Simultaneous Single Cell Gene Expression and EGFR Mutation Analysis of Circulating Tumor Cells Reveals Distinct Phenotypes in NSCLC

Abstract

While cancer cell populations are known to be highly heterogeneous within a tumor, the current gold standard of tumor profiling is through a tumor biopsy. These biopsies are invasive and prone to missing these clones due to spatial heterogeneity, and this bulk analysis approach can miss information from rare subpopulations. To noninvasively investigate tumor cell heterogeneity, a streamlined workflow is developed to scrutinize rare cells, such as circulating tumor cells (CTCs), for simultaneous analysis of mutations and gene expression profiles at the single cell level. This powerful workflow overcomes low-input limitations of single cell analysis techniques. The utility of this multiplexed workflow to unravel inter- and intra-patient heterogeneity is demonstrated using non-small-cell lung cancer (NSCLC) CTCs (n = 58) from six epidermal growth factor receptor (EGFR) mutant positive NSCLC patients. CTCs are isolated using a high-throughput microfluidic technology, the Labyrinth, and their EGFR mutation status and gene expression profiles are characterized.

1. Introduction

The power of targeted therapies designed to target specific molecular vulnerabilities, was shown in non-small cell lung cancer (NSCLC) patients with activating mutations in the epidermal growth factor receptor (EGFR) gene, who responded favorably to tyrosine kinase inhibitors (TKIs) compared to patients with wildtype EGFR.^[1] A patient’s eligibility to receive targeted therapy is determined by molecular information obtained from tumor biopsies. However, false-negatives resulting from spatial tumor heterogeneity and the risk of missing mutant tumor clones from regional tissue biopsies can occur. Additionally, tissue biopsies are not favorable for monitoring the evolution of clonal heterogeneity and resistance with treatment. The ability to monitor patients for the onset of resistance is important for improving patient care. NSCLC patients harboring activating mutations, L858R or exon 19 deletions, in EGFR have a dramatic response to TKI therapy, but develop resistance after about 10 months through secondary mutations, most commonly T790M.^[2–4]

Alternatively, circulating tumor cells (CTCs) and circulating tumor DNA (ctDNA) present in the blood can be accessed through a routine blood draw, termed a “liquid biopsy,” which may better capture tumor heterogeneity.^[5] Both ctDNA and CTCs have been used to monitor disease progression and have shown clinical significance across many cancer types.^[6–10] ctDNA, present in low levels in the blood, typically ranging from <0.1% to >10% of total cell-free DNA content,^[11] is frequently below the detection limit of technologies such as Sanger sequencing and quantitative polymerase chain reaction (qPCR), therefore many groups have focused on next-generation sequencing and more recently digital PCR (dPCR) to detect tumor-specific mutations.^[2,11]

Currently, the Roche Cobas EGFR mutation test v2 is the only FDA-approved liquid biopsy test, used for screening EGFR mutations from ctDNA as a companion diagnostic. However, the sensitivity and specificity of the system is only 58.4% and 80.4%, respectively, for the T790M mutation.^[12] Due to the relatively low efficiency of this test, it is used as a “rule-in” test, meaning if the mutation is not detected, it is not considered absent, and rather a clinician may consider recommending a repeat tumor biopsy before modifying treatment.^[12,13]

While ctDNA has shown promise, there is still debate about using ctDNA for tumor monitoring. ctDNA arises from lysed cells, and therefore represents a bulk snapshot of the tumor. Alternatively, CTCs may provide a real-time view of active disease status, and will likely be enriched for highly aggressive, treatment-resistant live cells. While CTCs are rare in the blood, typically on the order of tens of CTCs per milliliter of blood, many isolation techniques have been developed, most commonly using microfluidic technologies.^[14–16] This enables whole cell analysis, which can include genomic, transcriptomic, and proteomic analysis.

Initial work in CTCs focused on enumeration and bulk analysis for gene expression using techniques such as qPCR or RNA sequencing (RNA-seq). This approach is limited to characterizing the sample as an aggregate of the bulk population, therefore missing rare phenotypes and may be biased by sample purity from leukocytes remaining in the sample.^[14,17]

As a result, several groups have developed workflows for single cell analysis to study tumor cell heterogeneity, commonly through gene expression or genetic profiling. These analysis techniques tend to be labor intensive, time consuming, and expensive. Due to the low starting material many single cell analysis platforms require pre-amplification steps such as whole genome amplification (WGA)^[18,19] for genomic aberrations and whole transcriptome amplification (WTA).^[20,21] An inherent limitation of pre-amplification is the predisposition for PCR bias and gene dropout.^[22] Nonetheless, single cell RNA-seq (scRNA-seq) has emerged as a common method for transcriptomic analysis, important in exploratory studies for prognostic markers. scRNA-seq has revealed intra-patient CTC heterogeneity and signatures of highly metastatic cells.^[5,22] Despite its promises, groups have reported many inefficiencies and technical limitations, leading to few recovered cells proceeding through the entirety of the workflow from recovery, library preparation, and sequencing.^[22] It has been reported that after all these limitations, sequencing only achieves coverage of 15–50% of total transcripts.^[18]

Alternatively, dPCR is a highly sensitive approach facilitating precise and accurate detection and quantification of specific nucleic acid target sequences without the requirement of pre-amplification. In dPCR, individual DNA segments are partitioned into discrete reaction droplets through a water-in-oil emulsion following the Poisson distribution.^[23,24] Each droplet acts as an individual reaction for PCR amplification. The fluorescent intensity of each droplet is then measured in parallel with end-point PCR analysis. This converts data analysis into a binary positive/negative result based on fluorescent intensity, leading to simple data analysis without the need for standard curves for quantification, which could be quickly reported back to a treating physician.^[23,24]

Here, we have established a workflow for rapid single cell profiling of CTCs for simultaneous gene expression and mutation detection to evaluate how tumor subpopulations evolve over time. We screened 58 CTCs from six NSCLC patients with known EGFR mutations and identified intra-patient heterogeneous mutation profiles in these CTCs.

2. Results

2.1. Single Cell Co-Analysis Workflow

The development of a highly sensitive approach could overcome the limitations of liquid biopsies, such as rarity of the targets and low volume constraints, facilitating early detection of resistance mutations. Here, our workflow is optimized for ultra-low input and can consistently detect the presence of these EGFR mutations at the single cell level (Figure 1). To interrogate CTCs with single-cell resolution, we integrated our previously developed inertial microfluidic isolation technology, the Labyrinth^[25] and the commercial Fluidigm C1 integrated fluidic chip (C1 IFC).^[25] First, CTCs are enriched from peripheral blood using the Labyrinth, a high-throughput, inertial microfluidic device, which isolates CTCs from blood cells based on cell size differences. This label-free technology relies on curved channels and sharp corners to efficiently focus both CTCs and white blood cells (WBCs) into separate streamlines. It has been previously shown to yield >90% recovery of CTCs and >90% WBC removal.^[25] Without the requirement of antibody-based capture, this technology isolates heterogeneous CTC subpopulations into a collected cell suspension, and does not involve any additional CTC release step for downstream analysis. To improve CTC purity and facilitate better compatibility with current single cell isolation technologies, the remaining WBCs are depleted using modified immunoaffinity capture microfluidic device^[26] functionalized with a cocktail of antibodies against common WBC targets including CD45, CD15, and CD11b. Finally, the ultra-pure CTC suspension is loaded onto the C1 IFC, which streamlines on-chip single-cell capture, lysis, reverse transcription (RT), and targeted PCR pre-amplification of up to 96 single cells. For pre-amplification and gene expression, we chose a targeted analysis because it has been previously reported that single cell PCR offers all the benefits of bulk qPCR analysis including high sensitivity, specificity, and reproducibility.

Figure 1.

Circulating tumor cell sample processing schematic for single cell analysis. Blood is collected from EGFR mutant NSCLC patients and the blood sample is processed through the Labyrinth, a size-based sorting microfluidic technology to isolate CTCs (≈15–20 μm) from WBCs (≈10–12 μm). The CTC sample is loaded onto the Fluidigm C1 for single cell capture and processing. The resultant single cell sample is then used for gene expression and EGFR mutational profile co-analysis.

Due to the C1 IFC processing protocol, the generated complementary DNA (cDNA) is fragmented due to RT with random primers. This is optimal for the gene expression and EGFR mutation co-analysis because it allows for better coverage of the entire transcript during the reverse transcription step. The pre-amplification on the C1 IFC is comprised of a targeted gene panel (Tables S1 and S2, Supporting Information), using TaqMan-based assays. The amplified region of the gene consists of a short amplicon, while the regions of the EGFR mutations (exons 19–21) remain un-amplified. This provides an untampered view of not only the EGFR mutational burden but also the relative transcription of wildtype and mutant alleles for cells with heterozygous mutations.

To achieve the simultaneous gene expression and mutation detection from a single cell, a small portion (≈10%) of the pre-amplified cDNA was used for gene expression profiling of the pre-designed, targeted 96 gene panel (Tables S1 and S2, Supporting Information) with highly multiplexed qPCR via the Biomark HD Dynamic Array (Fluidigm, USA), as previously shown.^[25] The remaining single cell sample is analyzed for mutations using a duplex assay for the wildtype and mutant-specific sequences on the RainDrop Plus dPCR system (RainDance Technologies, USA).

2.2. Validation of EGFR Mutation Detection Using Digital PCR

Three major EGFR driver mutations, L858R, T790M, and exon 19 deletions, were chosen for this study because of their implications on patient sensitivity to TKI therapy. To validate the performance of the dPCR platform, we detected and quantified the mutant transcripts of bulk cDNA derived from H1975 (L858R/T790M) and H1650 (exon 19 deletion) and A549 (wildtype) lung cancer cell lines. Representative dPCR plots generated using each cell line are shown in Figure 2A.

Figure 2.

Validation of EGFR mutation detection using RainDrop Plus dPCR system. A) Representative dPCR plots of lung cancer cell line controls, H1975 (L858R/T790M), H1650 (exon 19 deletion), and A549 (wildtype), for the three tested EGFR mutations, L858R (left), T790M (middle), and exon 19 deletion (right). For the point mutations (L858R and T790M), TaqMan assays detect the wildtype (VIC channel) and mutant (FAM channel) variants. For exon 19 deletion, the assay screens for 19 common deletions (FAM channel). B) Dynamic linear range of positive droplet counts using serial dilutions of cDNA for L858R (left), T790M (middle), and exon 19 deletion (right) (Range = 0.05–50 ng). Droplets counts for L858R and T790M results used cDNA generated from H1975 cells and exon 19 deletion results used H1650 cells.

As shown, for point mutations, L858R and T790M, gates were used to identify mutant-positive and wildtype-positive droplets, based on H1975 positive control. In these two point mutation assays, there were no false-positive mutant droplets detected in cell lines with wildtype EGFR or containing a different EGFR mutation (Figures S1 and S2, Supporting Information).

For the exon 19 deletion, the assay used only screens for the mutation and contains a pool of 19 common exon 19 deletion variants, which caused a larger population spread, even in control cell lines. This lead to there being a small number of false positive-droplets being consistently generated in negative control cell lines (Figure S3, Supporting Information). To maximize droplets identified in positive controls, and minimize false-positive droplets counted, a quadrant-based gating was used, as shown in Figure 2A.

To determine the dynamic range and sensitivity of the system to detect low input samples, cDNA from positive control cell lines was analyzed using dPCR for the presence of EGFR L858R, T790M, and exon 19 deletions at loadings ranging from 0.05–50 ng (Figure 2B; Figures S1–S3, Supporting Information). Linear regression analysis of the fraction of positive droplets versus cDNA loading exhibited a linear relationship (R² = 0.99) for each mutation, demonstrating a linear dynamic range across three orders of magnitude and approaching single cell sensitivity (Figure 2B).

2.3. Validation of Robust Single Cell Gene Expression and EGFR Mutation Co-Analysis

To establish and validate the single cell co-analysis workflow, H1975 and H1650 cells were processed on the aforementioned C1 IFC chip followed by multiplexed qPCR and dPCR analysis. EGFR mutations were identified on dPCR using the single-cell cDNA product, as shown in Figure 3A–C, Figures S4–S6, Supporting Information. In 13/13 (100%) H1975 single cells EGFR wildtype and L858R mutant transcripts and in 14/14 (100%) H1975 single cells wildtype and T790M mutant transcripts were detected (Figure 3D,E). A heterogeneous expression of total EGFR was observed in H1975 cells with an average of 57 EGFR droplets per cell (range 17–88) for the L858R mutant duplex assay and 41 positive EGFR droplets per cell (range 13–72) for the T790M mutant duplex assay (Figure 3D,E). The H1650 cells tested with the exon 19 deletion assay showed an average of 11 positive droplets per cells (range 3–14). Due to the larger background signal in the assay, a threshold of 11 droplets was used to distinguish between positive and negative exon 19 deletion signal (Figure 3F).

Figure 3.

Validation of single cell workflow for gene expression and EGFR mutation analysis using lung cancer cell lines. Representative dPCR plots of H1975 and H1650 single cells for A) EGFR L858R and B) T790M point mutations and C) exon 19 deletion, respectively. H1975 single cells express heterogeneous total EGFR levels based on the combined mutant and wildtype droplet counts based on D) L858R (n = 13 H1975 cells) and E) T790M assays (n = 14 H1975 cells). F) H1650 single cells express heterogeneous mutant EGFR levels based on the exon 19 deletion assay (n = 5 H1650 single cells). Comparison of relative mutant and wildtype EGFR expression in single cell droplet counts in H1975 single cells using G) L858R (n = 13 H1975 cells) and H) T790M (n = 14 H1975 cells). Ratio of mutant:wildtype droplet counts shown above each cell. I) The average ratio of mutant:wildtype droplet counts in H1975 single cells compared to bulk cells in the L858R and T790M assays. L858R—bulk: n = 12, single cells: n = 13. T790M—bulk: n = 9, single cells: n = 14. J) Hierarchical clustering of H1975 and H1650 single cells show distinct gene expression profiles between the two cell lines and highlight heterogeneous gene expression within each cell line based on 28 genes from Biomark HD multiplexed qPCR. K) The correlation between gene expression and total droplet counts of mutation and wildtype in the same single cells. Each data point denotes a single cell (L858R: n = 10 H1975 cells, T790M: n = 10 H1975 cells). The gene expression and mutation droplet count data of each single cell are normalized to a cell that has the highest gene expression and mutation droplet counts.

In H1975 single cells, the mutant population contained approximately five times higher droplet counts than the wildtype population in both the L858R (ratio range 2.5–9.3) and T790M (ratio range 2.0–13.4) duplex assays (Figure 3G,H). The variations in the mutation-wildtype ratio demonstrate cell-to-cell heterogeneity, highlighting the utility of single cell analysis. However, the average of the H1975 single cells matched the bulk population for both mutation assays (Figure 3I). This higher expression of the mutant transcript is consistent with an increased EGFR mutant copy number in H1975 cells.^[27,28]

Even within a cell line, heterogeneous gene expression profiles with multiplexed qPCR (Figure 3J; Table S1, Supporting Information) and EGFR transcript counts based on dPCR were observed (Figure 3D–F), highlighting the need to dissect heterogeneity of tumor cells at a single-cell resolution. As can be seen, the expression of common NSCLC phenotype markers, such as EGFR, CD44, and epithelial cell adhesion molecule (EpCAM), showed cell to cell variability in each H1975 and H1650 single cells and could be used to characterize phenotypic subpopulations (Figure 3J).

To further confirm the results, the EGFR gene expression measured by qPCR was compared to the detected EGFR droplet count in the same H1975 single cells. The inferred EGFR expression using these two methods demonstrated a strong concordance (Figure 3K).

2.4. Demographics of Patient Cohort

After establishing this single cell workflow using cell lines, we applied this system to analyze CTCs. NSCLC patients were enrolled at the University of Michigan Rogel Cancer Center, under an IRB approved protocol (Table S3, Supporting Information). Six patients with clinical stage IV were enrolled in this study and contained known EGFR mutation based on primary tumor biopsy. Four patients had a single EGFR mutation, one with T790M, and three with exon 19 deletions. The other two patients each contained two mutations, one with L858R and T790M mutations, and the other with exon 19 deletion and T790M. The median age of the cohort was 66 years old (range: 45–78 years) and was evenly divided male and female and ranged from never, former, and current smokers.

2.5. Single Cell Characterization of NSCLC CTCs

CTCs were isolated from six metastatic NSCLC patients with known EGFR mutations (Figure 4) using the previously described Labyrinth microfluidic technology.^[25] A small portion of the CTC sample was used for enumeration with immunocytochemistry (ICC). CTCs were identified as being cytokeratin (CK) positive and CD45 negative (Figure 4A) and had high heterogeneous expression of epithelial to mesenchymal transition (EMT) markers. Some CTCs showed exclusive expression of epithelial marker, EpCAM, while others only showed expression of mesenchymal marker, vimentin (vim). A subset of CTCs showed dual expression of EpCAM and vim, suggesting an intermediate state within EMT (Figure 4A). We observed a wide range of CTC numbers across the patients (range 38.5–201.4 CTCs mL⁻¹ blood) (Figure 4B), and it was noted that the patients who were progressing at the time of blood draw tended to have higher CTC numbers than those who have stable disease (Figure 4C).

Figure 4.

Patient characteristics and CTC analysis. A) Representative images of heterogeneous CTCs and different EMT phenotypes based on EpCAM (epithelial) and vimentin (mesenchymal) ICC. B) CTC enumeration across the 6 patients based on immunofluorescence. (Range = 38.5–201.4 CTC mL⁻¹). C) Summary of patient characteristics including EGFR mutation status based on tumor biopsy, treatment information, and disease status at time of blood draw for CTC isolation.

From the six patients, the remainder of the CTC sample was used for single cell analysis via the C1 platform. The sample was stained on the C1 IFC for CD45, and only CD45 negative cells were processed for single cell analysis. The CTCs showed heterogeneous gene expression profiles (Figure 5A). Interestingly, the genes most commonly expressed in the CTCs were regulators of cell proliferation, such as estrogen receptor 1 (ESR1) and anti-apoptosis, such as b-cell leukemia/lymphoma 2 (BCL2), or differentiation, such as transforming growth factor beta 1 (TGFβ1). TGFβ1, known to be involved in EMT, expression showed a positive correlation with the expression of estrogen receptor (ER) in the CTCs. Interestingly, the CTCs with the highest ER and TGFβ1 also had BCL2 expression, an inhibitor of apoptosis.^[29]

Figure 5.

Patient tumor-matched mutations detected in CTCs. A) Hierarchical clustering showing heterogeneous gene expression of single CTCs from two NSCLC patients (n = 9 single cells). B) Summary of percent EGFR mutant positive single CTCs identified in each of the six NSCLC patients (n = 58). Primary tumor matched mutations detected in patients’ single CTCs for C) L858R (n = 3), D) T790M (n = 8), and E) Exon 19 deletion (n = 8).

*indicates matched gene expression data.

Additionally, individual CTCs were tested for the presence of tumor tissue-matched EGFR mutations (Figure 4C,B). In 5/6 patients, matched EGFR mutations were identified in at least one CTC (Figure 5B). In patients P01 and P03–P06, the CTCs contained patient-matched EGFR mutations, whereas for patient P02, dPCR signals were within the level of uncertainty to confidently classify the CTCs as containing exon 19 deletion. From patient P03, 3/4 (75%) (Figure 5C) of CTCs were positive for L858R mutation, but only 1/4 (25%) of cells tested positive for T790M mutation (Figure 5D). Similar heterogeneity of detection of different tumor-matched mutations was seen in patient P05, 1/2 (50%) of CTCs were exon 19 deletion positive, while 3/12 (25%) (Figure 5E) of CTCs were T790M positive (Figure 5D). We did not observe any correlation between the total CTCs mL⁻¹ and the percent of CTCs that tested positive for EGFR mutations.

For the subset of patients with either point mutations, L858R and T790M, in all mutation-positive CTCs we observed exclusively a heterozygous EGFR mutation expression, with variable mutation to wildtype expression ratios. For all tested cells, there was a higher abundance of wildtype expression compared to the mutant allele (Figure 5C,D).

EGFR expression was below the limit of detection on the qPCR analysis in the CTCs, but was detected using the dPCR (Figure 5). Notably, for patient P06, a subset of the CTCs demonstrated a high level of exon 19 deletion EGFR expression, and also exhibited gene expression profiles consistent with aggressive phenotypes.

3. Discussion

In this work, we describe a multiplexed, single-cell analysis method targeting gene expression and mutation detection from a single cell through the integration of ultra-sensitive technologies. We validated our method to simultaneously quantify mutant transcripts and profile the gene expression of a single cell using lung cancer cell lines. Within a cell line, we observed both heterogeneous gene expression profiles and ratio of mutant to wildtype EGFR expression. This heterogeneity would have been lost using bulk analysis techniques.

We then applied this method to analyze CTCs from EGFR-mutated NSCLC patients. Compared to current clinical testing, this multiplexed method enables rapid turnaround from sample collection to result, in as little as two days. While others have reported the use of single cell dPCR to screen CTCs for EGFR mutations, this work required WGA, and able to identify EGFR mutations in the DNA of single CTCs.^[30] Here, we highlight our workflow to screen single CTCs for mutant EGFR transcripts without the need for pre-amplification, and may reveal clinically-relevant, actionable information. In the case of NSCLC, co-analysis could reveal if specific clones harboring EGFR mutations are primarily utilizing EGFR-related pathways, or if another driver mechanism may be being utilized.

In all tested CTCs positive for L858R or T790M mutations, we observed higher wildtype than mutant EGFR expression. This could be caused by the altered copy number variation profiles within the CTCs or differential gene expression regulation, which have been demonstrated previously in lung cancer cell lines and tumors.^[28,31] This demonstrated the significance of single cell analysis to identify homozygous and heterozygous mutations, which would be lost in bulk analysis. Further, with single cell analysis we could also evaluate the relative expression of each allele. Interestingly, this mutant allele specific imbalance (MASI) is known to occur commonly in EGFR in lung cancer cell lines and tumors, usually with the mutant allele expression favored, but rarely with the wildtype allele expression being favored (reverse MASI).^[28] In cancer cells lines it was shown that those with MASI trended with sensitivity to gefitinib, while the cell lines with no MASI or reverse MASI tended to be resistant to gefitinib, although this relationship needed to be further studied.^[28] Other studies have found that in NSCLC patients with EGFR MASI on TKI therapy had a longer progression free survival, although it was not statistically significant.^[32] In this study, the authors did not include reverse MASI in their analysis. As we found in our study that reverse MASI is more prevalent in CTCs, it is possible that the CTCs with reverse MASI were able to survive longer in the circulation to be isolated, despite the patient receiving TKI therapy, although this would need to be further investigated.

We found some CTCs had increased exon 19 deletion expression in combination with high ESR1 expression. It has been previously shown that ESR1 expression and EGFR mutations tends to occur more frequently together in NSCLC.^[33,34] ESR1 expression in lung cancer has been associated with poor patient prognosis, while ESR2 correlation with patient prognosis appears to be dependent on cellular localization. Estrogen, through estrogen receptor signaling, can activate the signaling pathway downstream of EGFR, phosphoinositide 3-kinases/protein kinase B (PI3K/AKT). This can lead to EMT and promote cancer metastasis.^[34] Additionally, there have been clinical trials evaluating the efficacy of EGFR TKI therapy in combination with estrogen receptor antagonists and in a pilot study treatment was well tolerated and showed efficacy.^[30] Future work in designing combination therapies could include the co-analysis of gene expression and mutational burden of CTCs to measure the induced tumor changes.

4. Conclusion

In this study, we present a combined gene expression and mutation detection of single CTCs from EGFR-mutation positive NSCLC patients. Future studies should include this analysis approach of patients across multiple visits to investigate the changes in the tumor landscape as well as incorporating the screening of additional and resistance mutations through multiplexing the dPCR analysis. Due to the genetic instability and dynamic changes a tumor undergoes throughout treatment, real-time CTC monitoring of tumor evolution could help continually optimize a patient’s treatment plan, improving patient outcome. The single-cell resolution enables the early detection of emerging rare clones that could lead to therapeutic resistance from a routine blood draw, allowing for more predictive analysis of targeted therapy response. The coupled gene expression and mutation profiling using a simple workflow that does not require complex computational analysis could be easily integrated into a clinical setting, enabling real-time monitoring and could ultimately facilitate more timely treatment modification.

5. Experimental Section

Cell Culture:

Cells were maintained at 37 °C under normoxic conditions. Cells were grown to 70–80% confluence before subculturing using 0.05% trypsin-EDTA (Gibco). H1975 (EGFR L858R/T790M mutant), H1650 (exon 19 deletion), and T47D (EGFR wildtype) cells were grown in RPMI-1640 (Gibco), and A549 (EGFR wildtype) were grown in F-12 (Gibco), each supplemented with 10% FBS (Sigma) and 1% antibiotic-antimycotic (Gibco). Media was exchanged every 48–72 h between subculturing. Cell lines were routinely tested and reported negative for mycoplasma contamination (Lonza).

RNA Extraction and cDNA Synthesis:

For cell line experiments, total RNA was purified using miRNeasy mini kit (Qiagen) following the manufacturer’s protocol. RNA concentration and purity was evaluated using a NanoDrop ND-1000 spectrophotometer. For each sample, 2000 ng of total RNA was loaded into each reverse transcription reaction. cDNA was synthesized using SuperScript IV VILO Master Mix with ezDNase Enzyme (Invitrogen) following the manufacturer’s protocol. All purified RNA and cDNA products were handled in a PCR workstation (AirClean Systems) to prevent nuclease contamination.

Experimental Protocol for Labyrinth (Patient Sample Processing):

The experimental protocol was approved by the Ethics (Institutional Review Board) and Scientific Review Committees of the University of Michigan and all patients gave their informed consent to participate in the study. All patients had a diagnosis of EGFR mutant lung adenocarcinoma.

Briefly, blood samples were collected in EDTA tubes and processed through the Labyrinth within 2 h of collection. RBCs in the blood samples were removed using density separation with Ficoll-Paque PLUS Media (GE Healthcare) following the manufacturer’s protocol prior to processing in the Labyrinth.

The plasma and blood mononuclear cells (PBMCs) layers were collected and diluted with PBS (1:5). The diluted sample was collected into a syringe and processed through the Labyrinth at 2500 μL min⁻¹, and the product from outlet 2 was collected into a new conical (Corning). To achieve a higher purity, the second outlet’s products of the Labyrinth (single) were processed through another Labyrinth (double) into a second, new conical (Corning).^[25] To further purify the sample, a glass bottom graphene oxide (GO) chip was used for WBC depletion. The GO chip was made of a gold patterned silicon substrate and a PDMS top layer.

Previously described fabrication and protocol^[26] for sample processing of the GO chip were followed with a few modifications. Briefly, the PDMS top layer was bonded to a standard 1 in. × 3 in. glass slide (Fisher) with plasma surface activation of oxygen. The device was immediately injected and incubated with 3-mercaptopropyltrimethoxysilane (Gelest) for 2 h at room temperature, followed by rinsing with ethanol and adding N-gamma-maleimidobutyryloxy-succinimide (GMBS) (ThermoScientific). After a 30 min incubation with GMBS, the device was then washed with ethanol and treated with NeutrAvidin (Invitrogen). The device was then stored at 4 °C until future use. Before the experiments, cocktails of primary antibodies, including anti-CD11b, anti-CD15, and anti-CD45, were incubated on-chip for 1 h at room temperature. Following antibody incubation, the devices were blocked with 3% bovine serum albumin (BSA) for 0.5 h at room temperature. The second outlet’s products of the Labyrinth were first stained with Hoechst, and then collected into a syringe and processed through the GO Chips at 1 mL h⁻¹ by a syringe pump. The samples were collected into a microcentrifuge tube (Fisher Scientific) from the outlet of the chip for single-cell sorting using Fluidigm C1. With clinical CTCs samples, after the entire CTC enrichment processing, just prior to loading on the C1 system, cell concentration and viability was evaluated using trypan blue and visually inspected under the microscope using a hemocytometer. After sample processing, the immunoaffinity chip with captured WBCs was washed with PBS, fixed with 4% paraformaldehyde (PFA), and stored at 4 °C until imaged with fluorescent microscope.

Immunofluorescent Staining and CTC Enumeration:

The product of single Labyrinth from outlet 2 was processed using a Thermo Scientific Cytospin Cytocentrifuge. A poly-lysine coated slide was placed into the cytospin funnel and 250 μL of sample was added to each cytospin funnel and cytocentrifuged at a speed of 800 rpm for 10 min. Samples were fixed on the cytoslides using 4% PFA and cytocentrifuged at the same conditions as described above. Slide samples were permeabilized by applying 0.2% Triton X-100 solution for 3 min. Slides were then washed with PBS (×3) for 5 min and blocked using 10% donkey serum for 30 min at room temperature.

The panel of antibodies (anti-human CD45 (mouse IgG2a) (Bio-Rad), anti-human Pan-cytokeratin (CK) (mouse IgG1) (Bio-Rad), anti-human EpCAM, biotinylated (goat IgG) (R&D Systems), and anti-human vimentin (rabbit IgG) (Abcam) were used. The slides were then incubated with a cocktail of primary antibodies (anti-CD45, anti-PanCK, anti-EpCAM, and anti-Vimentin) overnight at 4 °C, followed by PBS wash (×3) for 5 min the following day. Slides were incubated in the dark with secondary antibodies goat anti-mouse IgG2a Alexa Fluor 488 (AF 488) (Invitrogen), goat anti-mouse IgG1 Alexa Fluor 546 (AF 546) (Invitrogen), goat anti-rabbit Alexa Fluor 647 (AF 647) (Invitrogen), and Stepavadin, Alexa Fluor 750 conjugate (Invitrogen) for 1.5 h at room temperature. Finally, slides were washed with PBS (×3) for 5 min and mounted using Prolong Gold Antifade Mountant with DAPI (Invitrogen). The stained slides were imaged using a Nikon TI inverted fluorescent microscope at 20× magnification for enumeration.

The tiled images generated from the scans were manually viewed and CTCs were determined based on their fluorescent signals in each channel. A CTC was counted as DAPI+/CK+ (AF 546)/CD45− (AF 488). The CTC phenotype was determined based on the presence/absence of phenotype markers. EpCAM was used an epithelial marker, and vimentin was used as a mesenchymal marker. CTCs (DAPI+/CK+/CD45−) were considered epithelial if EpCAM+/vimentin−, mesenchymal if EpCAM–/vimentin+, and EMT if EpCAM+/vimentin+.

Fluidigm C1 and Biomark HD:

Cell suspensions were loaded onto the C1 Single-Cell Auto Prep IFC for Preamp (10–17μM) (Fluidigm) following the company’s protocol with on-chip cell staining. The cells were stained with FITC pre-conjugated anti-human CD45 (Biolegend), as a negative marker. After loading, each cell capture site was manually imaged.

For targeted pre-amplification, a pre-designed panel of 96 genes implicated in cancer progression, phenotype, and aggressiveness was used to characterize the cells. The pre-amplified product from each of the 96 corresponding capture sites on the C1 IFC is harvested into a 96-well plate (Applied Biosystems). After the C1 run was complete, the sample was diluted using C1 DNA dilution reagent. This diluted product was split for dPCR (10 μL) and Biomark HD systems (2 μL). For the dPCR data generated using lung cancer cell lines, the C1 product was diluted to 28 μL, based on the manufacturer’s protocol. For the lung CTC samples, the C1 product was diluted to only 12–15 μL to keep the sample more concentrated for dPCR testing. 2 μL of this sample was further diluted to 4 μL using the C1 DNA dilution reagent and was used for gene expression analysis on the Biomark HD system (Fluidigm) following the manufacturer’s protocol, while the remainder was used for mutation detection. A limit of detection of 40 cycles was used for data analysis of gene expression profiles of single cell profiling of the lung cancer cell lines and lung CTC samples. For each Biomark HD run, water and no cell controls harvested from the C1 product were used as negative controls to ensure proper sample processing, and did not show any non-specific PCR amplification signal.

Mutation Detection Using dPCR:

The RainDrop Plus dPCR system (RainDance Technologies) was used for dPCR mutation detection. In brief, the PCR mix was prepared using TaqMan SNP Assay (Life Technologies), TaqMan Genotyping Master Mix (Applied Biosystems), and droplet stabilizer (RainDance Technologies). cDNA was mixed with the PCR mix in PCR tubes to generate 25 μL reactions and loaded onto the Source Chip (RainDance Technologies). The PCR reaction was emulsified with carrier oil (RainDance Technologies) into approximately 4 million, 5 pL sized droplets with single molecule loading, and collected into an 8-tube PCR strip (Axygen). After droplet generation, the PCR tubes were transferred to the thermocycler for 45 rounds of PCR amplification. The TaqMan SNP assays contained two probes, one for wildtype EGFR sequence the other for the mutant EGFR sequence, with VIC and FAM probes, respectively. The TaqMan exon 19 deletion assay contained probes for 19 common exon 19 deletion variants, all with FAM probes. The PCR tubes containing the amplified samples were then transferred onto the Sense Machine (RainDance Technologies) where the endpoint fluorescence intensity of each droplet was measured. Gating templates were generated using positive and negative cell line controls.

Statistical Analysis:

Statistical analyses were done using Prism, and error bars were generated based on average and calculated standard deviation. Gene expression analysis was conducted using the SINGuLAR Analysis Toolset (Fluidigm), which was operated through R. Data was compared and visualized using a Gene z-score, which normalizes the expression of each gene in terms of mean and standard deviation.