Background
A fundamental paradigm in cell biological studies involves evaluation of cell responses to molecular perturbations. These perturbations typically leverage gene disruptions or pharmaceutical intervention to affect protein abundance or activity. Often, the responses to such perturbations are measured on cells subjected to in vitro conditions under the assumption that the observed phenotypes would translate to similar phenotypes in vivo [
1,
2]. However, emerging evidence suggests that the three-dimensional characteristics of the cell microenvironment, including matrix composition, dimensionality, and stiffness affect cancer progression [
3‐
6] and thus it is unknown how the effects of drug perturbation on 2D assays would translate to more complex assays that in principle represent in vivo microenvironment more fully. However, these complex assays are often considered too expensive or difficult to implement for pharmaceutical candidate screening [
7‐
9].
An additional concern during drug screening involves differential responses to perturbation due to cell heterogeneity. Despite observations of drug resistance due to inherent or acquired cell heterogeneity in drug responses [
10‐
12] many high-throughput assays do not measure the single cell responses necessary to determine if a partial response is due to cell heterogeneity. For example, a compound could reduce proliferation in 100% of cells or kill 50% of cells while leaving proliferation unaffected in the remaining cells, but assays measure only cell number would not detect these differences. In addition to the inability to distinguish cytostatic from cytotoxic effects, high throughput assays that employ measurements such as cell luminescence or enzymatic conversion activity as proxy measurements for cell number could be affected by differences in cell state that are neither cytostatic or cytotoxic but simply affect activity of the proxy measure [
9]. Thus, there is a need for an assay that is capable of increasing microenvironmental complexity while retaining the simplicity necessary for high throughput drug screening studies.
Here, we demonstrate an approach for quantifying cell fates in response to drug treatment that may be implemented on common equipment in a reasonably high throughput fashion without sacrificing cell fate information or the ability to measure single cell responses. We demonstrate such a microscopy-based approach using a commonly available epifluorescence microscope and we provide analytical software to facilitate widespread adoption of this approach. Using this methodology, we evaluate the effects of drug treatment on melanoma cells cultured in 3D collagen-based matrices. Such 3D collagen matrices represent a common 3D tissue culture platform [
13], which has been used to measure changes to cell migration [
14,
15], invasion [
16], morphology [
17], and proliferation [
18] as well as to determine the effects of various drug treatments to cancer cells [
19]. Cells were treated with varying doses of the current standard of care for mutant BRaf V600E melanoma patients [
20], a combination of Dabrafenib (Dab), a BRaf V600E inhibitor, and Trametinib (Tram), a pan MEK inhibitor, a drug combination which we abbreviate as RMIC. Imaging was performed using a Nikon Ti epifluorescence microscope with motorized stage and image analysis was performed by a custom python program as well as the open source Cell Profiler program. We use this assay to show that cytostatic and cytotoxic effects occur at different drug concentrations and that the cell’s microenvironment influences drug effectiveness. We find that cells grown in 3D spheroids exhibit very similar drug responses to single cells in a 3D microenvironment, in contrast to cells grown on 2D plastic plates which are less sensitive than either single cells in 3D collagen or cells in 3D spheroids. Finally, we implement this assay on a system comprised of two different cell types in co-culture to illustrate how different responses to drug perturbations may be measured in heterogeneous cell populations.
Methods
Cell lines
A375 melanoma cells harboring the BRaf V600E mutation were purchased from ATCC (CRL-1619). Two PDX cell models PDX1 and PDX2 were acquired from the University of Michigan via Sean Morrison at UT Southwestern Medical Center. PDX1, like A375, is a BRaf V600E mutant melanoma cell line and PDX2 is a BRaf WT melanoma cell line. HFF-1 was purchased from ATCC (SCRC-1041). In order to discriminate cells in co-culture, the HFF-1 cells were lentivirally transduced with GFP-H2B nuclear marker using the PGK-H2BeGFP system as described by Addgene and selected by flow sorting for GFP positive cells.
Tissue culture materials
A375 cells were cultured in Dulbecco’s modified eagle media (DMEM), with high glucose and L-glutamine was purchased from Gibco (11965–167), supplemented with 10% fetal bovine serum (FBS), purchased from Thermo Fisher Scientific (11965–167). HFF-1 cells were cultured in DMEM supplemented with 15% FBS. Both PDX1 and PDX2 primary cells were cultured in Dermal Basal Medium purchased from ATCC (PCS-200-030), supplemented with adult melanocyte growth kit purchased from ATCC (PCS-200-042), which includes rh insulin, ascorbic acid, L-Glutamine, epinephrine, CaCl2, peptide growth factor and M8 supplement. Phenol red free DMEM with high glucose and L-glutamine for fluorescence imaging was purchased from Thermo Fisher Scientific (21063–045). The phenol red free media was supplemented with 10% FBS. Hanks balanced salt solution (HBSS), with no calcium or magnesium was purchased from Gibco (14170112). PGK-H2BeGFP, lentiviral vector to tag the nucleus of HFF1 cells was purchased from Addgene (Plasmid #21210). Trypsin/EDTA was purchased from Thermo Fisher Scientific (R001100). Trypsin Neutralizer (TN) was purchased from Thermo Fisher Scientific (R-002-100). For 3D culture, bovine collagen I with a molecular weight of ~ 300 kDa was purchased from Advanced Biomatrix (5005–100). Collagenase Type I powder to breakdown collagen was purchased from Thermo Fisher Scientific (17100017). Dimethyl Sulfoxide (DMSO) was purchased from Thermo Fisher Scientific (BP231–100). Dabrafenib, a BRaf V600-targeting inhibitor (GSK2118436), and Trametinib, a potent MEK1/2 inhibitor (GSK1120212) were purchased from Selleckchem. EtHd, a cell impermeant viability marker, was purchased from Thermo Fisher Scientific (E1169). Apopxin, a phosphatidylserine marker, was purchased from ABCAM (ab176749). Hoechst 33342, nuclear stain was purchased from molecular probes (H3570). Click-iT based imaging kit (C10340) and flow cytometry kit (C10632) were purchased from Thermo Fisher Scientific. CellTiter-Glo was purchased from Promega for 2D (G7571) and 3D (G968A) assays. Tetrafluoroethylene perfluoropropylene (FEP) tubes for spheroid light sheet imaging were purchased from BOLA, Germany (1815–04).
General cell culture
All 2D cell cultures were performed in 24 well plates. Cells at ~ 70% confluency were washed in 1X PBS, trypsinized, and harvested at 500 g for 5 min. The cell pellet was then re-suspended in 1–5 mL media depending on the cell density. The concentration of cells was determined using a cell counter and the dilution was adjusted to 500 μL per well for 24 well plates. The cells were seeded at a final density of 15,000 cells/well and incubated overnight to allow attachment to the bottom of the dish following which drug evaluations were performed.
3D cultured single cell, co-culture and spheroids
All 3D cell culture experiments were performed in 24 or 96 well plates. Sterile 10X PBS, 1 M NaOH and water were pre-warmed in a 37 °C water-bath. The multi well plate was pre-warmed in 37 °C incubator and an aliquot of 3.2 mg/mL collagen was brought to room temperature. Note: Cold reagents can lengthen the collagen polymerization time, resulting in cells settling to the bottom of the dish, which does not recapitulate the 3D morphology. To create the cell pellet, cells were trypsinized, harvested and re-suspended in 1–5 mL media depending on the cell density. The concentration of cells was determined using a cell counter and the desired volume of cells were centrifuged to acquire a pellet with the desired cell number. In the case of cells cultured as 3D single cells in 96 well dishes, a seeding density of 30,000 cells/well were used and for 24 well dishes a seeding density of 100,000 cells/well was used. To make spheroids a density of 10,000 cells/well was used. To prepare collagen at a final concentration of 2 mg/mL from 3.2 mg/mL stock, we combined 100 μL 10X PBS, 10 μL 1 M NaOH, 250 μL water and 640 μL of 3.2 mg/mL collagen stock. The pH was measured to be between 7 and 7.4 using pH strips. The cell pellet was re-suspended in 2 mg/mL collagen and added to the pre-warmed multi-well plate. The cells were incubated for 30 min in 37 °C incubator. Collagen polymerization was observed through change in color of the solution from a transparent solution to a cloudy gel. Media was then added to cells and incubated overnight.
For 3D co-culture experiments HFF1-H2B-eGFP cells were cultured 1:1 with A375 melanoma cells. Briefly, HFF1 mono-culture cells were seeded in 96 well dishes at 20,000 cells/well in a 96 well. We used 20,000 cells because cell numbers higher than 20,000 caused collagen contraction [
21]. For the co-culture A375 and HFF1 cells were cultured at 20,000 cells/well with each cell type at 10,000 cells/well. For the co-culture we used both DMSO and HFF1-H2B-eGFP as controls. The cells were setup in 3D like 3D mono-cultures.
A375 melanoma spheroids were created in 3D collagen by suspending single cells in collagen and culturing them over 10 days with frequent replacement of media to form 3D cultured spheroids. For RMIC evaluation in 3D, drug dilution was performed as described below.
Drug treatment as a function of concentration in 2D and 3D
A375, PDX1, PDX2, HFF1-H2B-eGFP mono-cultures and A375 + HFF1-H2B-eGFP co-cultures were setup in multi-well plates in 3D collagen and 2D culture platforms as described in the methods above. For drug evaluation, a 10 mM stock drug concentration was prepared in DMSO. The cells were incubated with three concentrations Dabrafenib and Trametinib: 10 nM Dab+ 1 nM Tram, 100 nM Dab+ 10 nM Tram and 1 μM Dab+ 100 nM Tram with 0.02% (v/v) DMSO as control. A 2X desired concentration of the drug was first prepared in media. This was followed by a 1:1 dilution in media, which yielded the desired final concentration. Controls received the same amount of vehicle (DMSO) as the drug treated cells (0.02% v/v). The cells with drug were then incubated for a total of 72 h. After 48 h of incubation EdU at a final concentration of 10 μM was added. This was done by making 20 μM EdU solution which was mixed with 2X drug and subsequently diluted 1:1 with media to give a final concentration of 10 μM EdU. This was subsequently incubated in cells for further 24 h to make a total drug incubation of 72 h.
Microscopy-based cell fate measurement
Following drug treatment, the media was aspirated from the cells and incubated for 30 min with 4 μM EtHd, 2 μL (per 100 μL media) Apopxin and 15 μg/mL Hoechst 33342 in fresh phenol red free DMEM supplemented with 10% FBS. Imaging was then performed with a Nikon Ti epifluorescence microscope with an OKO temperature and CO2 control system regulated at 37 °C with 5% CO2. The cells in 3D were imaged with a z-step size of 2 μm and a total of 251 steps. The filter set for the red channel had an excitation from 540 to 580 nm and emission at 600-660 nm, green channel had an excitation from 465 to 495 nm and emission from 515 to 555 nm and blue channel had an excitation at 340-380 nm and emission at 435-485 nm.
To identify cells undergoing S phase of cell cycle, cells were incubated with modified thymidine analogue EdU for 24 h and labeled using click chemistry. Following viability imaging after drug treatment, the cells in 2D and 3D were washed with 1X PBS. The cells cultured in 3D were fixed with 4% paraformaldehyde for 30 min at 37 °C, and the cells in 2D were fixed for 20 min at room temperature. The fixed cells were permeabilized with 0.5% Triton X-100 for 30 min in 3D and 20 min in 2D. Note: that the 0.5% Triton X-100 was prepared from a 25% stock solution. The 25% stock should be stored at 4 °C and diluted to 0.5% before use. The Click-iT reaction was then performed using the Click-iT EdU assay imaging kit with Alexa Fluor 647 fluorophore. The Click-iT reaction was partially modified from the manufacturer’s protocol such that the Alexa Fluor 647 azide was first diluted 1:100 in DMSO before adding the volume specified by the manufacturer’s protocol, resulting in a concentration 1/100th of the recommended concentration. This dilution needs to be optimized separately for each cell type. If the Alexa Fluor 647 is used directly as per manufacturer’s protocol in 3D the dye will non-specifically bind to the collagen, creating a high background.
RMIC treatment as a function of concentration on co-cultures
To evaluate identification of heterogeneous cell population using the image based assay we evaluated a co-culture based system. HFF1 cells tagged with H2B-eGFP were co-cultured with un-tagged wild type A375 melanoma cells. The cells in co-culture were seeded at 10,000 cells of each cell type per well in 3D. A control of HFF1-H2B-eGFP cells alone were also setup in 3D. The cells were treated with RMIC as a function of concentration as described above and incubated for a period of 3 days. Following RMIC incubation a modified viability assay was performed. Since the HFF1 cells were tagged with H2B-eGFP, we did not use Apopxin. Viability assays were performed using Hoechst and EtHd.
A control experiment was performed to identify H2B-eGFP pixels intersecting with Hoechst pixels, for HFF1 cells. This would inform how efficiently we can identify HFF1-H2BeGFP cells. Upon imaging and quantification a ratio of eGFP pixels to Hoechst pixels was then quantified for HFF1 cells upon RMIC treatment. Viability assay was then performed to determine the positive pixels for Hoechst, eGFP for HFF1 cells and EtHd for total dead cells. The intersection between the positive pixels for each channel were also acquired to identify cells that were both GFP and EtHd positive. This indicates death of HFF1 cells alone. The EtHd pixels that do not overlap with eGFP channel will indicate A375 melanoma cell death. This will give us information on the effect of RMIC on a co-culture based platform.
Image analysis using cell profiler and data processing
To analyze all the 2D and 3D images it was important to identify software that can be used with minimal computational experience. We identified Cell Profiler as an ideal candidate [
22] (
http://cellprofiler.org). We used Cell Profiler V 2.2. To operate Cell Profiler we established image analysis pipelines that can accept our multi-point z-stack images and analyzed data for positive pixels from each channel.
The z-stack images were acquired as individual TIFF files. A distinguishing parameter such as a unique name for each channel was used while acquiring 3D and 2D images. In the input module under “NamesAndTypes” we defined the distinguishing name for each channel to help the software separate images from different channels. We used a robust background thresholding method to distinguish foreground (i.e. marked cells) and background regions. The pixel areas of foreground regions were measured for each channel.
While using two dead cell markers (apopxin and EtHd) to identify different phases of cell death, we observed some cells that were marked by both dead cell markers. To identify cells that were counted twice we created a mask to identify overlapping regions and then subtract that area so they are counted only once (Additional file
1: Figure S1). For analyzing the cell proliferation data the pipeline was modified to two channels and the mask function was removed. Both the viability and proliferation data were exported to a spreadsheet for analysis.
To analyze the viability data we first subtracted the overlapping pixel area of the dead cell channels (apopxin and EtHd) from the total area of the dead cells channels. A log (ratio + 1) of the total pixels from hoechst to the dead cell pixels was calculated and averaged over all images acquired for a particular treatment. The average values for each treatment were normalized to the DMSO control; and these values were referred to as viability scores. Analogously, proliferation scores were computed based on averaging and normalization of the log (ratio + 1) of the total pixels from Hoechst to EdU pixels.
Parallel processing image analysis using a custom python program to increase throughput of image analysis
Three challenges with Cell Profiler were (a) the throughput of image analysis upon introducing signal correction functions like illumination correction (b) the absence of automated thresholding and (c) extended processing time. Illumination correction is essential as raw images might have a strong bias towards high intensity values towards the center of the image. To correct for these illumination effects, we first use a local Gaussian with a large variance to smooth the image. We then subtract the background from the foreground and set any resultant negative values to zero because negative intensity is impossible.
To fully automate the image processing, we need to automatically distinguish between background and foreground, hence we need automatic thresholding. To achieve this, we need to tackle two challenges: halo effects from out-of-focus cells, and the absence of biological signal. To deal with halo effects, we adopt a two-pronged strategy. First, we use the thresholding algorithm developed by Yen and colleagues [
23]. We chose this specific algorithm because it seeks to minimize the complexity of the thresholded image, in addition to the standard concerns about seeking a distinct signal level. The assumption of simplicity in the thresholded image makes sense in our domain because the stained biological specimens are all likely to be highly similar and uniform. As such, we observed that the resultant threshold values capture only the in-focus nuclei and not the irregular shapes caused by out-of-focus halo effects. Furthermore, to avoid the irregularities due to halo effects in the extremes of the z-stacks, we only use the images at the center of the z-stack to identify threshold values. Finally, to tackle images devoid of any meaningful biological signal, we compare the threshold value acquired in each image to the mean of the background in the same image. If the threshold is less than ten-fold intense when compared to the background, then we posit that the signal-to-noise ratio is weak due to the absence of any meaningful biological signal.
Finally, the image processing code needs to scale in order to facilitate drug screening. With this goal in mind, we built a highly modular and parallelized solution. In one standard compute node of our cluster, we can process 1800 images under 3 min. By submitting multiple parallel jobs to different nodes, we can process different wells or plates in parallel. When we leverage a high performance computing cluster to run 16 such jobs concurrently, we achieve a throughput of about 10,000 images per minute.
We compared the output of the python script with the output of the CellProfiler pipeline to determine if any systematic differences result from the different analysis approaches. Viability images of A375, PDX1 and PDX2 were analyzed by both approaches and results are compared in (Additional file
2: Figure S2). The data indicated similar trends to Cell Profiler at lower concentrations for all the three cell lines. However, at the highest concentration of 1 μM Dab + 100 nM Tram some difference could be observed for A375 and PDX2 models. A possible reason for this could be the illumination correction function that has been used in the Python script to remove background illumination light.
Evaluating effect of well size by comparing viability and proliferation in 24 and 96 well plates
To understand if well diameter results in variable drug efficacy, we evaluated the effect of RMIC on viability and proliferation as a function of concentration in A375 cells in 24 well and 96 well plates. A375 cells were plated in 2 mg/mL collagen in 3D. The cells were then treated with RMIC as a function of concentration for 3 days. On day 2 of RMIC treatment, 10 μM EdU was added for an additional 24 h. After 3 days total incubation, viability and proliferation assays were performed as described above.
Comparing imaging and flow cytometry for RMIC as a function of concentration
To validate the robustness of the 3D imaging platform, we measured RMIC effect on proliferation using the flow cytometry as an independent method. A375 cells were incubated with drugs as a function of concentration. The cells were incubated with drugs for 3 days and on day 2 of treatment, EdU at a final concentration of 10 μM was added to the cells. Following incubation with RMIC and EDU, the collagen was washed with Hanks balanced salt solution (HBSS). The collagen was then treated with collagenase as per manufacturer’s protocol for 30 min at 37 °C to breakdown the collagen completely. Media with 10% FBS was then added to neutralize the effect of collagenase. A flow cytometry based Click-iT reaction was performed as per manufacturer’s protocol. In brief, the cells were centrifuged to collect the pellet. The pellet was resuspended in 4% PFA for fixation and permeabilized with 1X Tween-20. The Click-iT reaction was performed, and the flow cytometry was gated using a negative control where the cells were not treated with EdU. To analyze the results, a uniform gating was setup by using the FlowJo software with a negative control consisting of cells not subjected to the Click-iT reaction and thus exhibiting only autofluorescence. The treated samples were the gated under the same settings.
Comparing imaging and CellTiter Glo for RMIC as a function of concentration
A375 cells were cultured in black walled, glass bottom 96 well dishes. Cells for 2D were seeded at 5000 cells/well and for 3D at 30,000 cells/well. The RMIC was incubated as a function of concentration as described above for a period of 3 days. After 3 day incubation the cell were treated with the standard CellTiter Glo for 2D and CellTiter Glo-3D reagent for the cells cultured in 3D. The cells in 2D were after addition of reagents were rocked for 2 min and further incubated for 10 min following which the plate was read using a BioTek plate reader. The cells cultured in 3D was incubated with 3D-CellTiter Glo were rocked for 10 min and incubated for a further before measurement.
Imaging cells in suspension using the click-iT flow cytometry kit
Since the Click-iT EdU based imaging kit was designed for adherent cells, we aimed to design an imaging based method to determine proliferation for cells in suspension. We thus tested the application to imaging of a flow cytometry kit that is designed for suspended cells. To do so we cultured A375 melanoma cells in ultra low adhesion 6 well plates to keep cells in suspension. The cells were then treated with RMIC and EdU as described above. The Click-iT reaction was performed with the flow cytometry kit as per manufacturer’s protocol and cells were also stained with Hoechst. A drop of the cell suspension after the Click-iT reaction and Hoechst treatment was put on a glass slide and placed under a glass coverslip to, after which images were acquired on the Nikon Ti microscope. The cell images for nuclei Hoechst staining and EDU proliferation were analyzed using Cell Profiler.
Melanoma spheroid preparation for light sheet fluorescence microscopy (LSFM)
Tumor cell spheroids were created from single melanoma cells by culturing cells at low concentration and allowing them to grow into large clusters. Sample holders for light sheet imaging were created as follows. FEP tubes were cut into small pieces with holes made along the walls using a 27G needle in order to allow media exchange across the tube. The tubes were cleaned with 30% bleach for 30 min and placed in 100% EtOH overnight. The sterile tubes were dried and pre-warmed in a 37 °C incubator. A375 melanoma cells were prepared in 3D collagen as described above and seeded at low density within the FEP tubes. The collagen was allowed to polymerize at 37 °C, fresh media was added, and then the cells were incubated for 10 days with regular replacement of media. The spheroids then underwent RMIC treatment for 3 days with 24 h EdU incubation. Following RMIC treatment and EdU incubation, the Click-iT imaging assay was performed, and the spheroids were imaged on the LSFM.
To image spheroids, a low magnification LSFM was constructed with a large field of view. To increase the field of view, and improve optical penetration, a conventional dual illumination LSFM was built that permits scanning of the beam in the Z-dimension as well as pivoting the light-sheet in the sample plane to reduce shadowing and stripe artifacts [
24]. For illumination, four lasers (405 nm, 488 nm, 561 nm, and 640 nm, Coherent, OBIS) are combined with dichroic mirrors (MUX Series, Semrock), spatially filtered and expanded with a telescope consisting of an achromatic convex lens (f = 50 mm, Thorlabs, AC254–050-A-ML), a pinhole (100 μm, Thorlabs, P100H), and an achromatic convex lens (f = 400 mm, Thorlabs, AC254–400-A). An achromatic Galilean beam expander (Thorlabs, GBE02-A) further increases the laser diameter by 2x. All solid-state lasers are directly modulated with analog signals originating from a field-programmable gate array (PCIe-7252R, National Instruments) that have been conditioned with a scaling amplifier (SIM983 and SIM900, Stanford Research Systems).
For light-sheet generation, a cylindrical lens (f = 50 mm, Thorlabs, ACY254–050-A) is used to focus the laser illumination into a sheet, which is relayed to the illumination objective (Nikon 10X Plan Fluorite, NA 0.3) with two mirror galvanometers (Thorlabs, GVS001), two coupling lenses (f = 50 mm, Thorlabs, AC254–050-A), and a tube lens (f = 200 mm, Thorlabs, ITL-200). The Z-galvanometer is conjugate to the back pupil of the illumination objective, whereas the pivot galvanometer is conjugate to the sample. To generate the second illumination arm of the microscope, a polarizing beam splitter (Newport, 10FC16PB.3) is placed after the second coupling lens, and two lenses (f = 200 mm, Thorlabs, AC508–200-A) relay the scanned illumination to a second tube lens and illumination objective. Because the counter-propagating light-sheets are orthogonally polarized, there is no interference from the two light-sheets, and a half-wave plate (AHWP10M-600) placed in front of the polarizing beam splitters is used to adjust the intensity of the two light-sheets. To control the light-sheet thickness, a variable slit was placed in the back-pupil plane of the cylindrical lens, which is conjugated to the back-pupil planes of both illumination objectives.
For detection, a 16x, NA 0.8 objective lens (Nikon CFI75 LWD 16X W) and a tube lens (f = 200 mm, Edmund optics, 58–520) form the image on a sCMOS camera (Hamamatsu, C11440, ORCA-Flash4.0). A laser line filter (Chroma, ZET405/488/561/640) is placed after the detection objective lens and a filter wheel (Sutter Instrument, Lambda 10-B) equipped with multiple bandpass filters is placed between the tube lens and the camera. The detection objective lens is mounted on a piezo-driven stage (Mad-City Labs, Nano-F450) that provides 450 μm travel range. The sample stage is a combination of a three-axis motorized stage (Sutter Instrument, MP285) and a rotation stage (Physik Instrumente, U-651.03). The microscope is controlled by custom-written LabVIEW software (Coleman Technologies, National Instruments).
3D co-culture experimental procedures and image analysis
To measure the effects of RMIC treatment on fibroblasts alone, we first test the GFP signal detection in fibroblast monoculture. The cells were treated with RMIC as a function of concentration for 3 days. Following treatment, the cells were imaged for viability with Hoechst and EtHd. Apopxin wasn’t used because the HFF-1 cells were GFP tagged. Since it was a mono-culture of only HFF-1 cells tagged with GFP and stained with Hoechst we first tested if all the Hoechst positive cells were also GFP positive.
To remove the effect of fibroblasts from our measures of cell fate, the H2B-eGFP pixels were subtracted from the Hoechst pixels. Since the A375 were only stained with Hoechst and not eGFP the subtraction would give us A375 pixels. Likewise, EtHd positive A375 melanoma cells were identified by computationally removing fibroblast cells from the EtHd count. To identify EtHd positive fibroblasts a mask was made to determine the intersection between GFP and EtHd channels and that intersection was then subtracted from the total EtHd positive pixels. Ratios of the EtHd positive A375 cells to Hoechst positive A375 cells were then acquired and normalized to DMSO to give viability results in the co-culture platform.
Statistical tests
All statistical measurements were performed using Holm-Sidak method, with alpha = 0.05. Each row was analyzed individually without assuming a consistent SD. Results are displayed using p-value of < 0.05: *, < 0.005: ** and < 0.0005: ***.
Discussion
To streamline drug discovery, we must develop methods to evaluate compound efficacy under increasingly complex extracellular microenvironments [
36,
37]. However, experimental complexity must be balanced with throughput, cost, reproducibility, and precision. Here, we describe an assay that enables high-throughput compound evaluation without sacrificing single cell resolution, readout specificity, or microenvironmental complexity. Indeed, although we evaluated this assay using 3D collagen scaffolds, it is equally amenable to other scaffolds and would be equally useful for identifying which ECM components affect drug efficacy. We show that wide field 3D fluorescence microscopy is sufficient to quantify cell fate in complex samples including spheroids; however, enzymatic extraction and 2D imaging enables the same level of precision via an even more widely available imaging platform if so desired.
The example results produced via this method also reveal critical insight into how cells respond to drug perturbations in different environments. Indeed, the observation that even a simple 3D collagen microenvironment affects how cells respond to treatment suggests that more careful consideration of the microenvironment may lead to more efficient drug evaluation. For example, the observation that cells in 3D collagen are more sensitive to drug treatment than cells in 2D suggests that many potentially useful compounds might have been ignored because they were tested on cells under 2D conditions. We also show that the standard RAF/MEK treatment is primarily cytostatic and that cytotoxic effects only occur at higher drug concentrations. This result suggests that RAF/MEK treatment may be more effective at slowing tumor growth than eradicating tumor cells and that if slowing tumor growth is the goal then low drug doses may reduce proliferation while reducing side effects compared to higher doses. Similarly, the RAF/MEK treatment affects BRAF wild type melanoma cells and fibroblasts similarly by killing cells only at high concentration, suggesting that the non-targeted effect of the MEK inhibitor Trametinib kills cells indiscriminately and mainly at higher concentrations.
The ability to observe single cell and heritable responses also provides insight into the drug resistance often observed in patients subjected to RAF/MEK inhibitor treatment. For example, if a fraction of melanoma cells are inherently resistant to drug inhibition and the drug effect is primarily cytostatic then it follows that expansion of a drug resistant population will occur. This insight would not be so easily gleaned from conventional methods of evaluating 3D cell culture assays such as chemiluminescence or enzymatic conversion that do not provide cell fate information or single cell resolution [
38‐
40]. Indeed, the effect of RAF/MEK treatment observed in melanoma cells is a combination of proliferation reduction and cell killing that is obscured when measured using Cell Titer-Glo. We also demonstrate the ability to discriminate between different cell populations by testing the hypothesis that fibroblasts in co-culture with melanoma cells can provide a protective effect from RAF/MEK inhibition. The ability to computationally remove fibroblasts from our cell fate measurements, while retaining their effect on melanoma cells, shows that fibroblasts offer little protective effect from RAF/MEK inhibitors for cells co-cultured in 3D collagen. We do observe a small decrease in cell killing at high doses of RMIC, which suggests that the viability effects of RAF/MEK inhibition observed only at high doses may be effected by fibroblasts. We also note that a bulk cell measurement of cells in co-culture would obscure these results - a bulk cell measurement of high RMIC on co-culture would overestimate the effect of RMIC treatment at high doses because it would include the effect of high RMIC doses on fibroblasts.
We also demonstrate how our approach may be used to test the hypothesis that melanoma spheroids offer a more realistic paradigm for drug evaluation than 2D cultures. This is indeed the case – A375 cells grown in spheroids are more sensitive to proliferation inhibition than cells grown on plastic dishes. However, we observe that cells grown as single cells in 3D are equally sensitive to RMIC when compared to cells grown in spheroids. While this phenomenon remains to be evaluated in other systems, the possibility that single cell 3D culture is sufficient to replicate 3D spheroid culture has to potential to greatly reduce experimental costs without information gained from an experiment. Hence, at least for the pathways tested here, drug effects can be tested in the much faster 3D single cell assay and do not require the growth of spheroids. It remains to be determined which condition more faithfully reflects the response of cells to RAF/MEK inhibition in vivo.