Mechanically tunable coaxial electrospun models of YAP/TAZ mechanoresponse and IGF-1R activation in osteosarcoma

Abstract

Current in vitro methods for assessing cancer biology and therapeutic response rely heavily on monolayer cell culture on hard, plastic surfaces that do not recapitulate essential elements of the tumor microenvironment. While a host of tumor models exist, most are not engineered to control the physical properties of the microenvironment and thus may not reflect the effects of mechanotransduction on tumor biology. Utilizing coaxial electrospinning, we developed three-dimensional (3D) tumor models with tunable mechanical properties in order to elucidate the effects of substrate stiffness and tissue architecture in osteosarcoma. Mechanical properties of coaxial electrospun meshes were characterized with a series of macroscale testing with uniaxial tensile testing and microscale testing utilizing atomic force microscopy on single fibers. Calculated moduli in our models ranged over three orders of magnitude in both macroscale and microscale testing. Osteosarcoma cells responded to decreasing substrate stiffness in 3D environments by increasing nuclear localization of Hippo pathway effectors, YAP and TAZ, while downregulating total YAP. Additionally, a downregulation of the IGF-1R/mTOR axis, the target of recent clinical trials in sarcoma, was observed in 3D models and heralded increased resistance to combination chemotherapy and IGF-1R/mTOR targeted agents compared to monolayer controls. In this study, we highlight the necessity of incorporating mechanical cues in cancer biology investigation and the complexity in mechanotransduction as a confluence of stiffness and culture architecture. Our models provide a versatile, mechanically variable substrate on which to study the effects of physical cues on the pathogenesis of tumors.

Graphical Abstract

1. Introduction

The high failure rate of drugs in early phase clinical trials can partially be blamed on the lack of translatability of current in vitro models [1, 2]. A fundamental tool for in vitro study of cancer is tissue culture on hard glass or modified polystyrene substrates. However, in vitro culture techniques suffer from myriad problems in recapitulating essential hallmarks of cancer including physical elements such as extracellular matrix stiffness and architecture [3]. Compounding the problem, these non-physiologic substrates have drastic effects on cell phenotype, expression of “biomarkers,” and therapeutic resistance profiles [4].

In recent years, the tumor microenvironment has been recognized as playing critical roles in cancer biology and therapy evasion [4–6]. The tumor microenvironment comprises a diverse array of elements including architectural cues, heterotypic cell interactions, biochemical signals, and mechanical properties. This convergence of factors facilitates the formation of a complex network of adaptive cell phenotypes that evade our best therapeutic options. Recent attempts at the identification of aggressive cell populations in cancer have involved suspension culture and encapsulation of cells in hydrogels of variable stiffness in order to generate populations of more aggressive phenotypes characterized by increases in cell-cell contact, drug resistance, plasticity, and tumorigenic potential [7, 8]. While these models have underscored the role of microenvironmental elements in tumor pathogenesis, many of these systems are not amenable to the interrogation of biophysical signals without confounding other elements such as the degree of cell spreading, cell shape, and nutrient or oxygen mass transport. Other approaches for interrogating the role of tumor microenvironment mechanical properties have included culturing cells on collagen-coated synthetic hydrogels of varying stiffness, although these generally lack the three-dimensional architecture of the microenvironment [9, 10]. More sophisticated strategies to fabricate models with greater complexity by tuning both mechanical properties and architecture or by limiting confounding factors such as changes in matrix composition and structure tend to be costly and low throughput [11, 12].

The mechanical environment is known to directly regulate lineage commitment of normal mesenchymal stem cells (MSCs) and generation of cancer stem cells [13]. However, it is unknown whether osteosarcoma cells, believed to derive from genetically aberrant MSCs or slightly more differentiated osteoblastic daughter cells, are equally susceptible to the biophysical forces inherent in the tissue, which vary widely in bone [14–16].

In this study, we set out to build electrospun models to investigate how osteosarcoma cells respond to their mechanical microenvironment. Utilizing coaxial electrospinning techniques, we fabricated a highly porous, mechanically tunable biomaterials suitable for studying the response of osteosarcoma to mechanical and architectural cues. We characterize our coaxial electrospun scaffolds as models for osteosarcoma with macroscale mechanical testing in a uniaxial tensile testing system and microscale testing with atomic force microscopy measurements of elastic moduli. We then assess the mechanoresponse of osteosarcoma cells by determining changes in localization and expression of Hippo pathway effectors, Yes-associated protein 1 (YAP) and transcriptional coactivator with PDZ-binding motif (TAZ) [17].

The insulin-like growth factor 1 receptor (IGF-1R)/mechanistic target of rapamycin (mTOR) cascade has been the target of recent clinical trials for osteosarcoma and other cancers [18–20]. Our group and others have previously shown that mechanical stress and culture architecture can affect drug sensitivity in vitro to both chemotherapy and IGF-1R/mTOR targeted therapy [21–23]. After establishing that osteosarcoma cells respond to mechanical elements in the microenvironment, we investigate the influence of our engineered substrates on the activation of the IGF-1R/mTOR pathway. We further characterize the differential response to combination chemotherapy and IGF-1R/mTOR targeted therapy in our 3D models. Our models, which are low cost and high throughput, make evident the phenotypic drift in osteosarcoma due to mechanical cues, and we evaluate the potential ramifications these changes have for stem cell characteristics, proliferation, and response to current therapy.

2. Materials and Methods

2.1. Statistical analyses

All statistics were performed using Microsoft Excel or GraphPad Prism. All data are displayed as raw values when possible or for proliferation and drug testing assays, normalized to an internal control after subtracting for baseline fluorescence or absorbance, respectively. Data are displayed as means with standard deviations, standard error, or 95% confidence intervals, as indicated in figure captions. Electrospun mesh evaluation, all in vitro cellular values, and patient sample staining values were compared between groups using paired t-tests, one-way or two-way ANOVA with post-hoc Tukey’s HSD as described in figure captions with *p < 0.05, **p <0 .01, ***p <0 .001, #p ≤ 0.001.

2.2. Electrospinning set up and technique

All meshes were electrospun using the YFlow Professional Lab Device V2.0. Fiber meshes with random orientation were spun onto an aluminum collecting plate. 60 mL syringes were loaded into the syringe pump pointed downward towards the collecting plate. Poly(ε-caprolactone) (PCL, average M_n 80,000, Sigma-Aldrich) and gelatin (Type B, Nitta Gelatin) were dissolved at 15% w/v in trifluoroethanol (TFE, Sigma-Aldrich) overnight in sealed round bottom flasks. For fibers consisting of purely gelatin or PCL, a 16G blunt needle tip was used, and polymer was extruded at a rate of 5 mL/hr at a distance of 10 cm from the collecting plate. The collecting plate was covered with aluminum foil and then non-stick parchment paper to prevent fibers from adhering to the collecting surface. For pure gelatin and PCL, meshes were spun for 45 min to achieve a thickness of at least 100 μm. The voltage applied ranged from 6–8 kV and the collecting plate was connected to ground. For coaxial fibers, a coaxial needle consisting of one blunt tip needle of 20G inside a 16G blunt tip needle was used (ramé-hart instrument co.). Polymer was extruded from each compartment at variable rates that totaled 10 mL/hr. A voltage of 9–11 kV was used to achieve a stable Taylor cone and the coaxial needle was placed 15 cm from the collecting plate. Coaxial fiber meshes were spun for 25–30 min each. Architecture of coaxial and homogeneous fibers was confirmed by briefly spinning fibers onto glass slides and viewing fibers with phase contrast microscopy under dry conditions and wet conditions after being briefly wet with PBS. Images were taken of individual fibers and processed by in grayscale with rolling ball background subtraction (r = 50 pixels) using ImageJ (Fig. S1).

For aligned fiber meshes, the flat collecting plate was replaced with a rotating mandrel of 25 cm diameter and covered with aluminum foil and parchment paper as before. The mandrel was rotated at 1000 RPM to ensure fibers were aligned. Electrospun meshes were allowed to dry overnight in the chemical fume hood and then removed, wrapped in foil and stored in a sealed desiccator prior to use.

2.3. Scaffold preparation from electrospun meshes

Electrospun meshes were removed from parchment paper on collecting plate by gently lifting the edges of the mesh. Scaffolds were prepared from electrospun meshes for 96-well plates by using a 6 mm dermal biopsy punch to collect scaffolds from thicker parts of the electrospun meshes. Similarly, for 6-well plate cultures, electrospun meshes were cut into circular scaffolds of 3.4 cm diameter. Gelatin-containing scaffolds were then crosslinked using glutaraldehyde (Sigma-Aldrich) in the vapor phase for 24hrs in a sealed desiccator as previously described [24]. Thickness was measured for dry scaffolds and scaffolds that were wet in PBS for 24 hrs as described in Section 2.9 using an EnduraTEC ELF 3200 uniaxial tensile testing system (Fig. S2).

All scaffolds were then placed in 50 mL conical tubes (BD Falcon) and sterilized for 12 hrs with ethylene oxide (AN73, Andersen Products) and allowed to vent for 24 hrs. Gelatin-containing scaffolds were then treated with sterile filtered 0.1M aqueous glycine solution overnight to block excess aldehyde residues following glutaraldehyde treatment [25]. PCL scaffolds were wet using an ethanol dilution gradient as described previously [22]. All scaffolds were then washed 3 times with PBS and placed in medium (DMEM, 10% FBS, 1% Pen-Strep, ThermoFisher) overnight in a humidified 37 °C incubator with 5% CO₂. Cells were then seeded the next day after cell culture medium treatment.

2.4. Scanning electron microscopy

Scaffolds were collected after electrospinning and placed on scanning electron microscopy (SEM) sample holders using double-sided magnetic tape (Ted Pella). Samples were sputter coated with gold and imaged with a FEI Quanta 400 emission scanning electron microscope (FEI, ThermoFisher).

2.5. Confocal microscopy of fiber meshes

To visualize and assess fiber swelling for all groups, we incorporated rhodamine B dye (0.01% w/v, Sigma-Aldrich) in the 15% w/v solution of PCL in TFE for the pure PCL scaffolds and all coaxial groups. For the case of pure gelatin, we incorporated 0.01% w/v rhodamine B as well. Both PCL and gelatin fluoresced in the blue wavelength range in response to light of wavelength 405 nm, allowing them to be visualized with the DAPI excitation and emission filter set. Therefore, we imaged the blue and red channel to image dry and wet samples of fibers of variable composition. We used a Nikon A1-Rsi confocal set up with the 405 nm and 561 nm laser to visualize the surface of fibrous scaffolds in dry and wet conditions. Scaffolds were placed on sample holders with 1.5 glass on the bottom (Ibidi 81218) and imaged at with a 40× water objective. Single plane images are displayed in figures unless otherwise stated in the figure caption.

2.6. Randomized fiber diameter assessment

To randomize fiber diameter measurements, images of fibers that were taken using confocal microscopy with a 3× zoom on a 40× water objective were analyzed in ImageJ with FIJI plugins (NIH). Each image was split into thirds horizontally. At every point where a dividing line intersected a fiber, that fiber was measured for diameter at that point using the measure tool in ImageJ. Images were analyzed until n = 30 fibers had been randomly selected and measured.

2.7. Porosity and pore size calculation

Confocal microscopy projections of 20 μm sections were generated from z-stack images of dry and wet scaffolds with incorporated 0.01% w/v rhodamine B dye as described. Using IMARIS image processing software (V8.4, Bitplane), we were able to obtain the solid volume of fibers by generating a volume mask of the red-fluorescing fibers. By subtracting the solid fiber volume mask from the total volume, we calculated the porosity of fibrous meshes in dry and wet conditions.

To calculate the theoretical pore radii, we employed a previously published theoretical model where pores are assumed to be uniformly cylindrical void volumes between randomly oriented straight fibers [26]. We utilized the following equation to calculate the theoretical pore radii, 〈r〉 :

$$\langle r\rangle \cong \frac{\omega }{\text{ln}\left(1/\epsilon \right)}$$

Where ω is the mean fiber diameter and ε is the porosity of the electrospun mesh.

2.8. Cell culture, cell culture reagents and proliferation assays

The MG63 osteosarcoma cell line was used in all experiments. Cells were cultured in DMEM, high glucose basal medium (ThermoFisher) with 10% fetal bovine serum (Gemini Bio-Products) and 1% Pen-Strep (ThermoFisher) and were sourced from the Characterized Cell Line Core Facility at The University of Texas MD Anderson Cancer Center. Cells were seeded as monolayers at 5,000 cells per well in 96-well plates and at 10,000 cells per well in all 3D conditions to account for decreased seeding efficiency. For 6-well plates, the same seeding density was used scaled to the growth area of the plate. For imaging of monolayers, 12 mm diameter 1.5 glass coverslips (ThermoFisher) were placed at the bottom of 24-well tissue culture treated plates (Corning) and cells were seeded at the same density.

Total cell count was calculated from analyzing DNA content using a Quant-iT™ PicoGreen™ dsDNA assay kit (ThermoFisher). Cells from monolayers were lifted from 96-well culture plates using TrypLE Express (ThermoFisher) and collected in 1.7 mL microcentrifuge tubes and resuspended in 200 μL of Milli-Q H₂O. for all 3D groups, 6 mm scaffolds were washed in PBS and placed in 200 μL of Milli-Q H₂O. DNA was extracted from cells by subjecting the samples to 3 freeze-thaw cycles in liquid nitrogen and 37 °C water. The lysate was combined with TE buffer and dye solution as previously described as per manufacturer instructions [22]. Excitation and emission wavelengths of 485 nm and 538 nm were used, respectively to measure the total dsDNA (FLx800 fluorescence microplate reader; BioTek Instruments). A standard DNA curve was used to ensure that measured values were within the linear fluorescence regime. Sample cell number was derived from a dilution of a known number of MG63 cells prepared similarly from monolayers run on the sample plate.

2.9. Uniaxial tensile testing

For uniaxial tensile testing, an EnduraTEC ELF 3200 uniaxial tensile testing system was used (Bose). A water tank was filled with filtered water and mounted onto the apparatus to test the electrospun meshes in aqueous conditions. All samples were loaded with a starting length of 2.0 cm and subjected to an initial strain of 10% per min (adapted from ASTM D 882–02). For randomly oriented fiber samples, the width was 2 cm and for aligned fiber samples (n = 5 per group), the width was 1 cm. Thickness was measured by subtracting the zero-point distance value of the instrument clamps from the change in distance of the zero-point distance value when various samples were placed in between the clamp heads with the seeding surface area oriented towards the clamps. Tensile moduli were calculated by fitting a straight line to the linear portion directly following the toe region of the generated stress-strain curves. The slope of the fitted linear portion of the curve was then taken as the bulk modulus (E_B) of the mesh. To calculate individual fiber tensile moduli (E_i), we corrected the modulus for the void fraction and assumed all fibers to be the same diameter using the equation:

$$E_i=\frac{E_B}{1-\epsilon }$$

where ε is the porosity of the electrospun mesh.

2.10. Atomic force microscopy

Atomic force microscopy (AFM) was conducted in the University of Texas Health Science Center AFM Core Facility using a BioScope IITM Controller (Bruker Corporation). The acquisition of elastic measurements was performed with the Research NanoScope software version 7.30 and analyzed with the NanoScope Analysis Software version 1.40 (copyright 2013 Bruker Corporation). This system was integrated to a Nikon TE2000-E inverted optical microscope (Nikon Instruments Inc.) to facilitate the bright field imaging of the microfibers.

Polymer solutions of gelatin and PCL were prepared as stated in section 2.2. Samples of single fibers were prepared by depositing electrospun microfibers on 25 × 75 mm glass microscope slides (Superfrost™, ThermoFisher) taped with electrical tape onto the aluminum collecting plate. After a stable Taylor cone was achieved, the syringe pump and needle apparatus was moved at a speed of 50 mm/second across the stationary glass slide at a distance of 15 cm from the blunt tip needle. Glass slides were then carefully cut from the aluminum foil on the collecting plate taking care to not stretch individual fibers. Fibers were allowed to dry overnight in the chemical fume hood. Fibers containing gelatin were crosslinked as above. For gelatin containing fibers, fibers were hydrated for 24 hrs by adding 1 mL of PBS to the glass slide prior to indentation. PCL fibers were first wet with ethanol and subsequently with PBS overnight.

The fibers that did not overlap were selected using optical microscopy (20×) to perform AFM force measurements. The elastic modulus of indentation was measured on hydrated microfibers using non-conductive silicon nitride DNP-S cantilevers (f₀ = 40–75 kHz, k = 0.32 N/m, ROC = 10 nm) purchased from Bruker Corporation. Force curves were probed using a ramp size of 2 μm and a scan rate of 0.5 Hz, with a force load of 50 nN. The probe spring constant was determined prior to each experiment using thermal tuning. The modulus was calculated by fitting to a standard Sneddon model for a triangular indenter and a Poisson’s ratio of 0.45 (given to hydrogels). A minimum of three force measurements were captured for individual fiber samples (n = 3), which resulted in at least n = 30 measurements for each group. The first 30 measurements were used in analysis excluding data points made in errors of alignment to the fiber. For AFM, data points that were deemed by the operator of the UTHealth AFM Core Facility as either missing individual fibers or indenting the glass surface were excluded due to being orders of magnitude different than average values. AFM data analysis was performed with the Bruker data processing software (NanoScope Analysis version 1.40 copyright Bruker Corporation) to estimate the elastic modulus.

2.11. Immunofluorescence staining for confocal microscopy

Samples of monolayers on coverslips or electrospun scaffolds were washed in 1× PBS and fixed in a 4% paraformaldehyde solution (ThermoFisher) for 15 min at room temperature. Samples were then washed 3 times in 1× PBS on a shaker for 5 min before being blocked with blocking buffer (1× PBS, 5% normal goat serum, 0.3% Triton™ X-100) for 60 min. Primary antibodies were then diluted 1:100 in antibody dilution buffer (1X PBS, 1% BSA, 0.3% Triton™ X-100) and samples were incubated in primary antibody overnight at 4 °C. Primary antibodies used for all in vitro immunocytochemistry were targeted to YAP (Santa Cruz, sc-271134), TAZ (abcam, ab-84927), IGF-1R (Santa Cruz, sc-461) and pIGF-1R (Santa Cruz, sc-101703). Samples were then washed 3 times with 1× PBS before incubating with secondary antibody (diluted 1:500, Cell Signaling) and phalloidin conjugated to iFluor 488 (Abcam, diluted 1:1000) in antibody dilution buffer for 1.5 hrs in the dark at room temperature. After 3 washes, cells were incubated for 10 min in a 1:2000 dilution of a 1 mg/mL Hoechst stock solution (ThermoFisher). Samples were then rinsed once with PBS. Coverslips were mounted onto slides using Prolong® Gold Antifade Reagent (Cell Signaling). 3D scaffolds were placed in 1 mL of PBS and imaged on 35 mm 1.5 glass bottom dishes as sample holders (Ibidi 81218). Locations of images were selected in order to maximize the area of the sample in the focal plane of the seeding surface of the scaffolds, constructs or coverslips and for increased cellularity to maximize the number of cells in the analysis. We used a Nikon A1-Rsi confocal set up to collect images from each group using the appropriate laser wavelength and filters for each stain.

2.12. Image segmentation and calculation of nuclear:cytoplasmic ratio

All image analysis was performed using IMARIS 3D image processing software (V8.4, Bitplane). Z-stack images were processed using the native “Cells” algorithm to segment individual nuclei and cytoplasmic volumes for each image set. Nuclei were identified, split by seed points based on areas of max intensity, and segmented in the 405 nm channel with nuclear diameter = 8.00 μm, smoothing filter = 2.00 μm after background subtraction with sphere diameter of 1.00 μm to reduce noise. Intensity values for defining nuclei segmentation were automatically set after background subtraction using the native auto-threshold algorithm. Because PCL autofluoresces in the 405 nm range, individual nuclei were identified by filtering out and discarding from analyses any identified volumes with a sphericity below 0.45 on a scale from 0 to 1. Cytoplasmic volumes for entire image z-stacks were identified by positive staining for fluorescent actin in the 488 nm channel [27]. Smoothing filter width was selected to be 2.00 μm as with nuclei. Z-stack images were then segmented for total cytoplasmic volume with the native auto-threshold algorithm in the 488 nm channel. Integrated intensity values for YAP and TAZ stains for individual nuclei and entire volumes of cells were then exported as .csv files and analyzed in Microsoft Excel.

Individual nuclei were able to be identified by the IMARIS algorithm and mean nuclear intensities of each marker in individual nuclei were calculated by dividing the integrated intensity of each marker by the volume of each nucleus. Total nuclear volumes and nuclear integrated intensities were subtracted from total cellular volumes and integrated intensities to determine the average volume and integrated intensities for the cytoplasmic domain across all cells in the z-stack volume. To determine the nuclear to cytoplasmic ratio (N:C ratio) of the markers, total integrated signal intensities were then normalized to individual nuclear volumes or total cytoplasmic volumes and calculated for each unique nucleus identified in the z-stack. The formula is shown below:

$$N:C\mathit{\text{ratio}}=\frac{\frac{\mathit{\text{Integrated}}\mathit{\text{Signal}}{\mathit{\text{Intensity}}}_{\mathit{\text{Individual}}\mathit{\text{Nucleus}}}}{{\mathit{\text{Volume}}}_{\mathit{\text{Individual}}\mathit{\text{Nucleus}}}}}{\frac{\mathit{\text{Integrated}}\mathit{\text{Signal}}{\mathit{\text{Intensity}}}_{\mathit{\text{Cytoplasmic}}\mathit{\text{Domain}}}}{{\mathit{\text{Volume}}}_{\mathit{\text{cytoplasmic}}\mathit{\text{Domain}}}}}$$

Segmentation boundaries we overlaid on the original merged image and verified for accuracy before analysis. Raw values and calculated N:C ratio are provided as Supplemental Data Set 1.

2.13. Protein isolation and western blot analysis

For monolayer conditions, cells from 3 wells in a 6-well plate were lifted with TrypLE Express (ThermoFisher) and pooled into one sample. For cells cultured in 3D electrospun scaffolds, three samples from individual wells in the 6-well plate were washed once with PBS and pooled into one sample in a 50 mL conical tube (BD Falcon). Cells were then lysed on ice in 250 μL of lysis buffer (1% Triton X-100, 50 mM HEPES, pH 7.4, 150 mM NaCl, 1.5 mM MgCl₂, 1 mM EGTA, 100 mM sodium fluoride, 10 mM sodium pyrophosphate, 1 mM Na₃VO₄, 10% glycerol) containing freshly added protease and phosphatase inhibitors (Roche) as previously described [22]. Protein concentration for samples were measured using a micro BCA protein assay kit (ThermoFisher). 25 μg of protein in lysate per sample were run immediately using SDS-polyacrylamide gel electrophoresis (SDS-PAGE) or stored at −80 °C. After SDS-PAGE, proteins were transferred to PVDF membranes (EMD Millipore). Membranes were blocked using SuperBlock™ Blocking Buffer (ThermoFisher) for 1 hr and then allowed to incubate overnight in primary antibody at a dilution of 1:2000 for GAPDH loading control antibody and 1:1000 for all other antibodies. Primary antibodies for protein detection on membranes were all purchased from Cell Signaling Technology with the exception phosphorylated pYAP which was purchased from Abcam. After three consecutive washes in PBS with 0.1% Tween 20 and a final wash of PBS, signals were captured using horseradish peroxidase-conjugated secondary anti-rabbit IgG (Cell Signaling Technology) and visualized using SuperSignal West Dura Chemiluminescent Substrate (ThermoFisher). Images of chemiluminescent bands were then recorded using a ChemiDoc™ Imaging system (Bio-Rad).

2.14. Dose-response and drug testing experiments

Osteosarcoma cells were seeded onto 96-well plates in monolayer or 3D electrospun scaffolds as described in Section 2.8. Cells were allowed to adhere for 24 hrs in 200 μL of basal medium. After seeding, 100 μL of medium was removed from monolayers and 100 μL of medium with various therapeutic agents was added. For 3D conditions, scaffolds were sterilely moved from an original 96-well plate to a new plate with 200 μL of medium containing therapeutic agents at the appropriate concentrations. To generate dose-response curves for doxorubicin (Selleck), stock solution was diluted to 10 μM and then serially diluted 1:2 for nine total doxorubicin concentrations tested along with a no-drug control. For combination drug experiments, ridaforolimus (10 μM, Merck) and dalotuzumab (10 μg/mL, Merck) were diluted with medium containing the appropriate concentrations of doxorubicin such that the inhibitor concentration remained constant in every well. The no-doxorubicin control wells in each combination drug experiment contained only ridaforolimus or dalotuzumab respectively. All drug concentrations were applied in triplicate wells in separate 96-well plates for all groups. After 3 days of treatment, 100 μL of medium was removed from each sample well and replaced with a WST-1 solution in medium (Sigma-Aldrich) to a final concentration of 10% WST-1 solution in 200 μL of culture medium. Cells were allowed to incubate at 37 °C for 1 hr per manufacturer instructions. After incubation, 100 μL was removed from sample plates and placed in a new, clear 96-plate. Absorbance of the medium with formazan product was read at 440 nm with a reference wavelength of 650 nm in a PowerWave X 340 plate reader (Bio-Tek Instruments). Baseline medium only absorbance was subtracted from all sample values. Sample values with no doxorubicin were averaged and considered and used as the 100% live cell control for each experiment.

3. Results

3.1. Electrospun coaxial fiber meshes are highly porous substrates that mimic bone tumor microenvironments

We modeled the highly porous structures found in trabecular bone by electrospinning meshes comprising randomly oriented coaxial microfibers of poly(ε-caprolactone) (PCL) and gelatin. In our coaxial setup, PCL was used as the core polymer and gelatin was used as the shell polymer. This orientation of polymers in our fibers was chosen because PCL is much stiffer than gelatin, and because gelatin has a similar composition to collagen found in bone extracellular matrix and behaves as a hydrogel and swells in aqueous solutions (Fig. 1A) [28–30].

Figure 1: Fabrication and characterization of a 3D bone tumor microenvironment comprising electrospun coaxial fiber meshes.

(A) Coaxial electrospinning set up with dual syringes extruding into a coaxial needle. Polymer solution is extruded at variable flow rates, Q1 and Q2, at a distance, d, from the collecting plate. Positive voltage, V1, is applied at the coaxial needle while negative voltage or a ground, V2, is applied to the metal collecting plate. (B) Top: cross section of a coaxial needle demonstrating a red poly(ε-caprolactone) (PCL) core flow rate Q1 and a blue gelatin shell flow rate of Q2. Bottom: graphical representation of cross sections of single fibers collected in electrospun meshes with variable composition of PCL and gelatin derived from varying flow rates. (C) Representative scanning electron micrographs of meshes comprising five fiber types of varying ratios of core-shell PCL-gelatin composition. Top: meshes at 150× magnification, scale = 200 μm. Bottom: meshes at 1000× magnification, scale = 20 μm. (D) Representative confocal microscopy images of a single plane depicting fiber swelling in aqueous solution over time. Top: dry fibers. Middle: fibers that have been in aqueous solution for 24 hrs. Bottom: fibers that have been in aqueous solution for 7 days, scale = 50 μm. (E) Swelling of individual fibers evaluated by measuring individual fiber diameter over time (n = 30 fibers for each group per time point). (F) Porosity of meshes comprising variable fiber composition in dry (red) and wet (blue) conditions after 24 hrs of submersion in aqueous solution (n = 3). n.s. = no significance, *p < 0.05, **p < 0.01, #p ≤ 0.001, two-way analysis of variance (ANOVA) with post-hoc Tukey’s honestly significant difference (HSD) for (E) and multiple paired t-test for (F).

We adjusted the extrusion rates to modulate the mass ratio of each polymer in individual fibers and consequently, their bulk mechanical properties (Fig. 1B) [29, 31]. We electrospun pure PCL meshes, pure gelatin meshes, and three coaxial fiber meshes in the gelatin:PCL mass percentage ratios of 20:80, 50:50, and 80:20, hereafter referred to as 20%, 50%, and 80% gelatin coaxial, respectively, and examined the fibers with scanning electron microscopy (Fig. 1C). To examine swelling behavior in aqueous solution, we utilized confocal microscopy to image these five fiber types in dry conditions and after 24 hrs or 7 days of incubation in phosphate buffered saline (PBS) at 37 °C mimicking cell culture conditions (Fig. 1D). As expected, PCL fiber diameter did not change from dry to aqueous conditions, whereas meshes containing gelatin increased fiber diameter after 24 hrs (Fig. 1E). Interestingly, pure gelatin and 80% gelatin coaxial fibers continued to swell up to 7 days from initial submersion in PBS. Dry meshes of all compositions exhibited similar porosities, ranging, on average, from 76.8% to 83.0% (Fig. 1F). We calculated the theoretical pore radius of each scaffold in aqueous conditions using observed fiber diameter and porosities (Fig. S3) [26]. PCL scaffolds had the lowest theoretical pore radii of 8.7 ± 2.1 μm while scaffolds comprising 50% gelatin coaxial fibers had pores with theoretical mean radii of 28.0 ± 7.5 μm. As fibers containing gelatin are fused together at crosslinked junctions limiting the overall bulk mesh swelling, meshes with high gelatin content saw a greater reduction in porosity upon wetting than meshes with low gelatin content.

3.2. Increasing gelatin shell components of coaxial fibers decreases elastic moduli in macroscale and microscale testing

To begin mechanical characterization of the composite electrospun meshes, we determined the bulk mechanical properties of meshes comprising randomly aligned fibers. Macroscale tensile testing of cut strips of fiber meshes was performed in an aqueous, uniaxial mechanical testing system. We characterized the bulk tensile modulus in aqueous conditions for randomly oriented fiber meshes treating the highly porous meshes as a solid thin film (Fig. 2A). Tensile moduli were calculated from the linear portion of stress-strain curves generated by gradually increasing the strain of the samples at a rate of 10% per min (Fig. 2B). However, the recorded force and resulting calculated tensile modulus in this system does not take into account the void fraction of the material and can be heavily influenced by fiber-fiber interactions including covalent crosslinking, electrostatic forces, and individual fiber orientation.

Figure 2: Bulk tensile modulus of single fibers decreases as gelatin fraction is increased.

(A) Representative stress-strain curves of randomly oriented fiber meshes tested in a uniaxial tensile testing system. Figure legend for group color is the same for entire figure. (B) Calculated moduli mean and standard deviations for randomly oriented fiber meshes (n = 5). (C) Representative stress-strain curves of aligned fiber meshes tested in a uniaxial tensile testing system. Inset is complete curve of PCL over 30% strain. (D) Calculated moduli mean ± SD after propagation of error from porosity calculation for individual fibers (calculated from n = 5 uniaxially aligned fiber meshes, summarized in Table 1).

Therefore, in order to characterize the tensile properties of individual fibers, we proceeded to fabricate and test meshes of aligned fibers oriented in the direction of uniaxial strain and subsequently correct for porosity in aqueous conditions. Aligned fiber meshes were generated using the same ratios of polymer solutions for each fiber type by electrospinning onto a grounded mandrel rotating at 1000 RPM. Porosity was assessed as before; alignment and shape of fibers was checked by scanning electron microscopy (Fig. S4). While fibers were not perfectly parallel, these meshes served to reduce the influence of fiber-fiber interactions during increases in strain compared to meshes of randomly oriented fibers. Further, aligned fiber meshes would reduce the variability of forces generated from fibers not oriented in the direction of the uniaxial strain in tensile testing. Stress-strain curves were generated for aligned fiber meshes where the uniaxial force was applied in the direction of fiber alignment (Fig. 2C) and tensile moduli of fibers were calculated after accounting for the scaffold porosity (Fig. 2D). PCL and 20% gelatin coaxial fibers had similar tensile moduli, 56 ± 15 MPa and 59 ± 14 MPa, respectively. For fibers comprising increasing gelatin content, moduli were progressively reduced, and at the low end, pure gelatin had a tensile modulus of 0.069 ± 0.009 MPa (Fig. 2D, Table 1). This range spans three orders of magnitude in tensile moduli and recapitulates a similar range found in sections of trabecular bone [32, 33].

Table 1:

Summary of mechanical properties, quantitative image analysis of phenotypes, and responses to therapy for in vitro groups.

	Modulus: Uniaxial Tensile (MPa)	Modulus: AFM Indentation (MPa)	YAP N:C Ratio	TAZ N:C Ratio	IC50 to Doxorubicin (95% CI)
	Mean ± SD (n = 5)	Mean ± SD (n = 30)	Mean ± SD	Mean ± SD	Doxorubicin (μM) IC50 (95% CI)	Doxorubicin (μM) + Ridaforolimus IC50 (95% CI)	Doxorubicin (μM) + Dalotuzumab IC50 (95% CI)
Monolayer	-	-	3.0 ± 0.6 (n = 43)	2.9 ± 0.7 (n = 43)	1.30 (0.57 – 2.94)	0.13 (0.12 – 0.14)	0.07 (0.06 – 0.08)
PCL	56 ± 15	71 ± 70	1.3 ± 0.2 (n = 74)	1.3 ± 0.2 (n = 74)	0.51 (0.41 – 0.64)	0.39 (0.24 – 0.62)	0.57 (0.36 – 0.89)
20% Gelatin Coaxial	59 ± 14	10 ± 2	1.4 ± 0.6 (n = 65)	1.6 ± 0.7 (n = 65)	0.46 (0.27 – 0.78)	0.89 (0.73 – 1.10)	0.91 (0.69 – 1.20)
50% Gelatin Coaxial	20 ± 3	4 ± 3	1.8 ± 0.3 (n = 54)	1.7 ± 0.2 (n = 54)	0.52 (0.42 – 0.63)	0.57 (0.43 – 0.77)	0.62 (0.54 – 0.71)
80% Gelatin Coaxial	2.2 ± 0.8	2.3 ± 0.7	2.1 ± 0.4 (n = 62)	2.2 ± 0.4 (n = 62)	0.92 (0.72 −1.18)	0.93 (0.65 – 1.32)	1.02 (0.70 – 1.51)
Gelatin	0.069 ± 0.009	0.004 ± 0.004	2.1 ± 0.4 (n = 60)	2.2 ± 0.5 (n = 60)	0.60 (0.41 – 0.88)	0.94 (0.68 – 1.31)	2.05 (1.68 – 2.51)

Previous studies that have sought to elucidate the role of the mechanical microenvironment on MSC behavior have utilized microscale measurements of nanoindentation in conjunction with atomic force microscopy (AFM) to assess the moduli of homogeneous hydrogels [17, 34]. We confirmed by visual inspection that fibers were coaxial with fluorescent microscopy (Fig. 3A). To ensure that individual fibers would remain in place for AFM indentation, we viewed single fibers with phase contrast microscopy and performed a raster of the surface of the fiber and plotted the deflection of the cantilever in aqueous conditions (Fig. S5A). Elastic moduli were calculated from retraction curves after indentation with the AFM probe (Fig. 54). Similar to uniaxial tensile testing we observed decreasing calculated moduli as gelatin content was increased (Fig. 3B, summarized in Table 1). PCL fibers had the highest modulus of all fiber groups, 71 ± 74 MPa, with fibers having a wide range of moduli. All coaxial fiber groups harbored fibers with tensile moduli in the range of 1 to 10 MPa with calculated means of tensile moduli steadily decreasing with increasing gelatin. Pure gelatin fibers were also observed to have a wide range of moduli, 0.004 ± 0.004 MPa, which were four orders of magnitude lower than PCL and three orders of magnitude lower than coaxial fibers. The presence of the inner PCL core appeared to affect the calculated elastic moduli from AFM as elastic moduli were on the order of three orders of magnitude stiffer in coaxial fibers compared to pure gelatin.

Figure 3: Atomic force microscopy of individual fibers reveals decreasing moduli as gelatin fraction is increased.

(A) Representative epifluorescence microscopy image with green color filter of two fibers comprising PCL:gelatin (core:shell; 1:1 mass ratio) electrospun onto a glass slide; imaged in dry conditions. White arrows denote shell polymer (gelatin) boundaries and red arrows denote core polymer (PCL electrospun with 0.01% fluorescein dye). Digitally enhanced to increase contrast and resolve the gelatin shell in the image. (B) Calculated elastic moduli of single fibers undergoing AFM indentation displayed as mean ± SD (n = 30).

Taking the results of microscale and macroscale testing together, we observed a wide range of mechanical properties across our electrospun microenvironments that indicated that increasing the shell gelatin fraction of coaxial fibers with PCL cores resulted in decreased tensile and elastic moduli of individual fibers.

3.3. Osteosarcoma cells respond to the microenvironment by regulating YAP/TAZ expression and localization

YAP and TAZ, in addition to being regulated by the Hippo pathway, are potentially druggable targets in sarcoma that are independently regulated by substrate stiffness and mechanical signals [17, 35, 36]. Localization of YAP and TAZ to the nucleus in MSCs is associated with cell culture on top of stiffer hydrogel substrates and increased propensity towards osteogenic lineage commitment whereas less stiff environments favor an adipogenic phenotype and cytosolic YAP and TAZ [34]. Because osteosarcoma is a tumor of mesenchymal origin, we hypothesized that osteosarcoma cells cultured on substrates of variable mechanical stiffnesses would modulate the localization of Hippo pathway effectors, proliferation of cells, and responses to therapy [37, 38].

We employed our randomly oriented electrospun meshes of variable mechanical properties, which mimic the structure of trabecular bone, as scaffolds for the culture of MG63 osteosarcoma cells to evaluate the effects of architecture and mechanical microenvironment on osteosarcoma. While we fabricated meshes with uniaxial alignment to fully characterize the mechanical properties of the microfibers, we did not use them as culture substrates to avoid surface topography and alignment as a confounding factor to cell response in our studies. Lower magnification images of cells grown in the 3D scaffolds indicate that MG63 cells penetrated the surface of fiber meshes and grew in and around fibers and adjacent cells in a 3D structure (Fig. S5).

To assess the mechanoresponsive ability of osteosarcoma cells, we co-stained samples for nuclei, actin cytoskeleton, YAP, and TAZ. We collected z-stacks of images using confocal microscopy that spanned 50 μm in depth and segmented the reconstructed 3D images for nuclear and cytoplasmic domains based on nuclear and actin stains, respectively (Fig. 4A) [27]. We then quantified the mean nuclear signal of YAP and TAZ for individual nuclei and the mean cytoplasmic signal throughout all cytoplasmic domains because individual cells were not able to be segmented in our dense cultures (Fig. 4B and 4C). Monolayers harbored the highest nuclear to cytoplasmic ratio (N:C ratio) of both YAP and TAZ which was expected given the supraphysiological stiffness of these culture environments. Interestingly, for 3D electrospun environments, as stiffness was decreased, YAP and TAZ appeared to have progressively increasing levels of nuclear localization of YAP and TAZ. N:C ratios of YAP and TAZ were lowest in PCL environments and highest in gelatin environments with a gradual increase in scaffolds comprising fibers with increased gelatin content and decreased bulk elastic moduli.

Figure 4: Osteosarcoma cells respond to 3D environments by modulating the expression and localization of Hippo pathway effectors.

(A) Representative confocal microscopy z-stack projections of MG63 osteosarcoma cells cultured for 5 days co-stained with Hoechst (nuclei, blue), phalloidin (actin, green), YAP (red), TAZ (cyan) with overlays of cellular and nuclear segmentation, scale = 25 μm. (B and C) Localization of YAP and TAZ based on ratio of mean intensity of nuclear staining to cytoplasmic staining (N:C ratio). Values were quantified from z-stack image sets from the surface of groups to a depth of 50 μm (n = 3). Single nuclei are displayed as points with mean ± SD. Letters indicate separate statistical groups p < 0.05, one-way ANOVA with post-hoc Tukey’s HSD.

Interestingly, western blot analysis of total cell lysate from cultures grown in each condition showed a progressive downregulation of YAP expression in environments with lower tensile moduli (Fig. 5). Conversely, TAZ expression levels remained unaffected in all group regardless of stiffness or culture architecture. Phosphorylated (inactive) YAP was readily detected in the monolayer control with fainter bands in all 3D conditions likely due to increased overall YAP in the stiffer conditions. Although YAP expression is dramatically shifted in our models, AXL, a receptor tyrosine kinase that has been implicated in other cancers such as hepatocellular carcinoma and lung adenocarcinoma as a key downstream target of YAP, was constant in all conditions [39, 40].

Figure 5: Osteosarcoma cells downregulate YAP but not TAZ in environments with decreasing tensile moduli.

Western blot highlighting YAP/TAZ total protein expression and activation in environments of decreasing stiffness.

Taken together, our results indicate that osteosarcoma cells modulate YAP and TAZ localization in response to bulk tensile properties of 3D environments and are likely differentially regulated in 3D models versus standard monolayer culture. YAP is drastically downregulated while TAZ expression remains constant, implying that YAP and TAZ are regulated separately in 3D models.

3.4. Osteosarcoma cells upregulate Sox2 and downregulate IGF-1R/mTOR in 3D environments

Key mediators of the IGF-1R/mTOR signaling cascade have long been attractive candidates for targeted therapy in many cancers including sarcomas since aberrant activation of this pathway is associated with an increase in aggressiveness and growth of tumors [41–43]. While preclinical models had shown great promise for IGF-1R/mTOR targeted therapy, the modest responses seen in patients receiving this therapy has led many to believe that if IGF-1R/mTOR targeted therapy is approved for sarcoma, it will be used in combination therapy and patients will likely need to be stratified based on IGF-1R/mTOR expression [18, 19, 44]. Having established that our scaffolds could induce an architecturally-driven, graded mechanoresponse in YAP and TAZ localization and that IGF-1R/mTOR targeted therapy may rely on phenotypes that are mechanically sensitive, we hypothesized that our mechanically tunable 3D models could be used to assess drug resistance patterns among osteosarcoma cell populations with variable levels of IGF-1R/mTOR activation. Specifically, we hypothesized that stiffer environments would induce greater IGF-1R/mTOR signaling evidenced by increased phosphorylated IGF-1R, mTOR, and its downstream effectors phosphorylated S6 (pS6) and phosphorylated eukaryotic translation initiation factor 4E (peIF4E). We further hypothesized that in less stiff environments and by mimicking the 3D architecture observed in osteospheres in a 3D scaffold such in as our models, cells would induce Sox2 expression and generate a chemotherapy-resistant phenotype [45–47].

We assessed the proliferation in our 3D models compared to monolayers to determine if rates of proliferation were drastically altered in our models. All culture conditions facilitated the growth of osteosarcoma cells after day 1. However, the 3D environments with higher tensile moduli (PCL, 20% gelatin coaxial, and 50% gelatin coaxial) did not foster proliferation of cells from days 3 to 5, possibly because the cultures had become confluent in the 3D environments (Fig. 6A). All cultures appeared confluent by Day 5 (Fig. S5).

Figure 6: Osteosarcoma cells downregulate IGF-1R/mTOR in 3D.

(A) MG63 osteosarcoma cell proliferation in variable environments quantified by total DNA and converted to cell count using a sample of known cell number (n = 3). Two-way ANOVA with post-hoc Tukey’s HSD, *p < 0.05, **p < 0.01, ***p < 0.001, #p ≤ 0.001, n.s. = no significance. (B) Representative western blots highlighting Sox2 expression and activation of the IGF-1R/mTOR axis activation.

Interestingly, all 3D models facilitated the upregulation of Sox2, a marker for osteosarcoma cancer stem cells, compared to monolayer controls regardless of substrate stiffness (Fig. 6B) [8]. We observed a marked downregulation of phosphorylated IGF-1R (pIGF-1R) in all 3D samples while total IGF-1R remained constant. Congruently, there was a reduction in mTOR detected in each sample for 3D groups. Further, there was a modest reduction in the downstream mTOR effector peIF4E but not total eIF4E in 3D groups compared to monolayer controls. However, phosphorylated and total S6 protein, a direct effector of mTOR, did not appear to be different across groups. Taken together, 3D environments facilitated the downregulation of active IGF-1R and mTOR but appear to have a lesser, but significant, effect on downstream effectors of the mTOR pathway.

3.5. Osteosarcoma cells grown in 3D conditions become resistant to IGF-1R/mTOR targeted therapy

Because both IGF-1R and Sox2 have been implicated as negative prognostic indicators but seem to be inversely correlated in our models, we decided to calculate a 50% inhibitory concentration (IC₅₀) for doxorubicin, a standard chemotherapeutic agent used in the treatment of osteosarcoma, using dose-response curves generated from serial dilutions of the drug (Fig. 7) [44, 48, 49]. Although the higher calculated IC₅₀ would suggest that monolayer controls are more resistant to doxorubicin treatment over 3 days, the 95% confidence intervals (95% CI) actually overlapped in all groups, indicating that culture architecture and substrate stiffness have little effect on response to doxorubicin on its own and calculated IC₅₀ values overall (drug responses summarized in Table 1).

Figure 7: Tumor microenvironments in 3D are resistant to combination chemotherapy and IGF-1R/Mtor targeted therapy compared to monolayers.

(A) Fit dose-response curves to doxorubicin alone and in the presence of 10 μM ridaforolimus (mTOR inhibitor) or 10 μg/mL dalotuzumab (IGF-1R targeted antibody). Results displayed as mean ± S.E. (n = 3 for each concentration tested in each condition). (B) Calculated IC₅₀ values for doxorubicin with 95% confidence intervals (95% CI) calculated from the inflection points of the fit dose-response curves in (A).

Ridaforolimus, a small molecule mTOR inhibitor, and dalotuzumab, a humanized monoclonal antibody to IGF-1R, have been the subjects of preclinical testing and clinical trials as targeted therapies for sarcoma and other cancers [18–20, 50, 51]. When either ridaforolimus or dalotuzumab were added to MG63 cultures at singular concentrations, the IC₅₀ of doxorubicin for monolayers plummeted to 0.13 μM (95% CI: 0.12 – 0.14 μM) and 0.07 μM (95% CI: 0.06 – 0.08 μM), respectively, from 1.30 μM (95% CI: 0.57 – 2.94 μM), which is consistent with previous studies showing that targeting mTOR or IGF-1R has a synergistic effect with doxorubicin in reducing proliferation of cancer cells [52]. In a stark contrast, these targeted agents did not act synergistically with doxorubicin to inhibit proliferation in 3D conditions but either had no effect or increased the IC₅₀ of doxorubicin, suggesting that the 3D environment induces a change in cell phenotype that actually makes them less susceptible to the drug when combined with IGF-1R/mTOR targeted therapy (Fig. 7B, Table 1). Interestingly, while overall proliferation rates of osteosarcoma in our models compared to monolayers remained similar for all groups, targeting the IGF-1R/mTOR pathway only affected the monolayer groups that harbored more active IGF-1R. Our results indicate that activation of the IGF-1R/mTOR pathway is at least in part dependent on architectural mechanical cues and affects the response to combination therapies that target this signaling cascade.

4. Discussion

The mechanical properties of the osteosarcoma tumor microenvironment are difficult to assess in patients. Normal bone is generally categorized as cortical bone, where the majority of mineralized calcium exists, and trabecular bone, an area comprising a hard, mineral component with a high degree of porosity. The “void,” non-mineralized portions of trabecular bone are not truly void but rather are constituted by a mix of collagenous matrix, fat, various cell populations and vasculature making heterogenous mechanical forces from the environment difficult to measure and model [34, 53]. Further complicating the assessment of bone tumor microenvironments in non-cancerous tissue, osteosarcoma destroys the native bone tissue and compromises the structural integrity to a degree where pathological fracture can occur and in some cases, is the presenting symptom [54].

YAP and TAZ have been the subject of recent clinical interest with some suggesting that expression level in tissue sections, localization to the nucleus, or upregulation of downstream effectors may herald chemoresistance and worse outcomes for osteosarcoma patients [36, 55, 56]. Other groups have shown that localization of YAP and TAZ in MSCs is sensitive to not only substrate stiffness but also to cell shape, cell confinement, and 3D environments [17, 27, 57, 58]. It remains less clear how these mechanical cues contribute to phenotypic changes and pathological YAP/TAZ signaling in tumors of mesenchymal origin such as osteosarcoma [13, 55, 59].

In this work we engineered a bone tumor niche comprising fibers that had tensile moduli in a range spanning three orders of magnitude in macroscale uniaxial tensile testing and elastic moduli in a range spanning four orders of magnitude assessed by microscale AFM measurements while maintaining similar 3D architectures. Calculated moduli for gelatin fibers were in the range of similar hydrogels [60]. Interestingly, the presence of the inner PCL core greatly affected the calculated elastic moduli from AFM as elastic moduli for coaxial fibers were on the order of three orders of magnitude higher than those calculated for pure gelatin. This could be a result of the selected retraction “trigger” force that caused the probe to interact with the core PCL component under the thinner and softer gelatin component in coaxial fibers but not in pure gelatin fibers where the trigger force was never reached [61].

In response to different mechanical environments in our models, osteosarcoma cells modulated the expression and localization of YAP and the localization of TAZ. YAP was downregulated in a dose dependent manner corresponding to decreasing calculated moduli of the substrate with monolayers on hard tissue culture polystyrene expressing the highest level. In contrast, TAZ expression appeared to remain unaffected in osteosarcoma cells cultured in 3D substrates of variable stiffness. This result suggests that YAP and TAZ expression and/or degradation may be regulated by different stiffness related mechanisms and by the architecture of the culture environment. Since our model system incorporates a gelatin shell, there is the possibility that changes in YAP and TAZ expression could occur in response to the cells sensing a microenvironment that degrades over time and thus either facilitates cell spreading, invasion, and migration or otherwise changes mechanical properties over time [62, 63]. However, the scaffolds were fully crosslinked with glutaraldehyde and maintained their structure in aqueous conditions over the course of 7 days (Fig. 1E) and the period of cell culture was only 5 day, and up to 8 days in the drug studies; we did not expect extensive scaffold degradation within that time frame, especially without the addition of exogenous active enzyme in the cell culture media [24, 64]. Because of the potential for mass loss, one limitation of the gelatin-containing model is that it is restricted to investigating phenomena such as cell response to architecture and combination therapy, as we did in the present study, over the course of days or weeks rather than months.

Previous studies probing the effect of substrate stiffness on MSCs have involved culture on the surfaces of hydrogels of variable stiffness [17, 34]. These studies attempted to engineer microenvironments much softer than standard tissue culture polystyrene surfaces, which typically have moduli in the range of 2–4 GPa [65]. These studies reported an increase in nuclear YAP/TAZ in cells cultured on progressively stiffer hydrogel environments. In contrast to those studies, we observed that in 3D electrospun microenvironments, less stiff environments contributed to a greater degree of YAP/TAZ nuclear localization, although less than standard culture methods on tissue culture polystyrene. These results suggest a complex regulation of YAP/TAZ that is greatly affected by 3D culture. Indeed, many have suggested that it is not substrate stiffness per se that affects YAP/TAZ localization but the resultant effects on the cytoskeletal organization that allow the cell to “sense” the mechanical microenvironment as the actomyosin cytoskeleton is required for mechanotransduction (reviewed in [66]). Given this, it is reasonable to hypothesize that 3D structures will have differential effects on cytoskeletal organization and therefore regulate mechanotransduction of substrate stiffness differently than monolayer cultures. Interestingly, the increased N:C ratio of YAP and TAZ as bulk tensile modulus is decreased in environments observed in our models is consistent with a previous study involving 3D encapsulated cells in hydrogels of variable stiffness which found differential, actin-dependent regulation of YAP/TAZ for cells encapsulated versus cultured atop hydrogels [27].

Furthermore, while all coaxial fibers and the pure gelatin fibers incorporated crosslinked gelatin at the surface of the growth substrate, increasing nuclear YAP and TAZ was observed as gelatin content was reduced suggesting that surface properties do not account for all changes in N:C ratio of these Hippo pathway effectors. Our system afforded us the opportunity to keep gelatin as the surface substrate in the coaxial and pure gelatin fibers. Thus, we were able to determine that osteosarcoma cells likely sense below the surface layer of gelatin to the much stiffer PCL substrate beneath which is reflected in our AFM measurements indicating a difference between coaxial fibers and pure gelatin fibers in elastic moduli. This result is consistent with previous studies involving MSC culture on collagen or polyacrylamide hydrogels that indicated that cells indeed sense substrates below the surface level to at least a depth of 10 μm and regulate gene expression and differentiation accordingly [67, 68].

A limitation in our models is that it is difficult to image more than a few tens of microns below the surface of 3D scaffolds as the PCL in fibers is opaque and both gelatin and PCL can distort fluorescent signal that is collected by the microscope. Nevertheless, we were able to capture z-stack images encompassing the surface of meshes to a depth of about 50 μm but did not include any cells that may have been residing further within our models in the analysis.

In the current study, we were interested in mechanotransduction in osteosarcoma, which is a highly cellular lesion. For this reason, cells cultured in our environments were grown to near confluency, and while individual nuclei were easily segmented for image analysis, corresponding cytoplasms are difficult to assign to individual nuclei. To combat this, we generated automatically thresholded image masks for the entirety of the cytoplasm in each z-stack and calculated the average intensities for the cytoplasmic volume in each sample. Therefore, N:C ratios calculated in our studies reflect values from individual nuclei and average cytoplasmic values. Because of the homogeneity in our culture substrate and the use of a monoclonal cell line, we believe this method was a valid way of assessing highly confluent cell cultures via image segmentation.

The IGF-1R/mTOR signaling cascade has significant crosstalk with mechanotransduction pathways that modulate the Hippo pathway effectors at multiple levels [69–72]. Additionally, the IGF-1R/mTOR cascade has been implicated as a driver of anti-apoptotic signals and aberrant growth in many cancers [73, 74]. Interestingly, as substrate stiffness in our 3D models was reduced, IGF-1R expression remained unchanged whereas mTOR appearance on western blot decreased in a dose dependent manner similarly to YAP. This result is consistent with previous studies, which report that YAP may potentiate mTOR activation [75, 76].

In addition to decreasing levels of mTOR, the almost complete abolition of pIGF-1R signal in all 3D groups coupled with relatively similar levels of downstream effectors of mTOR suggests a complex mechanical regulation at multiple levels of the IGF-1R/mTOR signaling cascade. The similar levels of pS6 and peIF4E across all groups in the context of decreasing total YAP as substrate stiffness was decreased suggests that nuclear YAP, as evidenced by the increased N:C ratio of YAP in groups with decreasing moduli, is a better indicator of mTOR potentiation. More studies will be required to assess to what degree N:C ratio and total expression of YAP govern the activation or potentiation of mTOR signaling in osteosarcoma. Downregulation of IGF-1R signaling mechanisms coupled with similar proliferation rates suggests the generation of an architecture-dependent osteosarcoma phenotype that is not reliant on IGF-1R/mTOR signaling for continued growth.

Others have reported the generation of Sox2-positive osteosarcoma stem cells that are recalcitrant to therapy in suspension cultures where there is increased cell-cell contact compared to monolayer cultures [8, 77]. Consistent with this observation, increased cell-cell contact mediated by our 3D environments upregulated Sox2 regardless of substrate stiffness. Further, the upregulation of Sox2 coupled with downregulation of the IGF-1R/mTOR cascade heralded increased resistance to combination therapy targeting IGF-1R/mTOR in osteosarcoma cultured in 3D environments compared to monolayer controls.

5. Conclusion

In this study we have sought to determine the role of substrate mechanical properties and culture architecture in osteosarcoma pathogenesis by engineering and characterizing mechanically tunable tumor microenvironments. We determined that 3D environments of decreasing stiffness had higher nuclear localization of YAP and TAZ in contrast to studies done atop hydrogels of variable stiffness. 3D cultures of osteosarcoma cells on substrates of varying mechanical properties induced downregulation of the IGF-1R/mTOR signaling cascade and upregulation of Sox2 compared to standard monolayers regardless of calculated moduli. The downregulation of IGF-1R/mTOR activation likely contributed to the decreased efficacy of IGF-1R/mTOR targeted combination therapy. Our results suggest that osteosarcoma cells sense a complex mechanical microenvironment that is greatly influenced by not only the stiffness but the architecture of the substrate. Electrospinning is a versatile technique that can produce conditions that are highly malleable with both synthetic and natural polymers. This versatility can be leveraged to model more complex and specialized tumor niches for which we have provided the framework and highlighted the importance.

Statement of Significance.

The tumor microenvironment plays a critical role in cancer pathogenesis. In this work, we engineered 3D, mechanically tunable, coaxial electrospun environments to determine the roles of the mechanical environment on osteosarcoma cell phenotype, morphology, and therapeutic response. We characterize the effects of varying macroscale and microscale stiffnesses in 3D environments on the localization and expression of the mechanoresponsive proteins, YAP and TAZ, and evaluate IGF-1R/mTOR pathway activation, a target of recent clinical trials in sarcoma. Increased nuclear YAP/TAZ was observed as stiffness in 3D was decreased. Downregulation of the IGF-1R/mTOR cascade in all 3D environments was observed. Our study highlights the complexity of mechanotransduction in 3D culture and represents a step towards controlling microenvironmental elements in in vitro cancer investigations.