A Microfluidic Liquid Biopsy Platform to Monitor Protein Biomarker Heterogeneity in Single Circulating Therapy‐Resistance Cancer Cell

Tumor cells display heterogenous molecular signatures during the course of cancer and create distinct tumor cell subpopulations which challenge effective therapeutic decisions. Detection and monitoring of these heterogenous molecular events at single cell level are imperative to identify tumor cell subpopulations and to engage the best therapeutic options for the individual patient. Herein, a microfluidic liquid biopsy platform to analyze circulating tumor cells (CTCs) at single cell level is reported. The individual CTCs are captured in an alternating current‐induced microfluidic platform and analyzed by using surface‐enhanced Raman scattering spectroscopy. This platform selectively captures single CTCs from the patient's peripheral blood mononuclear cells. Using cell line models and patient samples, it is shown that the assay can simultaneously detect multiple protein biomarkers on a single CTC. The platform can stratify the CTCs into different subpopulations based on their cancer‐associated protein signature changes in response to drug treatment. This enables the identification of CTC subpopulations that are probably not responding to treatment and may assist clinicians in specifically monitoring and eliminating therapy‐resistant cancer cells within a lesion. This single CTC monitoring chip will likely have high clinical importance in disease diagnosis and treatment monitoring, and advance the knowledge of cancer heterogeneity.


Introduction
Treatment efficacy in cancer depends on multiple factors, including the identification of best druggable targets and monitoring of drug response over time.However, the dynamic nature of the cancer progression cascade and tumor heterogeneity significantly compromises treatment efficacy, as cancer cells modulate their internal signaling pathways to survive drug treatment, resulting in drug resistance and disease progression and/or recurrence. [1]Recent research suggests that circulating tumor cells (CTCs) shed from the cancer mass to the blood can serve as a potential cancer biomarker and provide vital information about the dynamic changes in cancer cell's molecular profile within the tumor microenvironment during cancer progression as well as in response to the therapy. [2]In this regard, a liquid biopsy platform capable of profiling CTCs down to a single-cell level is highly desirable to reveal heterogenous molecular information of CTCs that would reflect the status of the parent tumor.However, detection and molecular profiling of CTCs are very challenging and require highly sensitive and multiplex techniques due to the low CTC abundance and their heterogeneity in blood. [3,4]19] However, characterizing single-cell heterogeneity based on their surface protein expression levels is still at an early stage. [20][23][24] However, these techniques require a large sample volume, and considering the rare occurrence of CTCs in bodily fluids, these techniques are not the best option for analyzing single CTCs in liquid biopsy. [25]n the past decade, significant advancements in miniaturized biosensors, especially in microfluidic platforms, have enabled the successful isolation of rare biomarkers, including CTCs, proteins, and extracellular vesicles, from complex biological samples. [26,27]33] In this study, we developed a single CTC analysis platform capable of concurrently identifying multiple surface proteins in single CTCs and showing that the expression levels of these proteins are highly heterogenous between CTCs and very dynamic during treatment.The platform consists of a device that is capable of fluidic flow manipulation at the nanoscopic scale and an SERS readout for detecting multiple protein biomarkers on the surface of a single cell.We selected melanoma cell lines and a nonsmall cell lung cancer (NSCLC) cell line as our model system for our initial experiments.The cell line model data showed that our platform could specifically capture single cancer cells and identify all the target biomarkers (melanoma-chondroitin sulfate proteoglycan (MCSP), melanoma cell adhesion molecule (MCAM), lowaffinity nerve growth factor receptor (LNGFR), and erythroblastic leukemia viral oncogene homolog 3 (ErbB3) for melanoma; programmed death ligand 1 (PD-L1) and epidermal growth factor receptor (EGFR) for lung cancer) on a single-cell surface.Distinct expression of target proteins on individual cell surfaces for cells under drug treatment revealed that each cell was different and responded differently to a specific drug.Most importantly, our platform identified the subpopulation of cells demonstrating higher biomarker expression profiles during drug treatment.Our patient sample data also showed that the CTCs were highly heterogenous during treatment and our assay identified the subpopulation of CTCs showing higher biomarker expression during treatment.This might help clinicians to select the right drug for the nonrespondent CTCs as well as investigate the underly-ing mechanism of the drug resistance and find a potential combination therapy for better patient survival.Finally, our platform provides a robust and sensitive system for quick and multiplex detection of single CTCs and their heterogenous profile in liquid biopsy.We believe that this miniaturized platform for multiplex protein analysis of single CTCs has a high potential for clinical application.

Results
A schematic workflow of our single CTC analysis platform is presented in Figure 1a.The operating principle of the platform is based on a sandwich immunoassay, where CTCs are captured on antibody-functionalized array of electrodes and labeled with SERS nanotags.The array of gold electrodes is interconnected and was patterned on silicon substrate.To facilitate sample flow over the electrodes, a polydimethylsiloxane (PDMS) layer containing a parallel microchannel structure was thermally bound to the silicon underlayer (Figure S1b, Supporting Information).The sample is introduced into the microfluidic channel and CTC are captured on the antibody-functionalized electrodes.The electrodes were functionalized with antibodies using biotin-streptavidin chemistry. [34,35]Previous reports suggest that melanoma and lung cancer-derived CTCs predominantly express MCSP and EGFR proteins, respectively. [8,36]Thus, we used anti-MCSP and anti-EGFR as our capture antibodies to capture the melanoma and lung cancer cells/CTCs for both cell line and patient samples, respectively.An ac electrohydrodynamic (ac-EHD) effect (described in Section 2.1) was applied during the capture process of the CTCs to induce fluid mixing within the nanometer range on the electrode surfaces.This ac-EHD nanomixing can improve target cell capture on the antibody-functionalized electrodes and can reduce binding of loosely bound nonspecific cells. [34,37]Following CTC capture, a solution containing different SERS nanotags that carried antibodies against target protein biomarkers and Raman reporters was passed through the microchannel to label the captured CTCs and complete the sandwich immunoassay.The SERS nanotags bound to CTCs were excited with the laser to collect SERS spectrum of Raman reporters (Figure 1c) and infer about biomarker expression levels.We hypothesized that by longitudinally tracking changes in protein biomarker expression on CTCs and tracking CTC heterogeneity, potential insights about CTC response to drug treatment might be obtained.
Compared to other techniques based on immune affinity using fluorescence detection, the integration of SERS to the microfluidic device has multiple unique advantages. [38,39]Different to the widely used fluorophore-based assays, the Raman reporters have narrow spectral peak widths and high photostability that can be used for in situ multiplexing on single cells using a single laser for excitation.Furthermore, our device facilitated the direct capture of CTC in peripheral blood mononuclear cell (PBMC) without requiring white blood cell deletion (i.e., CD45 depletion) that is often required in other immune affinity-based methods. [40,41]Upon capture of the CTC on the gold electrodes of the device, Raman mapping of single cells enables the analysis of protein expression and expression heterogeneity.This is advantageous compared to flow cytometry that typically requires  For our preliminary experiments, we used melanoma SK-MEL28 and NSCLC HCC827 cell lines.In the melanoma model study, we selected a panel of four melanoma cell surface markers which included MCSP, MCAM, ErbB3, and LNGFR.[44] LNGFR is a potential factor of melanoma tumor stem cells with a high propensity to establish tumors and is often associated with resistance development. [45][48] Specific SERS nanotags were used to achieve the characteristic peak intensities for each of the target proteins such as 5,5′-dithio-bis-[2nitrobenzoic acid] (DTNB, 1337 cm −1 ) [49] for MCSP, mercaptobenzoic acid (MBA, 1075 cm −1 ) [50] for MCAM, 7-mercapto-4-methylcoumarin (MMC, 1172 cm −1 ) [50] for LNGFR, and tetrafluoro-4-mercaptobenzoic acid (TFMBA, 1375 cm −1 ) [50] for ERBB3.The colors in Figure 1d-f denote the attached SERS nanotags (i.e., red = MCSP, green = ErbB3, blue = MCAM, and sky blue = LNGFR) on the single-cell surface (scale bars are 7 μm) through which different protein expression levels were detected.The irregular shapes of the captured cells in the images might be due to the drying effect during the imaging process. [51]wever, this does not affect our conclusion.For the NSCLC cell line model, we selected PD-L1 and EGFR proteins which were reported to be predominantly expressed in lung cancer cell CTCs. [52,53]

ac-EHD Mechanism and Optimization of the Operating Parameters
The ac-EHD phenomenon is first reported by Brown et al. [54] and subsequently, our research group has extensively utilized this nanomixing characteristic within microfluidic channels for enhancing biomolecular interaction with the captured molecules embedded on electrode surfaces for better specificity and sensitivity. [34,37]Briefly, upon application of an ac pulse (ac-EHD) across the asymmetric electrode pair embedded fluidic channel filled with ionic solution (i.e., sample fluid), a nonuniform charge double layer forms within a few nanometers of the electrode surfaces.This asymmetry in the charge double layer near the small and large electrode surfaces results in a gradient force that drives the ions across the electrode surfaces and induces a fluid flow from the smaller to larger electrode direction.
We initially determined the optimal conditions of the ac-EHDenabled single-cell analysis platform to achieve maximum cell capturing on the electrode surface.Because at low operating parameters (frequency and potential), the cell capturing might be affected due to nonspecific adsorption while at high operating parameters, the resulting force generated on the electrode surface may damage the sample and device which eventually could result in poor device performance.Therefore, we investigated different ac-EHD conditions (e.g., ac frequency, potential, and incubation time) so that the engendered force becomes adequate for capturing maximum cells by enhancing nanomixing, at the same time to shear off nonspecific molecules from the electrode surface, allowing better specificity of the system.As can be seen in Supporting Information Figure S2a, we functionalized three individual devices with 50 SK-MEL28 cells and applied three different EHD conditions (such as 500 Hz + 120 mV; 1000 Hz + 120 mV; 1500 Hz + 120 mV) for 10 min.The revealed maximum capture efficiency (≈85%) was achieved with 500 Hz frequency.In parallel, we also optimized the optimum potential to achieve maximum capture efficiency.As depicted in the Supporting Information Figure S2b, we functionalized three individual devices with 50 SK-MEL28 cells and applied three different ac-EHD conditions (such as 80 mV + 500 Hz; 120 mV + 500 Hz; 160 mV + 500 Hz) for 10 min.Here, we found maximum capture efficiency (≈85%) when we applied 120 mV potential.Moreover, we applied the optimized frequency and potential (500 Hz + 120 mV) to determine the optimum ac-EHD application time.As shown in Supporting Information Figure S2c, maximum capture efficiency (≈85%) was achieved with 10 min of incubation period with the optimized parameters (500 Hz + 120 mV).Moreover, we have performed sets of experiments for target capture within the microfluidic system under no ac-EHD condition and compared the result with target capture at optimized ac-EHD condition (Supporting Information Figure S2d).These data clearly demonstrate the influence of ac-EHD for increased capture efficiency compared to static incubation within the microfluidic system.

Specificity Analysis
To demonstrate the specificity of our assay, we used Melanoma SK-MEL28 cells and all four target biomarker proteins for melanoma.50 target cancer cells were passed through the chip functionalized with anti-MCSP capture antibody followed by four secondary antibody-conjugated SERS nanotags.A series of experiments were carried out under the optimized conditions (500 Hz + 120 mV) using i) a positive cell line (melanoma cell line, SK-MEL28 cells) + melanoma-specific capture antibody (anti-MCSP), ii) SK-MEL28 cells + anti-EPCAM as capture antibody (nonspecific to SK-MEL28 cells) instead of anti-MCSP, iii) a negative cell line (breast cancer cell line, SKBR3 cells) + anti-MCSP capture antibody (nonspecific to SKBR3 cells), and iv) no capture antibody.The selected capture antibody anti-MCSP (positive control) specifically captures/binds only the melanoma cells (SK-MEL28), not the breast cancer cells (SKBR3) while the other capture antibody anti-EPCAM (negative control) is specific only for the breast cancer cells, not for the SK-MEL28 cells.As shown in Figure 2a,b, the positive experiment with SK-MEL28 cells provided a strong signal for each of the target protein biomarkers for melanoma.On the other hand, the negative control experiments provided negligible signals.These data suggest that our platform is highly specific in capturing MCSP-positive cancer cells and identifying the biomarker protein expression within those cells.
We believe the high specificity of our assay induced relies on two factors: 1) ac-EHD nanomixing fluid flow increases collision between target cells and antibodies, and hence facilitates more capture.2) ac-EHD nanomixing fluid flow also facilitates strong binding and shears off the loosely bound nonspecific molecules from capture domains and therefore minimizes nonspecific adsorption.These data fully agree with our previous reports on the high specificity of ac-EHD platforms and therefore suggest that our platform is highly specific in capturing MCSP-positive cancer cells and identifying the biomarker protein expression within those cells. [8,34,37]Furthermore, we also validated the selectivity of our capture antibody anti-MCSP in capturing melanoma cells using the flow cytometry assay and found an excellent agreement with the SERS findings (Supporting Information Figure S3a).

Efficiency and Sensitivity
To demonstrate the efficiency of our assay in capturing a low number of cells, we took different numbers of target melanoma SK-MEL28 cells (50, 100, 150, 200, and 250) and passed them through our device.As shown in Figure 2c, the overall cell detection efficiency was higher than 80% for all the different numbers of target cells.We also found that the number of captured cells increased with the increase of inputted cell numbers (Supporting Information Figure S3b).The linear increase in SERS intensity with the increase in cell numbers indicates a high consistency of our detection approach.However, it is noted that the cell detection efficiency might be lower in blood samples where the assay is challenged by the presence of high concentration of other cells such as white blood cells.As a demonstration, we investigated the capability of the device to capture and detect CTC in a complex sample.We spiked 50 melanoma cells drug untreated (day 0) or drug treated (after 1 week) into 100 μL of PBMC and enumerated the number of captured cells after running the assay.As shown in Figure S3c,d in the Supporting Information, around 60% melanoma cells were captured.Subsequent single-cell analysis of the captured melanoma cells showed heterogenic expression of MCSP.The reasonably high recovery and relatively low standard error suggest the capability of the device to successfully capture and detect single CTCs in complex samples.
To demonstrate the sensitivity of our assay in analyzing single cells and to investigate their heterogeneity, we used melanoma SK-MEL28 cells as our target cells and MCSP as the target protein biomarker.The strong Raman signal shown in Figure 2d suggests that our assay successfully captured single cells and detected their surface marker expression levels.Each dot in Figure 2d represents the expression levels of MCSP protein biomarker on individual cancer cells.These data clearly demonstrate that each cell is different, and cells are highly heterogenous in terms of their surface protein biomarker expression.

Mapping On-Chip Cell Heterogeneity (Bulk vs Single-Cell Analysis)
To demonstrate the importance of single-cell analysis and identifying cellular heterogeneity, we performed a comparative analysis between bulk and single-cell analysis based on the expression of target biomarkers of interest.In this case, we used melanoma SK-MEL28 cells and MCSP, MCAM, ErbB3, and LNGFR as our target biomarkers.We passed 50 SK-MEL28 cells through our device and measured their bulk and single-cell expression of these four biomarkers.Figure 3a represents the average Raman intensity of the four surface proteins expressed in captured SK-MEL28 cells.The bulk data show that the average Raman intensity for the four markers is 1520 (±205) for MCSP, 1140 (±175) for MCAM, 482 (±50) for ErbB3, and 336 (±65) for LNGFR.However, one of the major limitations of the bulk analysis is that it cannot address cellular heterogeneity which is highly desirable for personalized treatment.Our single-cell data, shown in Figure 3b, clearly articulate the heterogeneity in the four-target biomarker expression in individual cells.For instance, the diversity of cells in expressing MCSP protein is significant and the Raman signal intensity ranges from 200 to 6000 a.u.Since MCSP plays a role in tumor progression and therapy resistance, the cells with higher MCSP may have a more active role in tumor progression and therapy resistance.Our single-cell (Figure 3c-e) analysis can identify these subpopulations (e.g., cells with signal 200-6000 for MCSP) which cannot be identified with bulk analysis.

Mapping On-Chip Cell Heterogeneity (Identifying Drug Nonrespondent Cells)
To map cellular heterogeneity during treatment, we treated the BRAF-mutated melanoma SK-MEL28 cells with PLX4720 and the lung cancer HCC827 cells with Erlotinib.PLX4720 is a BRAF inhibitor which selectively inhibits mutated BRAF present in ≈50% of melanoma. [55]Erlotinib is a tyrosine kinase inhibitor (TKI) which blocks EGFR pathways in NSCLC. [56]We treated our cells for 5 weeks and tested their expression levels on a weekly basis.As a "control study" (Figures S4 and S5, Supporting Information), we also tested the expression of the cells without drug treatment on a weekly basis for 5 weeks (untreated control cells).The bulk analysis as well as the average SERS intensity data in Supporting Information Figures S4 and S5 shows that the expression of MCSP, MCAM, ErbB3, LNGFR, EGFR, and PD-L1 was decreased during the treatment in comparison to the pretreated cells (day 0), while the expression levels of the biomarkers were almost unchanged between the pretreated (day 0) and untreated (control) cells.To check the statistical significance of treatment response during the drug treatment period, we performed Kruskal-Wallis test analysis of melanoma cells (Figure S6a-d, Supporting Information) and lung cancer cells (Figure S6e,f, Supporting Information).This nonparametric approach of analysis of variance (ANOVA) represents the median expressional pattern of i) MCSP, ii) MCAM, iii) ErbB3, iv) LNGFR, v) EGFR, and vi) PD-L1 on cell surfaces at different periods of drug treatment.In most cases, we found that the difference in biomarker expression for week 0 is statistically significant in comparison to the week 1, week 2, and week 3.The median intensity of all the melanoma biomarkers reduced in the first 2 weeks of treatment, and then started to increase from week 3 and continued until week 5.These bulk data indicate the occurrence of drug resistance during treatment.Furthermore, we performed a set of flow cytometry experimentations using the same pretreated and drugtreated SK-MEL28 cells to cross-validate the data acquired on our platform (Figure S7a, Supporting Information).We also crossvalidated the pattern of PD-L1 expression during the drug treatment of lung cancer cells using the flow cytometry assay (Figure S7b, Supporting Information).However, the bulk data could not identify the subpopulations of the resistant or persistent cells and neither their protein expression levels and thereby unable to provide clear understanding of the resistant pathways for better treatment suggestions.
Our single-cell analysis data shown in Figure 4a-f and the heatmap data of each cell expressing all the biomarkers for melanoma and lung cancer (Supporting Information Figure S8a,b, respectively) suggest that there was significant heterogeneity in the protein biomarker expression levels during each week of the drug treatment.If we consider the bulk data, the median SERS intensities for the biomarker expression level were ≈400 a.u.for MCSP and MCAM, ≈200 a.u.for LNGFR and ERBB3, ≈1000 for EGFR and PD-L1 after 1-2 weeks of treatment (Supporting Information Figure S6).However, our single-cell analysis data (Figure 4 and Supporting Information Figure S8) show that there was significant number of cells showing higher SERS intensities than the bulk SERS intensities for those biomarkers after 1-2 weeks of treatment.Furthermore, we analyzed individual cells based on all the target biomarkers expressed on each cell surface by applying linear discriminant analysis (LDA).As shown in the LDA analysis in Supporting Information Figure S9, the day 0 data are scattered, which indicates different phenotypes/heterogeneity of cells in biomarker expression.The data at week 1, 2, 3 closely grouped together but started to further scatter at week 4 and 5.The nonparametric one-way ANOVA data (Kruskal-Wallis test, Figure S6, Supporting Information) also indicate that the median expressional difference of most of the data sets (week 1 to week 5) in comparison to the baseline data (day 0) are statistically significant which supports the data of LDA.All these methods showed great coherence and shared a similar trend which further confirms the potential of our integrated platform.This suggests that we can identify the subpopulation of cells showing higher expression of our target biomarkers during treatment that could help physicians to select a proper combination of drug to target the persistent or resistant subpopulation of cells.

Clinical Sample Analysis
To demonstrate the applicability of our single-cell platform in detecting CTCs in patient samples, we initially performed  experiments with simulated patient samples.Specifically, 50 melanoma SK-MEL28 cancer cells from both the treated (week 1) and untreated (day 0) cohort were spiked in healthy subject blood-derived PBMCs and analyzed by our device.The Supporting Information Figure S10 represents the bulk and single-cell expression profile of the target protein biomarkers of the simulated sample.These data clearly resemble the data in Figure 4 for day 0 and week 1 of the treatment for pure samples.This suggests that our method can successfully capture cancer cells from a complex sample and analyze their biomarker expression profile.
Finally, we demonstrated the clinical potential of our platform by analyzing clinical samples (i.e., PBMC) isolated from ten melanoma and two NSCLC patients at different time points of their treatment.Melanoma patients were treated with BRAF inhibitor (BRAFi) monotherapy and/or combined BRAF and MEK inhibitor (MEKi) therapy and NSCLC patients (n = 2) were treated with conventional chemotherapy (see Supporting Information Tables S1 and S2 for the details).The patients were staged into stable disease, partial response, and progressive disease using clinical imaging.Figure 5 presents single CTC profiling data of six samples collected from three late-stage melanoma patients.The left heatmap represents the CTC profiles before treatment and the right heatmap represents the CTC profiles during treatment.Overall, our data show that the CTCs are heterogenous in expressing those four markers in patient samples before and after treatment.For instance, we captured five CTCs for the day 0 (i.e., the day before the patient started therapy) sample of patient 1.As shown in Figure 5 (patient 1), all the CTCs have different levels of biomarker expression, and most of them have high MCSP and MCAM expression.After 7 weeks of BRAF inhibitor therapy, this patient showed partial response according to the radiological imaging.Our single CTC analysis data show that some of the CTCs have low biomarker expression and five CTCs show relatively higher expression of the target biomarkers during treatment.Similarly, patient 9 (Figure S11, Supporting Information) showed a progressive disease response after 4 weeks of treatment.Here, our single CTC analysis data (patient 9) showed a significantly higher biomarker expression profile in three CTCs out of nine captured CTCs after 4 weeks of treatment.This suggests a significant heterogeneity in CTCs during therapy and indicates that the CTCs with higher biomarker expression might be the drug nonrespondent CTC subpopulation.This is particularly important for clinicians to plan possible treatment ahead of time and may help to find a combination therapy targeting those nonrespondent CTCs which could significantly increase patient survival.
The single CTC profiling of patient 4 (Figure 5) showed that most of the captured CTCs had higher expression of MCSP, MCAM, and LNGFR.After several weeks of BRAF and MEK inhibitor treatment, most of the captured CTCs showed relatively lower expression of the targeted biomarkers.This indicates a partial response to the treatment which correlated with the radiological imaging data taken after 3 months of treatment.Like the previous patient, our single CTC analysis identified one CTC (CTC 3) which is still showing similar or higher expression than the day 0 data which indicates that this CTC might be a nonrespondent one.Figure 5 also shows the data for day 0 and week 8 treated samples for patient 10.This patient showed lower biomarker expression levels in the captured CTCs before treatment in comparison to patient 1 and 4.After 8 weeks of BRAF inhibitor treatment, the biomarker expression levels of the CTCs either remained the same or even lower in some cases.The radiology suggested a partial response for this patient.Our CTC profiling data suggested a stable disease with partial response to the treatment.These data not only indicate a stable disease but also suggest that there might have been no nonrespondent CTCs for this patient.For the rest of the patients (patient 2, 3, 4, 5, 6, 7, and 8), our data match the radiological image analysis (Figure S11, Supporting Information).On the contrary, the bulk analysis of these samples in Supporting Information Figure S12a shows that the difference between average biomarker expression levels for most of the patients is insignificant.
Next, we analyzed six timeline samples from two NSCLC cancer patients (Figure 6).These patients received chemotherapy and the radiological staging suggested a disease progression after a certain period of treatment.Our single CTC analysis data showed that most of the CTCs for these patients had high EGFR and low PD-L1 expression at the baseline.However, EGFR expression levels increased notably after 1 week and 1 month of chemotherapy indicating a fast disease progression.Interestingly, in both cases, a number of CTCs showed a higher level of EGFR and PD-L1 expression during therapy.This suggests a combination of EGFR and PD-L1 therapy might help these patients and improve treatment outcome.It is noted that the small samples size might not be representative for treatment selection and requires further study on a larger cohort of NSCLC patients.The bulk analysis data for both patients shown in Supporting In-formation Figure S12b were not conclusive.These findings suggest that our single CTC analysis platform could provide informative CTC profiling for patients in a personalized treatment fashion.

Discussion
Analysis of CTC-derived molecular biomarkers can provide specific information regarding cancer status in real-time and helps to determine appropriate tailored treatments. [57]Due to the invasive and expensive nature of tissue biopsy methods, nowadays liquid biopsies as well as CTC analysis in peripheral blood samples have gained immense interest in the clinical field of cancer detection, characterization, and monitoring. [5,58]61][62][63] Niciński et al. reported a microfluidic chip-conjugated optical tweezers approach to detect specific Raman spectra of target cells. [64]Combining the nitrocellulose membrane and SERS imaging technique, Zhang et al. presented a platform for both CTC enrichment and detection. [65]Using folate-conjugated SERS-active nanoparticles and a magnetic tapping approach, Shi et al. reported the detection of cervical carcinoma (HeLa) cells from complex biological samples. [66]Wang et al. used SERSactive nanoparticles modified with epidermal growth factor (EGF) peptide as a targeting ligand for efficient CTC detection in blood plasma. [67]However, these techniques mostly depend on an average distribution of the cells based on a single marker, not the heterogeneity in a single-cell level with multiple markers, and thereby provide very little clinical information.Moreover, multistep isolation procedures and complex post analysis of the data are also obstacles for these techniques to be used for practical application.Most recently, Reza et al. reported a single CTC analysis technique based on SERS. [68]They used a microfluidic channel integrated with SERS that can perform a single CTC analysis.However, this method requires a pre-isolation of CTCs from PBMC samples using a CD45 depletion method which adds an extra step to the analysis and increases the analysis time.In contrast, our platform has provided an innovative approach to address these limitations.This platform does not require pre-isolation of CTCs from PBMC samples and therefore allows direct capturing of CTCs from PBMC.The workflow including CTC isolation (50 min), labeling of CTCs with SERS nanotags (10 min), and profiling individual CTCs by SERS (60 min) can be completed in ≈2 h, facilitating potential adoption of the test into the clinical setting for evaluation of patients during therapy.
Our assay also showed high specificity as shown in our cell line experiments (Figure 2a,b) and our spiked experiments with a known number of cancer cells spiked in normal/healthy PBMC samples (Figure S10, Supporting Information).Furthermore, our single-cell experiments with cell lines and patient samples showed that our assay can identify the heterogenous biomarker expression in individual CTCs.This suggests that our platform may identify the nonrespondent or resistant CTC subpopulation that can improve patient survival with a better treatment strategy if the corresponding marker proteins are known.However, our platform is based on antibody-based capture of CTCs which has limitations in terms of sensitivity.For instance, an antibody-based capture system may miss certain CTCs that do not express the target antibody and therefore may not represent the whole CTC population of the sample.We believe this limitation could be improved by immobilizing multiple target antibodies in the sensor surface to capture maximum number of CTCs.
In conclusion, our developed single-cell analysis platform determined multiple protein expression levels in individual cell surfaces and thus, demonstrated inter-and intracellular heterogeneity in expression patterns of protein biomarkers during therapy.This platform uses a nanomixing force that helps capture CTCs on a chip from complex PBMC samples.Once captured, the SERS technique can analyze the expression levels of multiple protein biomarkers on the CTC surface with high sensitivity.Using melanoma and lung cancer cell line, our platform specifically captured a single cancer cell and identifies four target biomarkers (MCSP, MCAM, LNGFR, and ErbB3 for melanoma and PD-L1 and EGFR for lung cancer) on a single-cell surface.In addition, our platform identified the drug nonrespondent cells demonstrating high expression levels of the target biomarkers during drug treatment.Most importantly, our single CTC analysis platform showed excellent performance in capturing and profiling single CTCs from melanoma and lung cancer patient samples.Our patient CTC data suggested that we might identify drug nonrespondent or resistant CTC subpopulation during therapy which could suggest alternative or combination therapy for better treatment outcomes.However, the relatively small patient cohorts limit the findings about treatment selection and warrants further clinical validation with larger cohorts.In addition to the high detection sensitivity and multiplex biomarker detection on single CTCs, the platform required only 500 μL of PBMC as input material and showed a potentially clinically amendable sample throughput of 2 h per sample.The throughput could be increased further by running multiple devices in parallel.We believe this simple technology will potentially help to improve current treatment strategies by providing critical information about cancer heterogeneity and helping clinicians to take therapeutic decisions.
top of the fabricated device by transferring the whole chip within a holder (Figure S1b, Supporting Information).
Preparation of SERS Nanotags: SERS nanotags were created by forming a mixed thiol monolayer of dithiobis (succinimidyl propionate) (DSP) and Raman reporters, then conjugating them with antibodies.In a nutshell, gold nanoparticles were made by reducing HAuCl 4 with citrate.To form a mixed thiol monolayer, 2 μL of 1.0 × 10 −3 m DSP in dimethyl sulfoxide (DMSO) and Raman reporters (5 mL of 1 × 10 −3 m MBA, 10 μL of 1 × 10 −3 m DTNB, and 10 μL of 1 × 10 −3 m MMC) in ethanol were incubated with 1 mL of AuNPs for 8 h.AuNPs were mixed with Raman reporters in the presence of the cross-linker DSP to make SERS nanotags (Figure S1c, Supporting Information).The succinimidyl ester of DSP bound to the amine groups of the antibody protein.The chemisorption of an aromatic thiol coupled Raman reporters to the AuNP surface.The colloid was then centrifuged for 10 min at 7600 rpm to extract the supernatants before being resuspended in 200 μL PBS buffer.After that, 2 μg of specific detection antibody (either anti-MCSP, anti-MCAM, anti-LNGFR, or anti-ErbB3 antibody for melanoma cell analysis; and anti-EGFR and anti-PD-L1 antibody for lung cancer cell analysis) was applied to the colloid and incubated at room temperature for 0.5 h.The colloid was then centrifuged for 8 min at 7600 rpm at 4 °C to remove the unconjugated antibodies before being resuspended in 200 μL of 0.1% bovine serum albumin (BSA) for 0.5 h at room temperature.BSA was combined with SERS nanotags as a stabilizing agent in order to minimize SERS nanotag aggregation as well as nonspecific protein binding on the SERS surface.
Device Functionalization: Initially, the fabricated microfluidic device was washed twice with 1× PBS to make it hydrophilic and then dried with nitrogen flow.Subsequently, 100 μL of biotinylated BSA (0.1 μg μL −1 ) was added and incubated for 2 h at room temperature.It was then followed by the addition of 100 μL streptavidin (0.1 μg μL −1 ) for 1 h at room temperature.In the next step, 100 μL of capture antibody (anti-MCSP for melanoma cells, anti-EPCAM for breast cancer cells, and anti-EGFR for lung cancer cells, 0.02 μg μL −1 for each) was added to the electrode, and after 1 h incubation, 50 μL of the blocking solution (1% BSA) was added for 30 min at room temperature.
Device Operation: Specific numbers of cancer cells (e.g., 50 to 250 cells in 100 μL PBS) or PBMC from patient samples were loaded into the serpentine channel of the microfluidic device and ac-EHD was applied for 10 min to fast-track the cell capture.Subsequently, the device was washed with 1× PBS (10 mm, pH 7.4) to remove any unbound cells.The application of ac-EHD induced a charge gradient between small and larger electrode arrays of the microfluidic channels and initiated a fluid flow toward the larger electrode, which led to improved analyte transport.The EHD fluid flow actively brought the CTCs in contact with the capture antibodies and efficiently reduced electrode adsorption of weakly bound, nonspecific molecules. [69]Next, 100 μL of a cocktail of secondary antibodyconjugated Raman reporters (i.e., SERS nanotags) was loaded into the microfluidic channels, and the ac-EHD was applied for 10 min.Washing with 1× PBS was done after each of the stated steps to further remove unbound reagents and cells.Finally, the dried microfluidic channels were analyzed using SERS.
Double Layer Thickness Calculation: For our optimization experiments, 1× PBS was used.The characteristic double layer thickness was given by Debye Length ( D = 1/) which was ≈1 nm for 1× PBS. [70]The double layer thickness of ac-EHD assay = 1.2 nm was previously calculated using the following equation [37,54]  = ( 2000F 2  ∩ rKT where 1  = double layer thickness, F is the Faraday constant, I = ionic strength = ½ Σc i z i 2 (c i = ionic concentration in mol L 1 , z i = valency).SERS Analysis: The captured cells on the electrodes of the device were analyzed individually using a SERS microscope (Witec alpha 300 R microscope (20X microscopy objective)).Initially, individual cells were identified using bright-field microscopy incorporated with the SERS instrument.Once identified, the microscope mode was changed to Raman scattering and the SERS signals were acquired from individual cells without moving the position of the stage.This approach enabled to count the total number of captured cells and individually recorded all the corresponding SERS spectra for single-cell analysis.SERS spectra were obtained at 1 s integration time with a laser power of 70 mW.The SERS mapping was performed at an area of 60 μm × 60 μm (60 pixels × 60 pixels) using a 633 nm laser and a highly sensitive electron-multiplying charge-coupled device (EMCCD).To analyze specific Raman spectrum, Vancouver Raman algorithm and GraphPad Prism 9 were used to identify individual spectral peak positions, intensities, and widths.
Clinical Sample Acquisition: This study was conducted according to the National Health & Medical Research Council Australian Code for the responsible conduct of Research and the National Statement on Ethical Conduct in Human Research.All patients had provided their written informed consent for the research study protocol, which was approved by the Human Research Ethics Committee of the Austin Hospital, Melbourne.Ethics approval was obtained from The University of Queensland Institutional Human Research Ethics Committee (Approval no.2011001315).Methods pertaining to clinical samples were carried out in accordance with approved guidelines.PBMCs were collected from blood drawn into heparinized tubes.PBMCs were separated using conventional Ficoll density gradient centrifugation.The PBMC layer was collected and washed once in 10 mL of RPMI.Subsequently, PBMCs were resuspended in 2 mL of RPMI containing 40% FBS (v/v) and 10% DMSO (v/v) and frozen.
Flow Cytometry Assay: Validation experiments were performed using a CytoFLEX flow cytometry machine (Beckman Coulter, USA) following a previously published protocol. [8]To functionalize the cancer cells for flow cytometry analysis, the desired number of cells was diluted in 100 μL of PBS (10 mm, pH 7.4) in each of the Eppendorf tubes corresponding to the target biomarkers.Then, 2 μg of the specific primary antibody was added to each of the specified tubes and incubated on a shaker at 37 °C for 30 min.The tubes were then centrifuged and the supernatant was discarded followed by the addition of 100 μL of PBS to resuspend the primary antibody-conjugated cells.Subsequently, a secondary antibody was added to the resuspended PBS solution in each tube for 30 min at 37 °C with shaking.Finally, the supernatant was removed after centrifugation and the pellet was resuspended with PBS to carry out the flow cytometry assay.Data were analyzed with CytoFLEX -CytExpert Software 2.4 (2.4.0.28).
Statistical Analysis: The nonparametric one-way ANOVA tests (Kruskal-Wallis test (used to compare three or more groups) and Mann-Whitney test (used to compare two groups)) were done by GraphPad Prism 9 (La Jolla, CA, USA) to compare the expressional difference between untreated cells (baseline) versus the drug-treated cells.The results were considered statistically significant if p values were <0.05.LDA was performed with SPSS 28.0 software package (SPSS Inc., Chicago, IL).

Figure 1 .
Figure 1.Schematic of CTC analysis platform.a) Individual CTCs are captured on the antibody-functionalized electrode surface analyzed by labeled with SERS nanotags to determine the expressional heterogeneity of the proteins on the CTC surface and to long.b) Scanning electron microscope image of the gold nanoparticles used for preparing the SERS nanotags.c) Typical Raman spectra of Raman reporters conjugated to SERS nanotags with characteristic peaks for MBA = 1075 cm −1 , MMC = 1172 cm −1 , DTNB = 1337 cm −1 , and TFMBA = 1375 cm −1 .Bright-field image with overlay of pseudocolor Raman image from d) a single untreated cancer cell with high SERS signal, e) a drug-responded single cancer cell with low SERS signal, and f) a drug-resistant cancer cell with high SERS signal.In panels (d), (e), and (f), the color red, green, blue, and sky blue denotes the attached SERS nanotags on the single cell surface (scale bars are 7 μm) through which different protein expression levels were detected.
larger volumes, thousands to millions of cells and cannot show protein localization on individual cells.

Figure 2 .
Figure 2. Specificity, detection efficiency, and heterogeneity analysis.a) Specificity study: a set of typical SERS spectra of the target SK-MEL28 and SKBR3 cells in different control experiments.The red spectra denote the positive control which shows four different peak intensities achieved based on the expressional status of corresponding biomarkers (MCAM, LNGFR, MCSP, and ErbB3), while spectra with other colors denote the negative controls.b) Corresponding bar graphs showing average Raman intensities of positive (red graph) and negative control (other colors) experiments.Here, the target melanoma cell line (SK-MEL28) with captured antibody (anti-MCSP) was the positive control, while the breast cancer cell line SKBR3 and EPCAM antibody was the negative control.c) Detection efficiency of the integrated platform: the bar graph depicts the capture/detection efficiency of the singlecell analysis platform in counting cells through a Raman microscope and SERS scanning.Each data point represents the average of three separate trials (n = 3), and error bars represent the standard deviation of measurements within each experiment.d) Heterogeneity analysis for MCSP expression in a population of captured cells.Each dot point represents the expression level (Raman intensity) of MCSP protein on individual SK-MEL28 cell surface which was captured on the electrode and thus depicts the heterogenous MCSP expressional pattern on individual cells.

Figure 3 .
Figure 3. Bulk versus single-cell analysis.a) Average intensity of all four protein biomarkers expressed on the melanoma cell surfaces.Each data point represents the average of three separate trials (n = 3), and error bars represent the standard deviation of measurements within each experiment.b) Cellular heterogeneity in protein biomarker expression levels on individual melanoma cell surface detected by the single CTC analysis platform.A number of SK-MEL28 cells (n = 50) were passed through the nanoscopic fluid flow-enabled microfluidic channels and all of the captured cells were scanned using SERS to reveal the protein expression (intensity) level in each individual cell.Each dot represents corresponding protein expression level on a single cell surface and thus reflects the heterogeneity of the captured cell population.The horizontal black bar in this figure denotes the average intensity of the corresponding proteins.Panels (c), (d), and (e) represents the images of captured single cell on the single-cell analysis platform's surface (scale bars, 7 μm).

Figure 4 .
Figure 4. Cell-to-cell and protein-to-protein heterogeneity analysis of a-d) melanoma cells and e,f) lung cancer cells during the drug treatment.The expressional pattern of a) MCSP, b) MCAM, c) ErbB3, d) LNGFR, e) EGFR, and f) PD-L1 on individual cell surfaces at different periods of drug treatment; day 0 (red dots), week 1 (blue dots), week 2 (purple dots), week 3 (green dots), week 4 (gray dots), and week 5 (orange dots).The horizontal black bar in this figure denotes the average intensity of the corresponding proteins.Each dot represents corresponding protein expression level on a single-cell surface and thus reflects both the cell-to-cell and protein-to-protein heterogeneity of the captured cell population.

Figure 5 .
Figure 5. Analysis of melanoma patient samples.The heatmap shows the SERS intensities (500 to 2500 a.u.) representing variation in expression levels of MCSP, MCAM, LNGFR, and ErbB3 proteins in CTCs detected from the samples collected from three different patients that underwent treatment over a period.The left side of this figure denotes the pretreated samples (baseline) and right side denotes the treated samples.The color intensity shifts from black to red represent the proportional increase of the expressional level of the corresponding protein biomarkers (MCSP, MCAM, ErbB3, and LNGFR).Each square box represents designated protein expression levels.While the numeric codes (1, 2, 3, etc.) in the X-axis denote the number/identification (ID) of the captured CTCs on the device.

Figure 6 .
Figure 6.Analysis of lung cancer patient samples.The heatmap shows the SERS intensities representing variation in expression levels of EGFR and PD-L1 proteins in CTCs detected from the analyzed samples collected from two different patients at three different timeframes.The left heatmap denotes the pretreated sample, mid-one (Treatment-1 week) denotes the treated sample collected after 1 week of treatment, and the right heatmap denotes the treated sample collected after 1 month of treatment.Each square box represents designated protein expression levels.While the numeric codes (1, 2, 3, etc.) in the X-axis denotes the number/identification (ID) of the captured CTCs on the device.