Patients with metastatic hormone receptor-positive breast cancer express PIK3CA oncogene mutational heterogeneity in circulating tumor cells

We present a single-cell application to determine PIK3CA mutations in CTCs, which uncovered the degree of intra-patient heterogeneity in patients with metastatic hormone receptor-positive breast cancer (HR+ MBC) and high CTC count (>10 CTCs/7.5mL). Using CellSearch and DEPArray we isolated circulating tumor cells (CTCs) and white blood cells (WBCs) from peripheral blood and sequenced PIK3CA exons 9 and 20 by targeted amplicon sequencing. Comparative analysis between the primary tumor (PT, n=27 patients), circulating cell-free DNA (cfDNA, n=31 patients), single (n=146 CTCs) and pools (n=70 CTC suspensions, ranging 5-120 cells/suspension) of CTCs from 26 patients and metastases/DTCs (n=11 patients) was performed. Mutations were frequent in PT (15/27 (55.5%)) and showed slight and substantial agreement with cfDNA (n=21; kappa=0.14) and CTCs (n=22; kappa=0.6733), respectively. A wild-type genotype in WBCs indicates a high speci fi city. Inter-compartmental concordance was observed in 13/18 (72.2%) patients and temporal heterogeneity in 4/18 patients (22.2%). CTC analysis reveals both mutational homo- and heterogeneity with cases showing the presence of di ff erent mutant and wild-type CTC subpopulations. Additionally, unique double-mutated CTCs were detected in 8/26 patients (30.7%). Th e developed liquid biopsy provides an insight into inter and intra-patient PIK3CA mutational heterogeneity in patients with HR+ MBC, paving the way towards the application of a more personalized medicine.


Introduction
Breast cancer is a heterogeneous disease, where specific genetic alterations can be associated with response or resistance to a given therapy [1,2]. The PI3K-Akt-mTor pathway is a validated target in the treatment of breast cancer [3,4]. The catalytic subunit of the phosphoinositide-3 kinase (PIK3CA) is mutated in up to 25% of patients with breast cancer, with mutation frequencies rising to 40% in the hormone receptor-positive (HR + ) subgroups [5,6]. Hotspot mutations in helical and kinase domain results in constitutively enzymatic activity, enhanced tumorigenicity and the capability to form heterogeneous tumors [7][8][9].
Genetic discordances between PTs and the metastasis have been repeatedly observed, emphasizing the need for real-time cancer genoand phenotyping. Since the sequential assessment of biomarkers in metastases is not routinely done, a real-time 'liquid biopsy' with the molecular characterization of circulating tumor cells (CTCs) may constitute an alternative approach [10,11]. In both localized and advanced breast cancer the enumeration of CTCs has demonstrated clinical validity with respect to progression-free and overall survival [12]. In addition, the molecular characterization of these cells may assist in the identification of relevant biomarkers [13][14][15].
Assessment of the inter-patient heterogeneity allows for a useful stratification of patients in histological, molecular and functional subtypes, but is not able to predict the prevailing driving mechanisms of disease progression in an individual patient at any particular point in time. The heterogeneous nature of cancer can be demonstrated by divergence and subtle differences of the molecular profiles between early and advanced disease [16,17]. These observations may be in part explained by the emergence of different tumor cell subpopulations within the primary and/or metastatic sites [18,19]. Clinically, this subclonal development whilst receiving treatment was translated into disease progression and is defined as drug resistance. Therefore, the molecular characterization of the progressing tumor burden and the longitudinal follow-up are an essential prerequisite for personalized medicine to become a reality. This emphasizes the attractiveness of a liquid biopsy using CTCs as real-time input material, which may infer the underlying PT and/or metastases [20,21]. Performing single cell sequencing allows the detection of rare subpopulations, the assessment of the level of genetic heterogeneity and its clonal evolution during therapy [22][23][24].
Within this study we utilized the previously reported methodology to obtain single CTCs by di-electrophoresis [25] and subjected these samples to a PIK3CA mutational analysis by massive parallel sequencing (MPS). We have assessed the degree of mutational

DEPArray purification and whole genome amplification of single cells from CellSearch-enumerated CTC cartridges
CellSave-collected samples were subjected to EpCAM-based immunomagnetical CTC enrichment and enumeration with the CellSearch CTC kit (Janssen Diagnostics, USA). Enumerated CellSearch cartridges were stored at 4°C until CTC cell suspensions were aspirated and transferred into an A300K cartridge for single and groups of cells isolation by di-electrophoresis using the DEPArray system (Silicon Biosystems, Italy), as described previously [25]. Following recovery and volume reduction, single or pools of CTCs or white blood cells (WBCs), 1 ng of unamplified genomic DNA (gDNA) as positive control and reaction mix water (negative control) underwent a whole genome amplification (WGA) using the Ampli1 kit (Silicon Biosystems, IT) [27]. Success of WGA was evaluated using the Ampli1 QC4 kit (Silicon Biosystems, Bologna, Italy).

Primer design and exon-targeted PCR
The MseI restriction, which is proper to the Ampli1 whole genome amplification, was taken into account during the fusionprimer design (Figure1) for downstream PIK3CA amplification and pyrosequencing. Additionally, we assessed the discriminating power of the exon 9 primer pair between the PIK3CA gene and a pseudo-gene on chromosome 22 (NCBI Reference Sequence: NG_027450.1) that has >95% sequence homology [29]. As shown in Table 1, existing target-specific sequences were used for PIK3CA Ex9 [28,29]. Fifty nanograms of extracted DNA from FFPE tissue sections, 2 μL cfDNA from plasma and 1 μL of the WGA product from single and groups of DEPArray-purified cells were subjected to PCR in a final volume of 25 µl containing 1xFastStart High Fidelity Reaction Buffer, 1.6 and 1.4 mM MgCl 2 (for Ex9 and 20, respectively), 400 nM of HPLC-purified fusionprimers (IDT Technologies, BE), 200μM dNTPs, 1.25U of FastStart High Fidelity Polymerase Enzyme (All Roche Molecular Systems Inc., USA) and PCR grade water. Program conditions were: 4 min at 95°C; followed by 30 cycles of 30s at 95°C, 45s at 61.8 or 65.6°C (for Ex9 and 20, respectively) and 1 min at 72°C; and 8 min at 72°C. PCR amplicons were purified with Agencourt AMPure XP beads (Beckman Coulter, Spain) at a ratio of 1:1.6 and stored at -20°C until use.

PIK3CA mutation screening by 454 pyrosequencing
Purified PCR products were measured on the BioAnalyzer (Agilent Technologies, USA), diluted to 1 × 10 9 molecules/µl in 1xTE buffer and equimolarly pooled to create a PIK3CA amplicon library at 1 × 10 7 molecules/µl. Amplicon libraries of the spiked and patient samples were subjected to targeted pyrosequencing in fifteen independent runs. Emulsion PCR of the libraries was carried out using the Lib-A Titanium emulsion PCR (emPCR) Kit, with an input of 0.5 molecules of library DNA per capture bead. In total 500.000 enriched beads were loaded on a Titanium PicoTiterPlate (PTP) and placed in the GS Junior instrument. The libraries were sequenced with Titanium reagents, and base calling was performed with the 'FullProcessingAmplicons' run processor settings in the GS Amplicon Variant Analyzer software (All Roche Molecular Systems Inc., USA). Reads were filtered for mixed and dot reads, adaptor trimmed and de-multiplexed using the 'Both' encoding multiplexer. Filter-passed reads were aligned against the PIK3CA reference sequence (NCBI NG_012113.2) and compared to COSMIC and dbSNP databases.

Statistics and performance analysis
During blood spiking experiments with the reference cell lines models, the level of true/false positives and negatives allowed for the calculation of performance parameters of the assay for PIK3CA mutation detection in single CTCs. Optimal sensitivity and specificity of the variant allele frequency (VAF) stringency were assessed by diagnostic odd ratios [30]. To study concordances and correlations between the occurrence of PIK3CA mutations in PT, plasma and CTC samples, standard statistical analysis techniques were used including Cohen's kappa test (concordance) and Pearson correlation coefficients. A two-sided P<0.05 was considered to be statistically significant. All statistical analyses and graphical representations were performed with SPSS and Microsoft Excel.

PIK3CA PCR assay validation and quality assurance for PIK3CA mutational analysis
Prior to pyrosequencing, PCR assays were validated by direct sequencing of MCF7, BT20 and healthy donor WBC genomic DNA on a 3730XL ABI DNA sequencer (Applied Biosystems, Foster City, CA) using the Big Dye terminator V1.1 DNA sequencing kit. PCR optimization for exons 9 and 20 resulted in 165bp and 301bp amplicons, respectively. Sanger sequencing revealed on-target specificity and identified heterozygously mutant (MT) and wild-type (WT) sequences in genomic DNA from tumor cell lines (MCF7 and BT20) and healthy donor white blood cells, respectively ( Figure 2). We validated our 454 pyrosequencing assay on replicate samples from fresh gDNA, obtained from cell lines (i.e., BT-20, MDA-MB-361, MDA-MB-231, MCF-7, SK-BR-3) and gDNA from externally validated FFPE-patient tissue sections (n=3). All inter-run technical replicates from cell line samples generated expected PIK3CA mutational profiles with a coefficient of variations ranging between 0-5,17%. Duplicate analysis of FFPE tissue sections revealed inter-run reproducibility and delivered concordant PIK3CA genotypes for both exons (Figure 3).

Single-cell PIK3CA mutational analysis in breast cancer cell lines
We assessed the performance of our workflow via independent sorting experiments with spiked tumor cells (TCs) in donor blood. After CellSearch enumeration we isolated single (n=64) and groups (n=9) of MCF-7, BT-20 and SK-BR-3 TCs and WBCs (i.e., internal negative control) by DEPArray. All group TC samples (n=9) obtained after spiking and DEPArray purification consisted of 20 cells per group. Upon WGA the QC PCR revealed a genomic integrity index (GII, scale 0-4) ≥ 1 and ≥3 in 92,2% and 60% of the single TCs, respectively. A GII=4 was obtained in 66,67% and 83,33% of the recovered pools of CTCs and WBCs, respectively. GII≥1 cells were selected for downstream analysis. We tested the ability of our assay to discriminate mutant cells from those with a WT PIK3CA genotype at various VAF cut-off levels ( Figure 4). Filtering variant reads of Ex9 at ≥5% delivered a sensitivity and specificity of 85.3% and 90%, whereas Ex20 reached highest discriminating power at ≥10% with a sensitivity of 71% and specificity of 100%. With the stringency filter set, 18 single and 2 pools of TCs and a pool of WBCs were analyzed and compared to genomic DNA (gDNA) from the parental cell line. Heterozygous WT/MT allelic frequencies were preserved in the DEPArray-recovered pooled TCs ( Figure 5). Single cell samples (n=54) could demonstrate an imbalance in the allelic frequency. A dropout of the expected mutant allele was observed in 7 and 5 cells for Ex9 and 20, respectively. Loss of the WT allele was observed in 6 and 1 cell for Ex9 and 20, respectively. Two false-positive E545A (VAF: 100%) missense mutations were detected, indicating pseudogene amplification. Qualitatively the single cell assay identified 29/36 (80,5%) and 13/18 (72,2%) of the expected mutations in Ex9 and 20, respectively. All SK-BR-3 and WBC samples revealed a WT PIK3CA genotype.

Patient population and sample collection
A blood collection for CTC enumeration was performed in patients with HR + MBC (n=60), before the start of a new line of systemic therapy. We defined a cohort of 38 patients (38/60, 63.3%) with five or more CTCs (Table 2). Median CTC count was 29 CTCs/7.5 mL, with 25 th and 75 th percentile CTC values corresponding to 13 and 182 CTCs, respectively. High CTC counts correlated with the presence of more bone metastases (p=0.024484, χ 2 -test) or with the presence of combined visceral and bone metastases (p=0.017163, χ 2 -test). Apart from CTC biopsy samples, FFPE PT samples (n=27), metastases (n=11) and plasma samples (n=31, CTC-matched) were collected from patients enrolled in this study ( Figure 6). The metastases encompassed nine FFPE tissues, consisting of an abdominal debulking sample (n=1), BM biopsy (n=1), pleural lesion (n=1), skin metastasis (n=2), bone metastasis (n=1) and lymph nodes (n=3). All, except the abdominal and one skin lesion, were temporally matched with the CTC blood draw. Additionally, we analyzed DEPArray-purified DTCs from ascites, pleura and/or bone marrow from two patients (15 single and 11 group samples, covering 139 DTCs in total).

CTC isolation and whole genome amplification
Enumerated CTC cartridges with ≥10 CTCs/7.5mL were selected for DEPArray sorting. Mean sample age (i.e., calculated as the time reveals presence of the missense E542K in exon 9 (upper) and synonymous T1025T mutation in exon 20 (lower) above a variant read frequency of 5 and 10%, respectively. B) 454 pyrograms demonstrate validity of picked-up mutations via comparative analysis of number of bases between reference and read base flows for exon 9 (upper) and exon 20 (lower). C) External quality assurance of sample B112 displaying both E542K (at detection limit of less than 5%) and T1025T mutations by sanger sequencing.  between CellSearch enumeration and DEPArray isolation) was 1.29 ± 0.61 years and did not correlate to overall WGA success rate ( Figure  7B). In total, we isolated 249 single CTCs (mean 8 and median 10 single CTC recoveries per patient), 134 CTC pools covering 2709 CTCs in total (mean 4 and median 3 CTC pool recoveries per patient, ranging from 2-120 CTCs/pool) and 69 WBC recoveries (5 single and 64 pools, n=20 WBCs/pool). CellSearch CTC counts correlated well with DEPArray detection (Figure 8) in the lower (<250 CTCs/7.5 mL) CTC range (Spearman's rho correlation, n=20, r=0.96, p<0.01). A mean recovery rate of 47% ± 22%, ranging from 5.3% to 80% (median 52.2%) was achieved. If calculated towards the theoretical expected value (accounting for the dead volume during cell suspension transfer from CellSearch to DEPArray) a mean recovery rate of 70% ± 35%, ranging from 7.9% to 114% (median 80%) was obtained.
We subjected 340 of 383 (88.7%) CTC recoveries (i.e., 223 singles, 117 pools), 54/54 (100%) DTC recoveries (24 singles, 29 pools) and 69/69 (100%) WBC recoveries (5 single and 64 pools) to WGA. We interrogated the genomic integrity index (GII) of DEPArray-recovered cells by Ampli1 QC4 analysis to identify single and group cell samples suitable for downstream analysis. A representative electropherogram of a WGA quality control experiment is presented in Figure 7, in which the number of detected amplicons (from 0 to 4) determines the GII value. Overall, whole genome amplification succeeded (defined as generating at least one PCR fragment upon Ampli1 QC4 analysis) in 87% and 95% of single and group CTC samples, respectively, and did
Intra-patient mutational analysis revealed the coexistence of both WT and MT CTCs, with the occasional occurrence of CTCs having a double mutated genotype. To explore the degree of this intra-patient mutational heterogeneity we subjected the hotspot loci to an in-depth analysis ( Figure 17). We defined pooled CTC samples, harboring a PIK3CA mutation, as homogeneous since preclinical performance studies demonstrated an imbalance in MT and WT allele amplification, making the assessment of the percentage MT cells within group samples fraught with some uncertainty. The majority of the mutational heterogeneity was detected in single cells. The hotspot genotypes allowed the classification of patients in 4 categories: WT, MT clonal (i.e., >75% of the CTCs having one hotspot mutation), MT subclonal (i.e., >25% of MT CTCs) and mix populations (i.e., displaying a mixture of both WT  Figure 18).

Analysis of metastases at disease progression:
We collected metastases from 11 patients, of which seven FFPE samples were analysed as intra-run replicates ( Figure 19). Three patients (     Focusing on the validated mutational hotspots (P539R, E542K, E54K and H1047R) PIK3CA status in PT and cfDNA or CTC was available in 21 and 22 cases, respectively. Comparison of a metastatic lesion and temporally matched CTCs and cfDNA was possible in 7 and 10 patients, respectively. The level of agreement between the compartments was determined by focusing on the mutational hotspots ( Figure 21). MT or WT statuses from PT tissues showed slight and substantial agreement with cfDNA (n=21; 33% disparity; κ=0.14) and CTCs (n=22; 14% disparity; κ=0.6733), respectively. Comparison of CTCs and temporally matched cfDNA samples revealed slight agreement (n=22; 36% disparity; κ=0.0833). When comparing CTCs and metastases, a proper agreement was observed (n=7; 29% disparity, κ=0.4167).

Comparative analysis of PIK3CA mutational status
PIK3CA data across PT, cfDNA and CTCs was available in 18 cases. In this group 12 patients harbored PIK3CA aberrations in the PT, which were undetected in plasma-derived cfDNA in three cases (3/12, 25%), whereas the corresponding CTCs demonstrated the expected mutation (patients 3301, 889 and 3470). It is worth mentioning that cfDNA samples from patients 3301 and 3470 did carry the expected mutations, however below the selected VAF stringency level. In 3/18 evaluable patients (16.7%) a mutation was detected in cfDNA whereas we did not detect any aberration in the PT and matched CTCs (patient Figure 19. Heatmap of detected variants, presented with the detected variant read frequency. Range: >2% to >50%. Lower: PIK3CA status exon wide (upper) and hotspot-focused (bottom). DTC samples are additionally presented with their genomic integrity score (see color scheme above patient ID). Blue denotes a GII of 4, white denotes a GII of 3, pink denotes a GII of 2 and red denotes a GII of 1. B) Mutation frequencies (%) in metastases (n=11) with E542K and E545K as most frequently detected substitutions. VR% denotes variant read frequency. MT denotes MT. WT denotes WT. MT ≠ denotes inconsistent variants between technical replicates. WT/MT denotes WT or MT genotypes between technical replicates. 3404, 2788 and 3936). Additional bone marrow and pleural DTCs from patient 3936 also revealed a WT genotype. A MT PIK3CA status across all compartments (PT, cfDNA and CTCs) was observed in 11/18 (61.1%) samples. In the six cases where a metastatic lesion was analyzed next to the cfDNA and CTC samples, a concordant PIK3CA status was observed in three cases. In cases 3380 and 3546, CTCs and cfDNA samples missed the MT genotype, respectively, which was present in the respective metastasis. In patient 3705 the WT status from the metastasis was not reproduced in the mutation-harboring CTCs and cfDNA. In general, mutational discordance between early and advanced disease was observed in four cases (patients 1529, 2139, 2648 and 3516). Gain of mutation was observed in two patients, which was confirmed by cfDNA in patient 2648.
In two cases, CTCs were capable to dissect the mutational heterogeneity of the hotspot mutations observed in PT or cfDNA. In patient 3564 we noticed how duplicate analysis of the PT demonstrated an E542K (VAF=38.85%) or E542K (VAF=40.52%) with low frequent E545K (VAF=2.94%) mutation. Duplicate analysis of cfDNA reproduced the double mutated genotype (VAFs>5%). Analysis of CTC samples revealed an overall homogenously E542K-mutated population, with one CTC harboring the double mutated genotype.
The same feature was seen in patient 3470, who's PT revealed a E545K (VAF=38.85%) or E545K (VAF=29.45%) with a low frequent H1047R (VAF=3.36%) mutation. These aberrations were present in one duplicate cfDNA sample. A single and two pools of CTCs revealed the existence of a mutually exclusive E545K and H1047R-positive subpopulations. Additionally, we identified a single CTC harboring the double-mutated E545K/H1047R genotype.

Discussion
We present an in-depth validation, feasibility assessment and clinical application of PIK3CA mutation detection in primary tumors, CTCs, cfDNA and metastases from patients with HR + MBC. All analyzed metastases samples (both tissue and disseminated tumor cell effusions) were temporally matched with the CTC blood draw, except the abdominal and one skin lesion. We previously reported the use of CellSearch and DEPArray to obtain pure CTCs (with a CellSearch to DEPArray recover efficiency of 62±12%) for molecular characterization down to the single cell level, and has demonstrated its applicability in different cancers [13,25,27,[31][32][33][34]. Using the FDA-cleared CellSearch CTC counts has proven its applicability as a prognostic biomarker in clinic. However, the technique is based on positive selection of EpCAM + cells, which may have missed tumor cells that underwent epithelial-to-mesenchymal transition (EMT). Recently, Sawada et al. have demonstrated the enhanced capture and enumeration efficiency of a novel marker-independent fluidic devise, when comparing to CellSearch. In patient samples, their platform detected significantly more CTCs than CellSearch, and detected CTCs in 9 cases, who were negative according to CellSearch [49]. Exploring PIK3CA mutational heterogeneity in CTCs obtained by marker-independent techniques would be a logical extension of this study.
We theoretically calculated that a CellSearch cartridge with 5 CTCs would allow us to minimally recover 2 and maximally 4 CTCs on DEPArray. However, due to variability in sample age and nature, keeping in mind the WGA unsuccessfulness in 7.8% of the spiked single cells, we empirically selected a subset of patients having more than 10 CTCs, to have at least 1-2 cells analyzed. In our cohort of 60 patients with HR+ MBC, 29 patients (ca. 50% of the total population) had equal to or more than 10 CTCs/7.5mL. We selected 26 patients for PIK3CA mutational analysis in whom it was feasible to determine the PIK3CA status in at least one CTC. CTCs were subjected to MPS of PIK3CA Ex9 and 20 after WGA, using a specific Ex9 reverse primer, which avoids pseudogene amplification [29]. Within this study we determined a rationale for a VAF stringency filter for the analysis of NGS data from single CTCs. By spiking breast cancer cells, carrying heterozygous PIK3CA mutations, and subjecting these cells to the same workflow of CTC enrichment, purification and WGA, we determined a VAF cut-off value for a reliable qualitative assessment of hotspot variants by calculation of the sensitivity and specificity for various VAF cutoff levels. Single cell data from spiked TCs revealed an WT/ MT allelic imbalance compared to fresh gDNA, which was also noted, albeit to a lesser extent, in pools of CTCs. The goal of our workflow was to assess whether a sample harbors a PIK3CA mutation or not. We report an imbalance in the detected variant allele frequencies when performing the mutational analysis on small pools of cells or on a single cell level. The origin of this imbalance is uncertain but might be traced back to manipulation steps, encompassing a cell fixation, permeabilisation and staining procedure. Furthermore, on average CellSearch cartridges were stored for 1.5 years at 4°C, upon which cells were subjected to dielectrophoresis and downstream WGA, which for all reasons described above could have resulted in bias. This observation has also been described in the recent report by Gash and colleagues. Regardless of which WGA method was used, a preferential amplification of one allele was noted in a subset of the analyzed CTCs [48]. Due to the variability and sometimes low TCC of PT, metastases and cfDNA samples, compared to purified CTCs, we empirically defined a VAF cut-off for these matrices, based on the coverage depth of the performed sequencing experiments. Additionally, we need to emphasize that when analyzing cfDNA samples we did not correct for WT cfDNA originating from lysed white blood cells between samples, which could have resulted in the poor concordance between PT/CTC and cfDNA.
Besides hotspot we detected other PIK3CA mutations within the coding sequences. These low-frequent and rare substitutions were detected in unique samples, and may in some occasions have gain-offunction activity [35,36]. We emphasize the nature of the PFA-fixed cells and tissues, which may have affected the DNA quality, resulting in fixation bias and subsequent sequencing errors [37]. Hence, this raises the question of false discovery due to technical errors. Therefore, lack of validation or reproducibility renders uncertainty regarding the validity of these rare variants. In this way it is unclear why there was variability in some duplicate primary tumor samples. During the sequencing experiments we occasionally performed a no template control (H 2 O used to generate PIK3CA sequencing libraries), which did not deliver any DNA sequences. Additionally, internal negative control samples (i.e., patient WBCs co-purified with CTCs) revealed a wild-type PIK3CA genotype on all occasions. However, we noticed how these inconsistent results were achieved during sequencing of primary tumor samples, which were performed in two independent sequence runs, of which one sequence run only delivered unidirectional reads, instead of bi-directional coverage of the targeted exons, which may form the basis for the observed inconsistent results. On the contrary, we validated our methodology to detect and reproduce hotspot P539R, E542K, E545K and H1047R mutations by different approaches. We analysed several breast cancer cell lines harboring a known PIK3CA WT or MT status, both on bulk and single cell level. We reproduced the mutational status of a FFPE tumor tissue section that was reported by an external laboratory to carry two distinct PIK3CA mutations, and finally we performed cross validation of obtained results for selected patient CTC samples by whole exome sequencing (WES). Also, to validate the somatic nature of the non-hotspot mutations we sequenced germline DNA from pools of white blood cells, both in the assay validation and patient analysis phase. Here, all WBC samples generated wild-type PIK3CA genotypes, which could indicate the true somatic nature of the non-hotspot picked-up variants.
We selected PIK3CA as it is the most prevalent mutated oncogene in HR + breast cancer with reported mutation frequencies in up to 40% of patients [38]. Over 80% of these alterations occur in the helical (Ex9) and kinase (Ex20) domains [39]. We chose to study the mutational heterogeneity in these hotspot loci by analysis of single CTCs as realtime liquid biopsy in patients with metastatic breast cancer. CTC analysis showed mutant PIK3CA genotypes in 21/26 (80%) patients, whereas previous studies have reported mutation frequencies ranging from 15% to 37% [27,33,[40][41][42]. This higher rate could be in part explained by selecting a pure HR + MBC cohort with a CTC count >10 CTCs/7.5 mL. This resulted in the interrogation of 147 single and several pools of CTCs, covering total of 1036 CTCs from 26 patients. The increased CTC sampling size may have enhanced our sensitivity to detect low frequent MT cells.
The concept of PIK3CA mutational heterogeneity in metastatic breast cancer has been described by other research groups, where different levels of intra-patient heterogeneity were observed [27,33,[40][41][42]. Here we performed mutational analysis on patients, all having a HR+ subtype at diagnosis and having advanced disease at the moment of blood draw, making it a more homogenous target population. Additionally, we want to remark that here we are among the first to demonstrate the dissection capabilities of intra-patient heterogeneity in mBC by comparative analysis of single circulating tumor cells to the primary tumor, circulating cell-free DNA and tissue and DTC metastases. Most recently, Gash et al. observed similar mutational patterns as the presented study with 4 cases having both PIK3CA wildtype and mutant subpopulations. Overall, the researchers concluded that mutational heterogeneity was frequent, as is the case in our presented study [48]. Additionally, different PIK3CA mutations were detected in asynchronously collected blood samples from the same patient. In our study, temporal heterogeneity was observed in four patients, with gain of mutation detected in two patients. This has also been observed in studies where comparison of asynchronously collected patient samples revealed mutational discordance in 7% up to 40% of the cases [43][44][45]. Besides temporal, mutational heterogeneity was detected between individual, synchronous collected CTCs, harboring different mutations or even double mutated and WT genotypes. Interestingly, it was reported how mutational discordance may occur in CTC samples from temporally-matched blood samples within a particular patient, indicating how one needs to approach PIK3CA mutational analysis by a single blood draw with caution [40,41]. We report how PIK3CA mutations are frequent and may occur in a homo-or heterogeneous manner in synchronously-collected CTCs. However, the relatively low number of patients analyzed limits the possibility to draw strong conclusions with regards to mutation incidence and frequency of mutational discordance between primary tumor, CTC, ctDNA and metastatic lesions. Therefore, the observed degree of heterogeneity emphasizes the need for an in depth single and pooled CTC analysis in the metastatic setting, via larger prospective clinical studies.
Single cell analysis allows the identification of minor CTC subpopulations. Besides the hotspot and rare PIK3CA alterations, we also observed patient samples harbouring a double mutated genotype within the hotspot loci. These events are rare, but indicate the heterogeneous nature of the PIK3CA oncogene [46,47]. Here, we observed two cases where the molecular heterogeneity of the PT and plasma samples were dissected by single and pooled CTC analysis. We identified how a CTC population with a double mutated PIK3CA or heterogeneous CTC subpopulations each containing one mutation may co-exist next to each other. Therefore, the exploration of the PIK3CA mutational heterogeneity demonstrates how within a given PT minor PIK3CA MT subclones can exist, which can be detected in circulation at an advanced stage of the disease. Performing the analysis on small pools of CTCs or down to the single cell level has allowed us to demonstrate the existence of these independent subpopulations. Our study provides an insight into the complexity of the mutational heterogeneity of PIK3CA in breast cancer at an advanced disease stage. We observed how the presence of PIK3CA mutations in CTCs is not associated with survival in patients with HR + MBC, which is in contrast with previous reported results [42]. Interestingly we did observe how the presence of clonal PIK3CA mutations in CTCs is associated with worse survival.
We provide a validated and sensitive MPS method to determine PIK3CA mutations in CTCs, allowing the study of mutational heterogeneity. The presented liquid biopsy can be performed in high throughput on several single and pools of CTCs with 100% purity from one or several blood samples, making it suitable for use in a diagnostic setting. We provide supportive evidence that PIK3CA hotspot mutations are frequently present in EpCAM-positive CTCs from patients with HR + MBC and may occur in a homo-or heterogeneous manner, which needs to validated via larger prospective clinical studies.
Besides mutational heterogeneity we observed temporal discordance when comparing early to advanced disease. Using our methodology, sequencing depth and VAF stringency levels, mutational analysis of CTCs may resemble better primary and metastasis tumor tissue sections, compared to cfDNA samples from plasma. However, we do emphasize the sequencing depth of our cfDNA samples and high VAF cutoff level. Nonetheless, using a mean sequencing depth of 1600x and a VAF cutoff of 2% allowed us to detect PIK3CA hotspot mutations in plasma from patients with advanced MBC which in 67% of the cases showed concordant results with the PT. The low sample input, sequencing depth and stringent VAF filtering may lay at the basis of the observed discordance. However, we want to emphasize that we targeted a patient population with advanced metastatic disease, having a high circulating tumor burden, in whom perhaps a lower sequencing depth is sufficient to detect clonal driver mutations, as the PIK3CA oncogene.
In conclusion, the study presents the utilization of a liquid biopsy on a single cell level, which dissected the PIK3CA mutational heterogeneity within the intravascular compartment of individual patients, thereby paving the way towards the the successful implementation of the liquid biopsy model as part of the personalized medicine in the management of patients with metastatic cancer.