A microfluidic chip for screening individual cancer cells via eavesdropping on autophagy-inducing crosstalk in the stroma niche

Autophagy is a cellular homeostatic mechanism where proteins and organelles are digested and recycled to provide an alternative source of building blocks and energy to cells. The role of autophagy in cancer microenvironment is still poorly understood. Here, we present a microfluidic system allowing monitoring of the crosstalk between single cells. We used this system to study how tumor cells induced autophagy in the stromal niche. Firstly, we could confirm that transforming growth factor β1 (TGFβ1) secreted from breast tumor cells is a paracrine mediator of tumor-stroma interaction leading to the activation of autophagy in the stroma component fibroblasts. Through proof of concept experiments using TGFβ1 as a model factor, we could demonstrate real time monitoring of autophagy induction in fibroblasts by single tumor cells. Retrieval of individual tumor cells from the microfluidic system and their subsequent genomic analysis was possible, allowing us to determine the nature of the factor mediating tumor-stroma interactions. Therefore, our microfluidic platform might be used as a promising tool for quantitative investigation of tumor–stroma interactions, especially for and high-throughput screening of paracrine factors that are secreted from heterogeneous tumor cell populations.


Results
Single-cell based microfluidic chip design for monitoring autophagy in tumor-stroma crosstalk. In a tumor microenvironment, cancer cells are involved in dynamic interactions with resident stroma cells (Fig. 1A). Recent evidence suggests that cancer cells induce autophagy in the surrounding stroma fibroblasts, and use digested materials and metabolites produced by them as a nutrient source. However, detailed mechanisms of the crosstalk have not been fully elucidated 18,19 . Therefore, in order to study the communication between a single tumor cell and the stroma niche, we developed a porous membrane-based single-cell trapping device (Fig. 1B). In this device, GFP-LC3 (Green fluorescent protein GFP-fused to MAP1LC3 protein) transgenic immortalized MEF (mouse embryonic fibroblasts) cells modeled cancer stroma fibroblasts. In fact, GFP-LC3 serves as a commonly used tool for autophagy detection. Recruitment of GFP-LC3 to autophagic vesicles resulted in the transformation of the diffuse GFP signal into dot-like puncta, allowing detection of autophagy activation in live cells under a fluorescent microscope ( Fig. 1A and B). MDA-MB-231 triple-negative breast carcinoma cells, which stably express the red fluorescent protein (RFP) (in short, MDA cells) were used as cancer cells (Fig. 1B).
The device platform was composed of a PDMS reservoir, a PDMS membrane, and a PDMS-coated thin glass ( Fig. 1B and C). The fabrication process of the microfluidic chip was explained in detail in Supplementary Fig. S1 and SI Methods. As confirmed by scanning electron microscope (SEM) analyses, the membrane has approximately 800 holes with a diameter of 30 μm, which was adjusted to ideally fit a single cancer cell having a size of ranging between 10-20 μm, and the distance between adjacent holes was 280 μm (Fig. 1D). After bonding the PDMS membrane with the PDMS reservoir, the top side of the membrane was covered with a cell repellent material, i.e., 0.4% (w/v) polyethylene oxide and polypropylene oxide trioblock co-polymer (PEO-PPO-PEO), while the bottom side of the membrane was coated with 10 μg/ml of fibronectin ( Supplementary Fig. S2A). Therefore, fibroblasts were entirely cultured on the bottom part of the membrane, while tumor cells settled in the membrane holes and not on the surface of the top side of the membrane (Fig. 1B). In addition, we confirmed in independent control experiments that the fibronectin coating did not leak to the top side of the membrane through the holes (Supplementary Fig. S2B and C). Furthermore, we checked that coating with PEO-PPO-PEO copolymer prevented non-specific binding of cancer cells on the surface of the PDMS membrane ( Supplementary Fig. S2 D-F).
Fibroblasts were cultured on the fibronectin-coated bottom side of the PDMS membrane. Then the membrane was flipped and the device was assembled. MDA cells were introduced to the top reservoir for single cell trapping in membrane holes. A different number of MDA cells and orbital agitation velocities were tested in order to obtain optimized single cell trapping ratio and exclusion of non-trapped cells from the top side of membranes ( Supplementary Fig. S3). The percentage of holes containing one and only one cell was determined in order to measure the performance of the single cell trapping ratio. Among the many conditions used, optimal single cell trapping ratio was obtained when 10 4 MDA cells were seeded, and under 100 rpm agitation for 5 min followed by 5 times washes (Supplementary Fig. S3B and C). We also confirmed that the viability of fibroblasts was not affected (Supplementary Fig. S4A) and autophagy was not induced in GFP-LC3 fibroblasts by mere culturing on the platform (Supplementary Fig. S4B and C). As shown in Fig. 1E, under optimized conditions, individual cancer cells trapped in multiple arrays of holes in the membrane were in contact with the layer of fibroblasts cultured on the bottom side of the membrane. Upon interaction of single cancer cells trapped in pores, autophagy was induced in the neighboring fibroblasts (Fig. 1F).

Autophagy activation by secreted paracrine factors in the tumor microenvironment. When
MDA cells were co-cultured with MEFs on cover slides in a 1:10 ratio and GFP-LC3 dot formation ratios were quantified, autophagy activation was clearly observed in the MEFs that were located near the MDA cells ( Fig. 2A). Furthermore, the fraction of autophagy positive fibroblasts, defined as cells with more than 20 GFP-LC3 dots, increased with time (Fig. 2B). Detailed criteria for imaging and autophagic cell counting are provided in Methods.
Among the many paracrine factors, TGFβ1 was proposed to play an important role in the formation of CAF from naive fibroblasts 21 . Moreover, TGFβ1 signals contribute toward myofibroblastic cell properties during fibrosis 22,23 and CAF formation 24 . Studies that were published during the preparation of this manuscript also indicated that TGFβ1 could induce autophagy in fibroblasts 25,26 . Therefore, in proof-of-concept experiments, we used TGFβ1 as a model cytokine that could mediate cancer-fibroblast interaction.
First, we confirmed TGFβ1 secretion in our system using the ELISA method. While these tests confirmed that fibroblasts also secreted TGFβ1, a significant contribution of cancer cell-derived TGFβ1 was observed in cancer cell-fibroblast co-cultures ( Supplementary Fig. S5). We observed a 27.5% increase in TGFβ1 levels compared to MEF-alone when MDA breast cancer cells were co-cultured with MEFs.
To prove that TGFβ1 was indeed an inducer of autophagy, we incubated GFP-LC3 MEFs in recombinant TGFβ1-containing medium in the presence of a TGFβ1 neutralizing antibody or a control antibody. As shown in Fig. 2C and D, while autophagy induction occurred in the GFP-LC3 MEFs that were incubated in cultured medium containing HEK 293T-derived recombinant TGFβ1 and control antibody, blockage of TGFβ1 using a specific neutralizing antibody prevented autophagy activation. Moreover, under similar conditions, classical molecular markers of autophagy activation, namely endogenous LC3-I/LC3-II conversion, p62 protein degradation, as well as free GFP production from GFP-LC3 protein tests confirmed that autophagy activation was TGFβ1 dependent (Fig. 2E).
To further prove that endogenous TGFβ1 that was secreted from MDA cells was the rate-limiting cytokine for the observed autophagy activating effects, we collected media from MDA cell cultures that were infected either with lentiviruses containing a shRNA against GFP or TGFβ1 and added them onto the GFP-LC3 MEFs in culture. Off note, TGFβ1 expression levels were significantly reduced in MDA cells that were infected with the shTGFβ1 virus compared to controls ( Supplementary Fig. S6). Conditioned media from control GFP shRNA (shGFP) infected MDA cells could significantly stimulate autophagy in GFP-LC3 MEFs compared to the control cells incubated with MEF-derived media. Strikingly, knockdown of the endogenous TGFβ1 in MDA cells using shRNA clearly prevented autophagy induction in fibroblasts ( Fig. 2F and G). These results were also confirmed using endogenous LC3-I/LC3-II conversion, p62 protein degradation, as well as free GFP production tests (Fig. 2H). These experiments proved that TGFβ1 which was produced and secreted by the MDA cells was one of the key rate-limiting and important factors for the activation of autophagy in fibroblasts.
In order to analyze dose and time kinetics of the cytokine, we performed GFP-LC3 dot formation assays in the presence of purified recombinant TGFβ1 protein (Rec. TGFβ1). 1 ng/ml was determined as the minimum TGFβ1 concentration that was required to induce autophagy in GFP-LC3 MEFs at 2 h and 6 h incubation times ( Fig. 3A and B). Autophagy induction at concentrations under these conditions was confirmed using endogenous LC3-I/ LC3-II conversion, p62 protein degradation, as well as free GFP production tests (Fig. 3C). To determine the minimum concentration of TGFβ1 that was capable of inducing autophagy, GFP-LC3 MEF cells were incubated for 12 h or 24 h with decreasing concentrations of TGFβ1. The lowest autophagy-inducing dose of TGFβ1 was determined as 10 −2 ng/ml at 12 h and 24 h incubation time points (Fig. 3D-F). We assumed that an empty hole is surrounded by the neighbor cancer cells. As the array of holes in the system was periodic, a single periodic unit (represented by the continuous line a-c in Fig. 4A) was taken as the domain of interest (DOI) for simulation. And, the symmetric condition was applied to the boundaries.
The geometrical dimensions, including hole diameter, distance between neighboring holes, and space between the biochip membrane and glass slide, were identical to the real experimental platform. Concentration of TGFβ1 that was secreted from MDA cancer cells was around 250 pg/ml following 24 h of incubation ( 27 and our observations). Using this value and time point, the secretion rate was calculated as 4.88 × 10 −14 mol/m 2 •s when the cancer cell size was assumed as 11 µm (measured by average diameters of trypsinized 100 MDA cells). The diffusion coefficient (D) of TGFβ1 in the reservoir was calculated as 6.4 × 10 −11 m 2 /s, by using the equation of D = 1.72 × 10 −8 (M.W.) −0.552 ( 28 ) and the molecular weight (M.W.) of TGFβ1 as 25 kDa. Fur thermore, the media b etween the membrane and glass was assumed to b e p orous , : porosity and : tortuosity) since it was packed with fibroblasts. We determined basal concentration (C 0 ) of TGFβ1 in the DMEM medium supplemented with various FBS concentrations (1~10% (v/v)) using an ELISA assay ( Supplementary Fig. S5A). When ELISA results were corrected according to cell numbers on chips (66,000 cells in 300 µl medium), TGFβ1 amount secreted from MEFs after 6 h of culture was calculated as 105.93 pg/ml. Basal concentration of TGFβ1 (C 0 ) was considered as the sum of TGFβ1 that comes from 3% FBS containing media (266.72 pg/ml) and MEFs in culture (105.93 pg/ml) ( Supplementary  Fig. S5). The corresponding molar density is 1.50 × 10 −8 mol/m 3 .
TGFβ1 concentration difference profile (∆ = − β C C C TGF 1 0 ) around an empty hole (center) that is surrounded by eight cancer cell-containing hole is shown in Fig. 4B. Figure 4C represents the time evolution of ∆ β C TGF 1 at specified positions (a, b and c points) (Fig. 4A). ∆ β C TGF 1 near holes, that are occupied with a tumor cell (a), rises steeply at the beginning and then increases gradually to reach to approximately 0.66 ng/ml at 12 h, meanwhile the concentration at the empty hole (c) increases slowly and does not reach 0.18 ng/ml even after 12 h. Figure 4D shows the distribution of ∆ β C TGF 1 along the path (a-c distance in µm) in different time points. According to our experimental results, a net 10 −2 ng/ml TGFβ1 concentration was required to induce significant levels of MEF autophagy in our system at 12 h ( Fig. 3D and E); and the same concentration did not induce autophagy after 6 h incubation under similar conditions ( Supplementary Fig. S7). Taking these 6 h and 12 h data into consideration, concentration of TGFβ1 that was required for autophagy induction in MEFs was modeled as a function of time ( Fig. 4E and F). As shown in Fig. 4E, the minimum amount of TGFβ1 exposure for highly probable autophagy activation was fitted as the upper bound (red symbols) and the maximum amount of TGFβ1 exposure that is unlikely to induce autophagy was fitted as the lower bound with a hyperbolic function ). The parameter of A corresponds to the accumulated amount of TGFβ1 in constant secretion condition and b is the fitting parameter. The contour plot of exposure amount for TGFβ1 ( β J TGF 1 ), which is divided into high, intermediate, and low probable regimes for autophagy activation, was drawn by integrating ∆ β C TGF 1 with respect to exposure time ( ) and the distance from single cancer cell (Fig. 4F). Optimized exposure time was decided in the light of these results. 6 h exposure time was long enough to induce high to intermediate levels of autophagy in MEFs around a defined tumor cell (a to b: 0 to 155 µm). Yet, at this time point, autophagy in MEFs around neighboring holes was still at low levels (b to c: 155 to 310 µm). Therefore, 6 h time point was chosen and used in the biochip experiments.

Single cell-level monitoring of autophagy in fibroblasts interacting with a cancer cell.
In the light of tumor-stroma autophagy model, we wanted to confirm that autophagy activation in response to tumor-secreted TGFβ1 in MEFs followed a gradient. In order to reduce heterogeneity in the system, we first tried to improve control vs TGFβ1 signal using chloroquine (CQ, lysosomal pH modifier) or rapamycin (Rapa, an mTOR inhibitor and autophagy inducer). But, there was no improvement (Supplementary Fig. S8). We then created monoclones of both GFP-LC3 MEF cells and MDA cancer cells. These monoclones were similar to parental cells with respect to their TGFβ1 secretion, and autophagic response to TGFβ1 treatment and starvation ( Supplementary Fig. S9). In the biochip system, MEF monoclone cells were cultured alone or with single MDA monoclone cells captured in holes. Images of autophagy activation were taken from 3 different "region of interests" depending on the distance from the center of a hole. Cells that were found at a 50 μm (S), 100 μm (M), 155 μm (L) radial distance from a hole were evaluated. Percentage of GFP-LC3 dot positive cells in these 3 regions were quantified in a cumulative manner (Supplementary Fig. S10). Here, GFP-LC3 dot positivity in MEFs followed a clear gradient on biochips. MEFs that were closer to tumor cell-occupied holes had higher autophagic , the exposure condition space can be divided into high, intermediate, and low probable regimes for autophagy activation. The optimized exposure time, which is sufficiently short so as not to affect autophagy induction via TGFβ1 diffusion from neighbor tumors and sufficiently long to induce autophagy activation at their own position, can be decided by the contour plot.
activity, compared to those that were farther away. However, lowest standard deviation levels and more homogenous autophagic response were obtained when autophagy activation in cells that were within 155 μm (L) radial distance were quantified. As expected, MEFs around empty holes did not activate significant levels of autophagy. Therefore, autophagy activation in biochips followed a gradient and was not random. Activation in empty holes most probably reflected basal autophagic activity in the cells. In the light of these results, we decided to evaluate autophagy activation in MEFs at a 155 μm radial distance from holes.
Next, we performed time kinetics experiments of cells in biochips. We checked autophagy activation in MEFs on biochips at 1 h, 2 h, 3 h and 6 h. We observed that the most significant autophagy activation was obtained after 6 h of co-culture ( Fig. 5A and B). A live cell imaging video of real-time dynamics of autophagy activation in fibroblasts were also recorded (Supplementary Movie S1). Following optimization of the autophagy detection system on biochips, the role of TGFβ1 in cancer cell-stroma interaction was tested. MDA cells were infected with control shRNA (shGFP) or shTGFβ1 constructs and GFP-LC3 MEF autophagy induction levels were quantified. In line with our results in in vitro cell culture dishes, TGFβ1 knockdown in MDA cells (Supplementary Fig. S11) did significantly attenuate autophagy activation in MEFs on biochips (Fig. 5C and D).

Facile retrieval of single tumor cells and downstream molecular analysis.
In order to demonstrate that the proposed platform could identify and isolate autophagy-inducing single cancer cell clones, we performed the following blind tests by using a mixture of wild-type MDA cells and shTGFβ1 infected MDA cells. We performed the single cell analysis of the MDA shTGFβ1 cells after viral infection to decide the ratio of cells in the mixture. As shown in Supplementary Fig. S12, only 8 out of 20 (40%) single MDA cells actually show the shT-GFβ1 band although we could obtain 80% infection efficiency. Therefore, a mixture of wild type MDA and shT-GFβ1 MDA cells with 1:4 ratio was used for blind test. After 6 h of co-culture between cancer cells and fibroblasts on biochips, GFP-LC3 dot status of MEFs were documented. Regardless of their autophagy status, single MDA tumor cells at their center were pinpointed at random, and collected from the chips using a microscope-based single-cell picking device (Fig. 6A). Genomic DNA isolation from the picked cells was performed. Using primers that are specific for the shRNA vector, we were able to PCR amplify and identify the TGFβ1 shRNA sequence in the vector. PCR results confirmed that out of 36 randomly selected single cancer cell clones, 10 cells contained the shTGFβ1 sequence in their DNA. When blindly taken cell pictures were analyzed retrospectively, most MEF cells with less or no autophagy were around holes with MDA cells containing the shTGFβ1 sequence ( Fig. 6B and C).
The cut-off value was for the percentage of GFP-LC3 dot positive MEF cells near a hole was previously decided as 15% based on the capacity of the biochip system to differentiate the CNT MDA cells from shTGFβ1 containing MDA cells with a highest accuracy (82.23% in Fig. 5D). In other words according to our criteria, to label a hole as autophagy positive, the hole should have around it (155 µm radius area) more than 15% of the MEFs containing >20 GFP-LC3 dots per cell. Among 36 MDA cells which were randomly picked-up from 3 independent chips, pictures of MEF cells that were around 21 holes showed higher autophagy signals than the cut-off value, so they were considered as autophagy positive ( Fig. 6B and C). Among cancer cells retrieved from these 21 positive holes, 20 cells were indeed CNT cells and only one cell was shTGFβ1 positive. Remaining 15 holes were autophagy negative. 9 of 15 cells were indeed shTGFβ1 positive MDA cells ( Fig. 6B and C). whereas 6 cells were identified as CNT MDA cells. Figure 6D summarizes the results of these tests.
Therefore, the sensitivity (or, true positive rate) which is defined as the percentage of the holes, whose signal were higher than the cut-off value among total number of holes containing wild type tumor cells, was 77% (20/26). On the other hand, the specificity (or, true negative rate), which is defined as the percentage of the holes whose signal were lower than the cut-off value among total number of holes containing shTGFβ1 MDA cells, was 90% (9/10). As a result, we can differentiate CNT MDA and shTGFβ1 MDA cells based on their single-cell level capability of inducing autophagy in neighboring fibroblasts with 81% accuracy. These results suggest that our platform can be used in shRNA-based screens of cancer cells and allow a high-throughput study of the mechanisms of autophagy activation in fibroblasts in a tumor microenvironment.

Discussion
In this work, a single cell-based screening platform was developed, in which the degree of autophagy activation induced by the intercellular communication between tumor cells and fibroblasts was used as criteria. In these chips, trapping of single tumor cells in the membrane holes of the microfluidic chip followed by co-culture with adjacent fibroblasts was possible. In proof-of-concept tests, we demonstrated in cell culture dishes and on microfluidic chips the effect of TGFβ1 on paracrine autophagy activation in fibroblasts. While this manuscript was in preparation, in line with our results, Cai et al. confirmed autophagy-inducing effects of TGFβ1 on fibroblasts and in the formation of CAF phenotype in a tumor microenvironment 29 .
To the best of our knowledge, this is the first study to demonstrate at a single cell level that autophagy in tumor stroma, can actually be used as a tool to screen for autophagy mediators in individual cancer cells. Therefore, the technique has the potential to reveal key mediators secreted from tumor cells by performing a genome-wide analysis of target cancer cells after autophagy analysis in fibroblasts.
Importantly, the microfluidic platform was suitable for the retrieval of single tumor cells of interest, allowing genomic, transcriptomics and even metabolomics and determination of clonal backgrounds of individual cells. As such, the platform is suitable for large-scale unbiased omics screens.
Future large-scale studies using this microfluidic chip platform might provide novel insights into tumor biology, and allow characterization of key molecular mediators of autophagy during tumor-stroma communication and other biological and pathological phenomena. Indeed, our platform has the potential to be applied to different cellular systems (e.g. tumor cell-macrophage, tumor cell-stem cell couples etc.) and reveal further information on tumor-stroma interactions at a single cell level.

Lentiviral transduction.
Lentiviruses were produced by the co-transfection of pRSI9-U6-(sh)-UbiC-TagRFP-2A-Puro plasmid with helper plasmids psPAX2 and pMD2.G or by transfection of shTGFβ1/ shGFP (kindly gifted by Tamer Önder) with helper plasmids pCMV-VSVG and pCMV-dR8.2 into HEK 293T cells. Culture media were harvested at 48 h and 72 h after transfection, and used to infect cells or stored at −80 °C. To create stable cell lines, MDA cells were infected at 60% confluence for 24 h with lentiviral supernatants diluted 1:1 with fresh culture medium in the presence of 5 µg/ml polybrene (Sigma, H9268) or 5 ng/ml protamine sulfate (Sigma, P4505). 24 h later, fresh culture medium was added. Infected cells were selected for 4 weeks in culture media containing 2 µg/ml puromycin.
Conditioned culture media preparation. Culture media of MDA cells and TGFβ1 or pcDNA3 transfected HEK 293T cells were changed with DMEM medium, containing L-glutamine, 100 U/ml penicillin and streptomycin and 3% FBS. After 72 h incubation, conditioned culture media were collected and concentrated using an Amicon ® Ultra-4 centrifugal filter device (Millipore, UFC800324). In order to avoid autophagy activation by impoverished media, conditioned media were concentrated and diluted with fresh culture media before adding onto MEFs.
After 16 h, MDA cancer cells were added onto MEFs in order to obtain a cancer cell: fibroblast ratio of 1:10. Cells were examined 24 h or 48 h later.

Recombinant protein incubations and antibody neutralizations. Conditioned media from TGFβ1
overexpressing HEK 293T cells or culture media containing recombinant TGFβ1 protein (Sigma Cat #: H8541) were used right away or mixed with 30 µg/ml TGFβ1 neutralizing antibody (R&D, Catalog #: AB293) or control serum (rabbit serum: Sigma, R9133). Following rotation at 4 °C for 2 h, fresh media (DMEM; 10% FBS; L-glutamine; 100 U/ml Pen/Strep.; 1X nonessential aminoacids) was added. GFP-LC3 MEFs were cultured in this mixed media for 24 h and then analyzed. GFP-LC3 dot analyses. Following indicated treatments, GFP-LC3 MEFs were fixed in 4% paraformaldehyde (Sigma, 158127) for 20 min, washed with PBS, mounted and inspected under 60x magnification using a BX60 fluorescence microscope (Olympus, BX60). Dots per cell were counted (at least 100 cells) and basal autophagy threshold was determined as 20 GFP-LC3 dots per GFP-LC3 MEF under non-treated conditions. At least 150 cells per condition were counted and autophagy was expressed as a percentage of dot positive cells (>20 dots) to the total cell population.
Analysis of basal and induced autophagy in microfluidic chip system. Fibroblasts were seeded in the microfabricated PDMS membrane platform, the bottom of which was coated with 10 µg/ml fibronectin. The medium in the reservoir was changed after 16 h of fibroblasts culture. After 6 h of interaction with cancer cells, the autophagy was measured by counting the percentage of fibroblasts with more than 20 GFP-LC3 dots under