PIK3CA hotspot mutations in 190 plasma-ctDNA samples from (a) patients with early BrCa (n = 77), (b) patients with metastatic BrCa (n = 73), and (c) healthy donors (all female, n = 40). We further performed a direct comparison study of PIK3CA hotspot mutations in DNAs isolated from CellSearch Ò cartridges and paired plasma-ctDNA in 56 samples of 43 patients with early BrCa and in 27 samples of 16 patients with metastatic breast cancer. Pretreatment and posttreatment peripheral blood samples were available for 13 and 11 of these patients with early and metastatic BrCa, respectively, while corresponding formalin-fixed paraffin-embedded (FFPE) primary tumor samples were available for 16 of these patients. The study was conducted in accordance with the 1964 Declaration of Helsinki and was approved by the ethics and scientific committees of the participating Institutions. All participating patients signed an informed consent form to participate in the study, which was approved by the ethics and scientific committees of our institutions. The clinicopathological characteristics of the patients are shown in Table S1. We also included in our study 42 DNA samples isolated from CellSearch Ò cartridges from an independent group of 17 metastatic BrCa patients, before and after first cycle of treatment. For 8 of these patients, blood sampling at progression of disease was also available. The majority of these patients were ER+/PR+/ HER2À.

Isolation of plasma-ctDNA
Whole blood samples were collected into venous blood collection tubes using K 2 EDTA (BD Vacutainer, Plymouth) as anticoagulant. Samples were mixed thoroughly, and plasma was isolated within 2 h from sample collection by centrifugation at 530 g for 10 min at room temperature. Once isolated, plasma samples were further centrifuged twice at 2000 g for 10 min, before transferring into clean 2-mL tubes and freezing at À70°C until further processing. The QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany) was used to isolate ctDNA from 2.00 mL of plasma according to manufacturer's instructions.

Isolation of gDNA from CellSearch â cartridges
CellSearch Ò analysis was performed according to the manufacturer's instructions; 7.5 mL of PB was used for each patient with metastatic disease, while in patients with early BrCa, three CellSave tubes were used for each patient (in total 22.5 mL PB). Following CTC analysis, CellSearch Ò cartridges were stored in a dark place at 4°C until gDNA isolation. CTCs and WBCs (prestained with antibody to CD45, pan-CK, and DAPI) were aspirated from the CellSearch Ò cartridge and underwent downstream gDNA extraction using the QIAamp DNA Micro Kit (Qiagen) according to manufacturer's instructions.

gDNA isolation from FFPEs
Formalin-fixed paraffin-embedded tissue sections of 10 mm containing > 80% of tumor cells were used for DNA extraction. gDNA was isolated with the QIAamp DNA FFPE Tissue Kit (Qiagen), according to the manufacturer's instructions.

Whole genome amplification
In the independent group of 42 clinical samples analyzed, the archival DNA material was not sufficient for downstream analysis. For this reason, a whole genome amplification (WGA) step for all these samples was included. Amplification with REPLI-g kit (Qiagen) was performed according to the manufacturer's recommendations. A blank and DNA from human cell lines, MCF-7 and T47D, were included for each experimental procedure as negative and positive controls, respectively. DNA concentration was determined using the Quant-iT TM PicoGreen TM dsDNA Assay Kit (Invitrogen TM , Waltham, MA, USA), and DNA was diluted according to the manufacturers' manual. To verify the specificity of our assay after WGA, 10 plasma samples from healthy donors were checked before and after WGA step.

Sample preparation
To avoid contamination, different rooms, dedicated labware, and dedicated areas were used for all procedures. All DNA preparation and handling steps took place in specific laminar-flow hoods under DNase-free conditions. DNA concentration in all cases was measured in a NanoDrop 1000 spectrophotometer (Thermo Scientific, Waltham, MA, USA) , calibrated with the recommended CF-1 (concentrated aqueous potassium dichromate, K 2 Cr 2 O 7 ) standard solution. The isolated DNA samples were stored at À70°C until further use.

Quality control
Synthetic gene fragments were synthesized as gBlocks by Integrated DNA Technologies (Coralville, IA, USA) and were used as positive controls for PIK3CA hotspot mutations (sequences are available upon request). Synthetic oligos for E545K and H1047R mutation sites were designed so that each sequence was represented by a unique gBlock that mutation position would be included. Lyophilized gBlocks were suspended in Tris/EDTA buffer to a stock  concentration of 10 ngÁmL À1 and were further diluted to working concentrations as needed. Also, two negative controls were included in each experimental procedure, one as negative control of PCR reaction (blank) and the second one as wild-type control, which contained wild-type as DNA template. All samples were checked for their DNA integrity prior to analysis, by amplifying a region of PIK3CA exon 20 that includes the hotspot mutation site.

Spiking experiments
To further verify the analytical sensitivity of the assay, we collected peripheral blood samples from two healthy donors in 2 different types of tubes: (a) K2EDTA (BD Vacutainer) and (b) CellSave (Menarini Silicon Biosystems). In both cases, the first 5 mL of blood was discarded to avoid skin epithelial cell contamination. In each tube, MCF-7 cells (100 cells enumerated in a Malassez Hemocytometer) were spiked into peripheral blood (10 mL in EDTA tube and 7.5 mL in CellSave tube), and mixed immediately after spiking. In the EDTA tube, CTC isolation was performed by adding red cell lysis buffer and capture beads, coated with the monoclonal antibody BerEP4 against the human epithelial antigen EpCAM. In the CellSave tube, CellSearch Ò analysis was performed according to the manufacturer's instructions. Following CTC analysis, CellSearch Ò cartridges were used for downstream analysis of EpCAM-positive CTCs. In both cases, DNA was isolated using QIAamp DNA Micro Kit and DNA concentration was measured in a NanoDrop 1000 spectrophotometer that was calibrated with the recommended CF-1 standard solution.
2.10. Droplet digital PCR for the detection of PIK3CA E545K hotspot mutation We analyzed 37 of these DNA samples using droplet digital PCR (ddPCR) technology for PIK3CA E545K hotspot mutation. All clinical samples were analyzed with the QX200 Droplet Digital PCR System (Bio-Rad Laboratories, Inc., Hercules, CA, USA) using TaqMan hydrolysis probes for detecting and quantifying wild-type PIK3CA, as well as PIK3CA E545K in exon 9. One probe was specific for the wild-type sequence (wild-type reference assay/HEX, Channel 2), and the other was specific for each PIK3CA E545K hotspot mutation (mutation assay/FAM, Channel 1).
The PCR mix for one well was set up by mixing 2 lL DNA sample, 10 lL 2 9 ddPCR Supermix for Probes (No dUTP) (Bio-Rad Laboratories), 1 lL 20 9 target primers/probe (FAM), 1 lL 20 9 reference primers/ probe (ΗΕΧ), in a reaction volume of 20 lL, adjusted with sterile water, according to the manufacturer's instructions. The cycling profile of the PIK3CA mutant detection assay was as follows: 95°C for 10 min (1 cycle), 94°C for 30 s, and 55°C for 60 s (55 cycles), and infinite hold at 12°C. Each experimental procedure included a negative control (nontemplate control) and a positive control (MCF-7 cell line), which contained enough mutant and wild-type DNA to create a double-positive cluster of greater than 100 droplets. PCR products were loaded onto the QX200 Droplet Reader (Bio-Rad Laboratories, Inc.), and the data were analyzed with QuantaSoft Software.

Statistical analysis
Statistical analysis was performed via SPSS version 24.0 (IBM Ò SPSS Ò Statistics, Armonk, NY, USA). A P-value ≤ 0.05 was considered to be significant.
3.3. PIK3CA mutational status in CTCs and paired plasma-ctDNA changes before and after treatment
3.5. Comparison study between the combination of allele-specific, asymmetric rapid PCR and melting analysis assay and ddPCR for the detection of PIK3CA E545K hotspot mutation The results of the clinical sample analysis were based on the definition of threshold line for each subpopulation of mutant or wild-type droplets. All positive droplets, those above the threshold line of 3.500 droplets, are scored as positive, whereas the negative droplets those above the threshold line of 3.500 are scored as negative (background). In case of clinical samples, at least one droplet of PIK3CA E545K DNA-mutated sequence was considered as positive sample for PIK3CA E545K hotspot mutation by ddPCR. For this study, we analyzed in total by ddPCR 37 samples: 17 genomic DNAs isolated from CellSearch Ò cartridges (EpCAM-positive cells) and 20 plasma-ctDNA samples for PIK3CA E545K mutation. More specifically, 14 and 3 DNAs isolated from CellSearch Ò cartridges (EpCAM-positive cells) from early and metastatic patients, and 11 and 9 plasma-ctDNA samples from early and metastatic patients, respectively, were used for this study.
In total, the concordance between our assay and ddPCR for PIK3CA E545K hotspot mutation was 30/ 37 (81.1%) ( Table 3). More specifically, 21 samples were found positive by both methods, and 9 samples found negative by both methods. There were only 2 samples that were found positive by our method and negative by ddPCR, and 5 samples that were found positive by ddPCR and negative by our method. Our data support that our assay and ddPCR give comparable results (P = 0.001).
3.6. PIK3CA hotspot mutations in CTCs from an independent group of DNA samples isolated from CellSearch â cartridges In the independent patient group studied, PIK3CA hotspot mutations were detected in 25/42(59.5%) DNA samples isolated from CellSearch Ò cartridges most of which were positive for CTCs (Fig. S1). PIK3CA E545K hotspot mutation was detected in 12/ 42(28.6%) DNAs isolated from CellSearch Ò cartridges, whereas PIK3CA H1047R hotspot mutation was detected in 17/42(40.5%). CellSearch Ò detected at least one CTC in 37/42(88.1%) samples, and PIK3CA hotspot mutations were identified in 22/37(59.5%) DNAs isolated from CellSearch Ò cartridges positive for CTCs. Moreover, PIK3CA hotspot mutations were also detected in 3/5(60%) cases that were found negative for CTCs by CellSearch Ò . These data are in concordance with our previous results, indicating that PIK3CA mutations are also present in EpCAM-positive cells where CellSearch Ò gave negative results for the presence of CTCs. In this independent patient group, we found in total 4/42(9.5%) samples where both PIK3CA mutations were detected. As can be seen in Fig. S1, in all these four cases, CTCs were detected in the CellSearch Ò . Moreover, in 15/42(35.7%) cases where CellSearch Ò was positive for CTCs, no PIK3CA mutations were detected.

Discussion
Liquid biopsy is now becoming a highly important tool for following up cancer patients in real time with a variety of solid tumors (Alix-Panabi eres and Pantel, 2016; Bardelli and Pantel, 2017;Bedard et al., 2014;Bettegowda et al., 2014;Ignatiadis et al., 2015;Lianidou and Hoon, 2017;Lianidou et al., 2015;Malapelle et al., 2017;Stathopoulou et al., 2002;Wang et al., 2017). However, some questions still need to be answered (Bardelli and Pantel, 2017). One common question is whether CTCs and ctDNA give identical or complementary information, and which of the two is the best choice to follow up patients over time.
PIK3CA mutational status has already been studied in CTCs and plasma-ctDNA using different methodologies such as NGS and ddPCR; however, there are no studies till now where PIK3CA mutational status was evaluated in CTC and plasma-ctDNA in identical blood draws using the same methodology. In the present study, we addressed this question for the first time by performing a direct comparison study of PIK3CA mutational status (E545K, H1047R) in DNA isolated from CellSearch Ò cartridges and paired plasma-ctDNA from the same blood draws in early and metastatic BrCa, using our previously described highly sensitive real-time PCR methodology (Markou et al., 2014). PI3K is one among the most important downstream molecules in the pathway of tyrosine kinase growth factor receptors and one of the most promising targets for personalized medicine. PIK3CA mutations have been reported in 18%-40% (Barbareschi et al., 2007) of BrCa cases; three 'hotspot' mutations (E542K, E545K, and H1047R) comprise more than 80% (Campbell et al., 2004) of all PIK3CA mutations and are localized in exons 9 and 20. Moreover, early clinical studies have indicated that the presence of PIK3CA mutations is linked to the acquirement of enhanced sensitivity to PI3K pathway inhibitors (Janku et al., 2011). Although the presence of PIK3CA mutations in archival tissues has been evaluated in many clinical trials such as BOLERO-2, FERGI, BELLE-2, and BELLE-3, the PFS benefit from PI3K or mTOR inhibitors seems to be largely maintained irrespectively of PIK3CA genotype (Hortobagyi et al., 2016); it is important to note that the detection limit of the nextgeneration sequencing methodology used in this study for mutation allele frequency (MAF) was 5%.
However, after taking into consideration the advantages of noninvasive liquid biopsy analysis, we expect that identifying key mutations in CTCs and plasma-ctDNA from peripheral blood samples may provide more actionable information on the molecular profile of metastatic tumors. Results of the BELLE-2 and BELLE-3 clinical studies, based on the combination of a pan-PI3K inhibitor (buparlisib) with endocrine therapy, indicated a better PFS in the PIK3CA mutant subgroup identified through ctDNA analysis, but not for the PIK3CA mutant subgroup as identified in primary tumors (Baselga et al., 2017;Di Leo et al., 2018). However, the BOLERO-2 study which examined the benefit of an inhibitor of the PI3K/AKT/ mTOR pathway (everolimus), has shown a benefit for patients when everolimus was added to endocrine therapy regardless of the presence of a PIK3CA mutation (H1047R, E545K, and E542K) in either tumor tissues or ctDNA samples analyzed by droplet digital PCR (ddPCR), suggesting that PIK3CA mutation status is not a predictive determinant for everolimus benefit (Moynahan et al., 2017). mTOR, however, is downstream of PI3K, and studies have shown that AKT-independent mechanisms of mTOR activation are involved in breast carcinogenesis (Tee et al., 2003). Recently, the mTOR inhibitors temsirolimus and everolimus and the PI3K inhibitors idelalisib and copanlisib have been approved by the FDA for clinical use in the treatment of a number of different cancers. Very recently, it was shown that early ctDNA dynamics in the commonly truncal mutations in PIK3CA may predict sensitivity to palbociclib, a CDK4/6 inhibitor (O'Leary et al., 2018a(O'Leary et al., ,2018b. Novel compounds with greater potency and selectivity, as well as improved therapeutic indices owing to reduced risks of toxicity, are clearly required. In addition, biomarkers that are Table 3. Comparison study between the combination of allelespecific, asymmetric rapid PCR and melting analysis assay and droplet digital PCR for detecting PIK3CA E545K hotspot mutation in early and metastatic BrCa patients Combination of allele-specific, asymmetric rapid PCR and melting analysis assay (Markou et al., 2014) ddPCR predictive of a response, such as PIK3CA mutations for inhibitors of the PI3K catalytic subunit a isoform, must be identified and analytically and clinically validated (Janku et al., 2018).
PIK3CA mutational status has already been studied in CTCs (Bingham et al., 2017;Gasch et al., 2016;Markou et al., 2014;Schneck et al., 2013) and plasma-ctDNA (Moynahan et al., 2017;Takeshita et al., 2017). When PIK3CA hotspot mutations were analyzed in CTCs isolated using a label-free technology and corresponding ctDNA in 9 patients using 23 identical blood draws with colorectal cancer, mutational status of PIK3CA showed a match between CTCs and ctDNA in six blood draws and discordance in two, with 15 blood draws showing no mutations; however, in this study, PCR-based Sanger sequencing (analytical sensitivity 7.5%) was used for mutation detection in CTCs and targeted panel NGS for cfDNA (Kidess-Sigal et al., 2016). In another recent study, in a limited number of five metastatic breast cancer patients with more than 100 CTCs/7.5 mL PB (enumerated using the CellSearch), single EpCAM-positive CTCs were isolated by DEPArray and compared with matched cfDNA and primary tumor tissue by targeted NGS for about 2,200 mutations in 50 cancer genes; according to this study, in these five patients, cfDNA profiles provided an accurate reflection of PIK3CA mutations seen in individual CTCs (Kidess-Sigal et al., 2016;Shaw et al., 2017). According to these studies, the percentage of PIK3CA hotspot mutations in CTC and plasma-ctDNA is around 30-40%. Our results indicate a much higher percentage of PIK3CA-positive samples in both CTC and plasma-ctDNA and this could possibly be explained by the higher analytical sensitivity of the molecular assay that we used (analytical sensitivity 0.05%) (Markou et al., 2014).
According to our results, in both early and metastatic BrCa, PIK3CA hotspot mutations were detected in the large majority of DNAs isolated from Cell-Search Ò cartridges but in a lower number of paired plasma-ctDNA samples. It is interesting to note that in most cases where CTCs were detected by Cell-Search Ò , PIK3CA hotspot mutations were identified either in ctDNA or in CTCs or both, in the early and metastatic BrCa setting. There were two cases where CTCs were detected at high numbers, but PIK3CA mutations were not detected in CTCs. Moreover, there were many cases where CTC were not detected by CellSearch Ò but our downstream analysis in DNA material isolated from these cartridges was positive for PIK3CA mutations. This finding could be possibly explained by the fact that some EpCAM-positive cells can be negative for CKs (8,18,19) due to EMT. These findings are in accordance with our previous results, where we identified PIK3CA hotspot mutations in EpCAM-positive cells that were negative for CK-19 mRNA expression (Lampignano et al., 2017;Markou et al., 2014). We have also shown in one of our previous studies (Markou et al., 2018) by using exactly the same blood draws that we can detect a lot of molecular alterations in EpCAM-positive CTCs, in identical samples that were 'officially' negative for CTCs when using the CellSearch.
We realize that we report a high percentage of PIK3CA mutations in primary tumors; however, in all cases where PIK3CA E545K hotspot mutation was detected in FFPEs, it was also detected in DNA isolated from CellSearch Ò cartridges. However, we have to emphasize that the number of primary tumors that we tested was very low, since these are not randomly selected samples, but highly selected paired samples, where ctDNA and CTCs were also available, and our main aim was not to show the incidence of PIK3CA mutations in primary breast tumors but to compare the PIK3CA mutational status in primary tumors, CTCs and ctDNA. Our findings are in accordance with our previous paper (Markou et al., 2014), and we confirm again that there was no concordance in CTCs, ctDNA, and primary tumors.
Our aim was to evaluate the potential of our previously published assay for detecting PIK3CA hotspot mutations in DNAs isolated from CellSearch Ò cartridges and paired plasma-ctDNA samples in early and metastatic breast cancer patients. The patients' selection was performed independently from the receptor status of these patients. Our aim was clearly not to provide evidence regarding the potential utility of PIK3CA mutation status by plasma-ctDNA or CTC in predicting response to specific therapy. This is not a clinical study, and the samples analyzed were not selected based on the clinical characteristics of the patients. On the contrary, the BELLE-2 and BELLE-3 clinical trial studies included a large number of clinical samples, and patients' selection was based on specific inclusion and exclusion criteria, demonstrating the incidence of PIK3CA mutations in specific target populations.
Our direct comparison study revealed for the first time that in early breast cancer, PIK3CA hotspot mutations are present in CTCs at a higher percentage before than after treatment, when compared to paired plasma-ctDNA samples. This finding may suggest that in early BrCa, standard treatment was more often successful in targeting a number of cancer cells, thus leading to the release of DNA derived from these dying cells into the systemic circulation. However, in metastatic BrCa, the percentage of PIK3CA hotspot mutations in DNAs isolated from CellSearch Ò cartridges was significantly higher in post-treatment samples. This could be possibly explained by increased resistance to standard therapy which was not targeted therapy for PIK3CA hotspot mutations. Therefore, this nontargeted therapy seems to be ineffective to reduce the CTC/tumor burden in metastatic BrCa patients harboring PIK3CA hotspot mutations. In the same samples, the presence of PIK3CA hotspot mutations in plasma-ctDNA was decreased after treatment, possibly indicating that the cells remained intact, hampering the release of ctDNA in peripheral blood. The concordance of PIK3CA mutations between CTCs and ctDNA was relatively poor in early breast cancer (48.2%) and higher in the metastatic setting (66.6%).
Since the high incidence of mutations found could raise questions, we performed a comparison study in the same samples using a commercially available highly sensitive methodology for PIK3CA mutations based on ddPCR, for ctDNA samples and CTCs that were available. Our results revealed a high concordance between our assay and ddPCR (81.1%, P = 0.001).
Our results in CTCs could only be compared with similar studies using highly sensitive methodologies, like ddPCR; however, till now so highly sensitive methodologies have not been applied in CTCs isolated using the CellSearch system. In the study of Schneck et al., the mutation rate reported for PIK3CA mutations in CTCs was only 15.8%. This percentage is much lower than the one that we report; however, this discrepancy can be clearly explained by the much lower analytical sensitivity of the SNaPshot methodology that is 5 to 10% (Hurst et al., 2009). Our results in ctDNA of metastatic breast cancer patients are in agreement with a very recent study by O'Leary et al. (2018b) where a percentage of 56.3% (9/16) is reported for PIK3CA mutations, in ctDNA samples, since we also report for metastatic breast cancer a similar percentage, even lower, more specifically 47.9% (35/73) in ctDNA samples. In the independent patient group, PIK3CA hotspot mutations were present at high frequencies (59.5%) in DNA samples isolated from Cell-Search Ò cartridges. In most cases where CellSearch Ò detected at least one CTC, PIK3CA hotspot mutations were also identified. However, in 3 cases PIK3CA hotspot mutations were also detected in samples that were found negative for CTCs according to the CellSearch Ò . These data are in concordance with our previous results, indicating that PIK3CA mutations are also present in EpCAM-positive cells where CellSearch gave negative results.

Conclusions
We performed for the first time a direct comparison study on the presence of PIK3CA hotspot mutations in CTCs and plasma-ctDNA using identical blood draws and a highly sensitive and analytically validated methodology. Our results indicate that: (a) PIK3CA hotspot mutations are present at high frequencies in DNAs isolated from CellSearch Ò cartridges and paired plasma-ctDNA both in early and metastatic BrCa, (b) the concordance between plasma-ctDNA and CTCs is higher in the metastatic setting, (c) PIK3CA mutational status significantly changes after therapeutic intervention, and (d) PIK3CA mutation detection in CTCs and plasma-ctDNA provides complementary information. Our data reflect significant differences in CTCs and ctDNA before and after treatment and point toward the necessity of following patients over time using an integrated liquid biopsy approach. and EL edited the manuscript. AK, AP, DM, and VG collected the clinical specimens. EL designed the study and supervised the experimental procedures.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig S1. Independent group: PIK3CA hotspot mutations in DNA isolated from CellSearch Ò cartridges, before and after treatment from metastatic BrCa patients (n = 17). Table S1. Clinicopathological characteristics of early and metastatic BrCa patients used in the direct comparison study. Table S2. Correlation of PIK3CA hotspot mutations and clinicopathological characteristics of early and metastatic BrCa patients used in the direct comparison study.