ABSTRACT
There is an urgent need for accurate, scalable, and cost-efficient models of the complexity and heterogeneity of the tumor microenvironment. Here, we detail how to fabricate and use the Metabolic Microenvironment Chamber (MEMIC) – a 3D-printed ex vivo model of intratumoral heterogeneity. A major driver of the cellular and molecular diversity in tumors is the accessibility to the blood stream that provides key resources such as oxygen and nutrients. While some tumor cells have direct access to these resources, many others must survive under progressively more ischemic environments as they reside further from the vasculature. The MEMIC is designed to simulate the differential access to nutrients and allows co-culturing different cell types, such as tumor and immune cells. This system is optimized for live imaging and other microscopy-based approaches and it is a powerful tool to study tumor features such as the effect of nutrient scarcity on tumor-stroma interactions. Due to its adaptable design and full experimental control, the MEMIC can provide novel insights into the tumor microenvironment that would be difficult to obtain via other methods. As a proof of principle, we show that cells can sense gradual changes in metabolite concentration, and tune intracellular cell signaling to form multicellular spatial patterns of cell proliferation. We also show that ischemic macrophages reduce epithelial features in neighboring tumor cells highlighting the power of this system to study cell-cell interactions and non-cell autonomous effects of the metabolic microenvironment. We propose that the MEMIC can be easily adapted to study early development, ischemic stroke, and other systems where multiple cell types interact within heterogeneous environments.
INTRODUCTION
The tumor microenvironment is a complex cellular ecosystem (Fig. 1A). This complexity includes a large diversity of malignant and non-malignant cells such as fibroblasts and immune cells (Junttila and Sauvage, 2013; Merlo et al., 2006; Tabassum and Polyak, 2015). These cells secrete and consume molecular signals, growth factors, and metabolites, creating an intricate biochemical landscape in the tumor interstice that in turn affects tumor cell behaviors and cell-cell interactions (Bader et al., 2020; Buck et al., 2017; Mehta et al., 2017; Wiseman and Halliwell, 1996).
We and others have shown how insufficient tumor vascularization can produce predictable spatial patterns of cell phenotypes (Carmona-Fontaine et al., 2017b). Cells located near functional blood vessels are constantly perfused with nutrients and oxygen. In contrast, cells distant from vasculature, or proximal to faulty vessels commonly found in solid tumors (Mazzone et al., 2009), are in a microenvironment poor in nutrients and profuse in potentially toxic metabolic waste products (Carmeliet and Jain, 2000a; Gatenby and Gillies, 2004; Hobson-Gutierrez and Carmona-Fontaine, 2018a; Thomlinson, 1977a). The role that nutrients and other metabolites have in modulating the behavior and phenotypes of tumor – and especially immune cells – has been the subject of recent interest (Buck et al., 2017; Hobson-Gutierrez and Carmona-Fontaine, 2018a; Olenchock et al., 2017a; Pavlova and Thompson, 2016a). However, the lack of amenable models of the metabolic microenvironment of tumors hampers the ability to predict and control the effect of environmental metabolites on tumor cells.
Animal models are a fundamental tool to study the complex and heterogeneous tumor microenvironment (Day et al., 2015; Gould et al., 2015). However, the complexity of animal physiology – while crucial in pre-clinical studies – can challenge the isolation of individual experimental variables, and their use for large experiments is severely limited by practical, economical, and ethical concerns (Bert et al., 2017; Bressers et al., 2019). On the other side of the spectrum, conventional in vitro experiments offer much better experimental control and can be easily used in high throughput approaches. However, these cultures do not model the metabolic heterogeneity and other essential features of the tumor microenvironment. The recent resurgence in the use of three-dimensional tumor organoids – or tumoroids – as a tool to model different aspects of tumor biology does offer some of these features (Clevers, 2016). Tumoroids can recapitulate key histopathological tumor characteristics, and they can be used to screen for patient-specific drug responses (Boj et al., 2015; Gao et al., 2014; van de Wetering et al., 2015). However, the organization of tumoroids emerges spontaneously and thus the visualization, quantification, and prediction of their organization remains challenging (Fig. 1B).
We previously developed a microphysiological system that mimics the complexity of the tumor microenvironment in a well-controlled and predictable manner. This Metabolic Microenvironment Chamber (MEMIC) is suitable for high resolution microscopy analyses and can be easily adapted to the complexity and throughput that different experimental scenarios may need (Carmona-Fontaine et al., 2017b). Cells in the MEMIC are gradually limited in their access to fresh media, generating gradients of extracellular metabolites and oxygen across the chamber where they are cultured. This metabolic heterogeneity can be accompanied by the addition of other components of the tumor microenvironment, such as stromal cells, an extracellular matrix, and perturbations with carcinogens or drugs. Compared to the methods mentioned above, the spatiotemporal complexity that emerges in the MEMIC is predictable, reproducible, and measurable.
Here, we expand on key features of the MEMIC and provide detailed guidelines on how to use this system. We determined key parameters that shape metabolic gradients in the MEMIC, which we describe alongside detailed information on how to assemble the platform, how to set up cultures of tumor cells – alone or in co-culture – and how to monitor these experiments using live imaging and fixed endpoint microscopy assays, such as immunofluorescence. We demonstrate that the MEMIC accurately captures the cellular response to nutrient and oxygen deprivation that occur in vivo, and we show that nutrient-deprived macrophages reduce epithelial features in neighboring tumor cells. Finally, we provide an image analysis pipeline designed to obtain information at the single-cell level from MEMIC images suitable for users without any coding experience. We propose the MEMIC as a complement to standard in vitro and in vivo experiments, diversifying the tools available to accurately model, perturb, and monitor the tumor microenvironment, as well as to understand how extracellular metabolites affect other processes such as wound healing and stem cell differentiation.
RESULTS
MEMIC – an overview
A hallmark of the microenvironment of virtually all solid tumors is the presence of hypoxic and poorly nourished niches (Gatenby and Gillies, 2008; Hobson-Gutierrez and Carmona-Fontaine, 2018b; Lyssiotis and Kimmelman, 2017; Thomlinson, 1977b). These conditions are the result of the increased growth of tumor cells and insufficient blood perfusion (Baish and Jain, 2000; Carmeliet and Jain, 2000b; Pavlova and Thompson, 2016b). Because tumor growth and tumor vascularization are not uniform, they create a heterogeneous ‘metabolic microenvironment’ where some cells experience near physiological conditions, whereas others endure severe ischemia, and potentially cell death, due to lack of nutrients and accumulation of toxic waste (Carmona-Fontaine et al., 2013; Gatenby and Gillies, 2008; Thomlinson, 1977b). The MEMIC is a 3D-printed microphysiological culture system specifically designed to model this spectrum of metabolic conditions (Video 1). In addition, the MEMIC allows co-culturing any number of cell types to study how different cells interact and behave in different metabolic niches (Carmona-Fontaine et al., 2017b).
To generate these gradients of metabolic conditions, cells in the MEMIC grow inside a small chamber that is connected – only from one end – to a large reservoir of fresh media (or outer chamber, Fig 2A,B). The concentration gradients of extracellular metabolites in the MEMIC are established by a balance between their diffusion rate and their rates of consumption and secretion by cells in the small chamber. Cells proximal to the opening, therefore, have a constant supply of nutrients and oxygen and the metabolic byproducts they secrete are quickly cleared by fresh media from the reservoir. In contrast, cells at the opposite end of the small chamber endure a microenvironment where oxygen and nutrients are scarce and metabolic byproducts accumulate (Fig 2C).
The MEMIC framework described here is composed of 12 independent MEMIC chambers facilitating multiplexing, technical replicates, and no-gradient controls (Fig 2A). The small chamber where cells grow is sealed by optical glass coverslips, and thus it is ideal for high resolution microscopy analyses (Fig 2B). The 12-MEMIC framework has the same footprint as a multi-titer plate and is therefore compatible with conventional equipment such as microscope stages. The main requirement to fabricate a MEMIC is access to a 3D printer, which is inexpensive when compared to most lab equipment, and they are often available as a shared resource at universities and research institutions. In the Supplementary Information linked to this article, we provide a detailed protocol and a video on how to 3D print and assemble this system (Fig S1, Video 2, SI Fabrication).
Cellular responses to nutrients and oxygen gradients
To observe the cellular responses to metabolic gradients, we quantified the activation of key nutrient and oxygen sensing pathways. The mammalian target of rapamycin (mTOR) pathway is a major nutrient sensor and a regulator of anabolic metabolism (Saxton and Sabatini, 2017). We first cultured human colorectal adenocarcinoma cells (DLD1) in the MEMIC for 24h and we then used immunofluorescence to detect phosphorylated S6 (p-S6) as a read-out of TORC1 pathway activation. For a detailed description on how to seed cells in the MEMIC and to process them for immunofluorescence, please refer to the Supplementary Information (Fig S2, SI Seeding and processing). Cells in these cultures, show high levels of S6 phosphorylation near the opening of the MEMIC but these levels decrease dramatically under ischemic environments (Fig 2D). Conversely, Hypoxia-inducible factor 1-alpha (HIF1α) is a transcription factor that regulates the cellular response to hypoxia. In the absence of oxygen, HIF1α is stable and accumulates in the nucleus, otherwise it is targeted for proteasomal degradation. HIF1α is barely detectable near the opening of the MEMIC, but its levels steeply increase in ischemic regions (Fig 2D). A similar decrease in activation of the mTOR pathway (Palm et al., 2015) and increase of HIF1α (Zhong et al., 1999) has been observed in solid tumors in vivo.
To acquire information at the single-cell level, we wrote software that rapidly detects cell nuclei, segments individual cells and obtains key parameters such as their position, area, and fluorescence intensity (Fig 2E). This approach is similar to flow cytometry in that it quantifies the fluorescence of individual cells, but unlike flow cytometry, it retains critical spatial information. To help the wider research community use this image cytometry approach, we provide MATLAB scripts and a step-by-step tutorial fit for scientists without any coding experience (see Image cytometry section in Supplementary Information).
We used this image cytometry approach to analyze the images shown in Fig 2D. We first plotted HIF1α level for every cell versus its p-S6 levels (Fig 2F). To incorporate spatial information, we color-coded each cell on the plot according to its distance from the MEMIC opening. To emphasize the role of positional information and the spatial patterns that emerge we can plot the fluorescence levels of each cell versus its distance from the opening of the MEMIC as shown in Fig 2G. The quantification shows an exponential decrease in mTOR activation as cells are further away from the opening – consistent with a gradual decrease in nutrient availability. By contrast, we observed a sharp increase in nuclear HIF1α within ischemic areas of the MEMIC. Although oxygen levels are expected to form a smooth gradient, we observe a stepwise increase in HIF1α levels. This digital response emerges from the threshold of oxygen tension that determines whether HIF1α is degraded or not (Lee et al., 2015; Wong et al., 2011). Overall, these results illustrate the formation of metabolite concentration gradients in the MEMIC and show that these gradients are translated into finely tuned intracellular signaling pathways.
Shaping metabolic gradients in the MEMIC
Interstitial levels of extracellular metabolites in vivo are determined by a balance of metabolite diffusion and cellular activities such as consumption and secretion rates. As such, the slopes of concentration gradients of these metabolites vary according to tissue crowding, nutrient transporter levels, growth rate, and other parameters that are difficult, or impossible, to untangle in vivo. Gradients in the MEMIC are formed by the same principles, however, in this system we are able to tune parameters independently and shape these gradients according to specific experimental needs.
To accurately measure and tune parameters in the MEMIC, we used a fluorescent hypoxia reporter as a readout for metabolic gradients ((Vordermark et al., 2001) see SI for details). We engineered human breast tumor cells (MDA-MB-231) to express GFP under the control of five HIF1α binding sites (Fig 3A). As a fluorescence reference, we also expressed a membrane-bound form of mCherry in these cells (mCherry-CAAX, see SI for details).
Denser cultures should form steeper gradients as more cells consume nutrients and oxygen. To test the effect of cell density on these gradients, we seeded different cell numbers into the MEMIC and measured GFP levels after 24 hours. With the lowest cell density (125 cells/μL or ∼12% confluency), GFP signal was not detectable, indicating that the collective oxygen consumption was not enough to activate a hypoxic response even in the far end of the chamber (Fig 3B). At higher cell densities, however, GFP signal becomes evident and its intensity gradients are steeper, have a larger amplitude, and peak closer to the opening. At the highest cell density (1000 cells/μL or >100% confluency) GFP levels peak almost at the opening of the MEMIC (Fig 3B). From there, GFP levels drop as cells become anoxic and necrotic (Fig 3B). Together, these results confirm that denser cultures produce steeper metabolic gradients. In subsequent experiments, we used intermediate cell densities (250-500 cells/μL) that create strong gradients but that are shallow enough to clearly distinguish well-nurtured, nutrient-deprived, and necrotic regions within the MEMIC (e.g. Fig 2C).
The formation of the ischemic gradient is a dynamic process. To establish how these gradients develop over time, we seeded the same engineered MDA-MB-231 cells into the MEMIC and tracked their fluorescence levels using live microscopy and image analysis. While constitutive membrane-mCherry levels remain constant (Video 3), GFP levels dramatically increase after about 12h in the distal regions of the MEMIC (Fig 3C, Video 3). Over time, the amplitude of the gradient increases and it becomes steeper. In a control culture where cells were well-nurtured throughout the well, no hypoxia was detectable throughout the entire experiment (Fig 3C).
We expected that the intensity and shape of metabolic gradients will also depend on the geometry of the MEMIC chamber. Using mathematical modeling we predicted that the height of the MEMIC is the critical geometric parameter in determining the position and slope of metabolic gradients (Carmona-Fontaine et al., 2013). This model also predicts that the width of the MEMIC’s opening should not affect the shape of the gradient. To test these predictions, we seeded a constant number of MDA-MB-231 cells expressing the 5xHRE/GFP hypoxia reporter in MEMICs with different heights or with different opening widths and we imaged them after 24 hours (5×105 cells, Fig. 3D,E). As predicted by our model, changing the width of the opening did not affect the shape of metabolite gradients. Cells seeded in MEMICs with different opening widths increased GFP expression at a constant distance from the opening. When we plot the GFP levels versus distance to opening, the signal from all experiments collapse into a single curve regardless of the opening width used (Fig 3D). While the width of the MEMIC opening has little physiological relevance, it has an important experimental implication: the exact position of the top coverslip does not matter as long as the edge of the glass is used as a reference point. This observation makes the fabrication of MEMICs easier and more reproducible.
As shown in figure Fig 3E, the height MEMIC chamber is indeed key in shaping ischemic gradients. Cells cultured in thin MEMIC chambers (short height) have less media, resulting in steeper gradients that peak close to the opening (Fig 3E). In taller chambers, ischemic gradients are shallower and peak further inside the chamber. The length scales in the MEMIC are much larger than in vivo where cells are tightly packed, and thus, their equivalent of the MEMIC height parameter is very small. As a reference point, tumor cells are completely hypoxic at about 150-200µm from the closest blood vessel (Carmona-Fontaine et al., 2017a; Hobson-Gutierrez and Carmona-Fontaine, 2018a; Thomlinson, 1977a). This distance is about an order of magnitude smaller than gradients in the MEMIC and spans only about 10-20 cell diameters. The shallower gradients in the MEMIC stretch spatial patterns to a few hundred cell diameters, greatly increasing the spatial resolution of our experiments. Despite this being an advantage for our experiments, it is important to keep in mind these scale differences and adapt the MEMIC’s geometry if needed.
Effects of metabolic gradients on cell proliferation
Nutrients provide the biomass and energy that all cells require to proliferate, and tumor cells are no exception. Cells in perivascular, and other well-perfused tumor regions, proliferate actively, while cells within nutrient-deprived tumor regions often show lower growth rates, senescence, and they can ultimately die (Carmona-Fontaine et al., 2013; Gatenby and Gillies, 2008; Thomlinson, 1977b). We wanted to test whether we can reproduce these spatial patterns of cell proliferation in the MEMIC. We used cells expressing FUCCI (Fluorescent Ubiquitination-based Cell Cycle Indicator, Fig 4A), a dual fluorescent reporter that enables the visualization of cell cycle progression (Sakaue-Sawano et al., 2008). Using our image cytometry approach (Fig. 2E), we quantified fluorescence levels per nuclei. To avoid confounding effects due to aneuploidy and polyploidy that most tumor cells have, we used RPE1 cells – a nearly diploid immortalized epithelial human cell line with a relatively stable genome (Davoli and Lange, 2011; Watkins et al., 2020). RPE1 cells proliferate rapidly, as shown by the many cells in M phase in the no gradients control culture (Fig 4B). The same was true in well-nurtured regions of the MEMIC. However, ischemic cells showed a stark arrest at G1 denoted by high mKO2-CDT1 and low mAG-GMNN levels (Fig 4C). A similar arrest at G1 has been reported in tumors in vivo (Yano et al., 2014). De novo nucleotide synthesis is very sensitive to nutrient limitation (Ducker and Rabinowitz, 2016), suggesting that ischemic cells in the MEMIC – and in tumors– may arrest in G1 as they fail to synthetize enough nucleotides to duplicate their genomes. We detected a similar spatial pattern of cell proliferation in tumor cells using immunofluorescence. Lung adenocarcinoma cells proliferate rapidly as evidenced by a high number of nuclei positive for phosphorylated Histone H3 (p-H3). As shown in Fig 4D, these numbers decline rapidly in ischemic regions of the MEMIC (Fig 4D, E). Overall, these results show that metabolic gradients in the MEMIC induce changes in cell proliferation that are consistent with nutrient availability and in vivo observations. More broadly, these proof-of-principle data show the important role that extracellular metabolites play in producing spatial patterns of cell behavior.
Untangling the effects of different metabolites in the MEMIC
The fast proliferation rate of tumor cells requires exogenous sources of multiple nutrients including glucose (Gatenby and Gillies, 2004; Heiden et al., 2009), aspartate (Birsoy et al., 2015; Sullivan et al., 2015), non-essential amino acids such as glutamine (Son et al., 2013), glycine (Jain et al., 2012), serine (Maddocks et al., 2012; Possemato et al., 2011), and asparagine (Halbrook et al., 2020; Krall et al., 2020; Pavlova et al., 2018), as well as diet-derived essential amino acids, vitamins, and lipids (Bose et al., 2020). Ischemic tumor cells in vivo experience reduced levels of oxygen and all these numerous nutrients – and the accumulation of secreted metabolic byproducts such as lactate. Modeling this complex environment is difficult, and thus, most cell metabolism studies focus on limiting a single substrate.
In the MEMIC cells spontaneously form complex metabolic gradients, better approximating in vivo conditions. While these more physiological conditions can be advantageous, they also can also challenge the experimental uncoupling of individual metabolites and their hierarchical relationships. Here we provide an experimental approach to tackle this challenge. First, in the MEMIC we can easily distinguish effects that require hypoxia from changes produced by nutrient scarcity. To do this separation, we take advantage of the gas permeability of polydimethylsiloxane (PDMS (Cox and Dunn, 1986)). PDMS is a biocompatible and transparent polymer that we shape into membranes the size of the conventional glass coverslips that we use in the MEMIC (Fig. S1). Cells in MEMICs sealed with a PDMS membrane – instead of a glass coverslip – will experience gradients of nutrients but no hypoxia as oxygen is rapidly equilibrated across the PDMS membrane (Fig. 5A). To illustrate this approach, we cultured cells expressing the 5xHRE/GFP hypoxia reporter and we grew them in MEMIC constructed with glass, PDMS, or no coverslips (Fig 5A). As shown in Figure 5, only cells in a MEMIC sealed with glass showed GFP signal gradients (Fig 5C). To test if hypoxia is sufficient to produce a particular effect, these MEMIC experiments can be complemented with conventional in vitro cultures in hypoxic incubators where oxygen is set at specific tensions. Similarly, to determine the effects of specific soluble nutrients we can simply remove them from the media formulation, or we can add them at saturating concentrations. For example, we demonstrate that a decrease in media pH-directly induced cell death by increasing the buffer capacity of culture media (Carmona-Fontaine et al., 2013). Overall, this section illustrates that cells in the MEMIC create complex metabolic gradients while still allowing to untangle, identify, and manipulate specific microenvironmental factors.
Ischemic macrophages decrease epithelial features in neighboring tumor cells
The complexity of the microenvironments within the MEMIC can be increased as needed. For instance, we can incorporate additional cell types to approximate the in vivo microenvironment and to study cell-cell interactions under different metabolic conditions. Extracellular metabolites can directly modulate cell phenotypes and tumor-immune interactions (Buck et al., 2017; Olenchock et al., 2017b).
We and others have previously shown that tumor-secreted lactate and low oxygen tension induce a pro-angiogenic phenotype in tumor-associated macrophages (Carmona-Fontaine et al., 2017a; Colegio et al., 2014; Wenes et al., 2016). Here we focused on how ischemic macrophages affect neighboring tumor cells. We first observed that virtually every ischemic tumor cell cultured in the MEMIC became more dysplastic, lose their multicellular epithelial arrangements, and adopt a round or spiked morphology. These effects were exacerbated when tumor cells were co-cultured with Bone Marrow-Derived Macrophages (BMDMs). We thus hypothesized that ischemia and the presence of macrophages synergize to alter the morphology of tumor cells.
To formally test this idea, we used clones derived from MMTV-PyMT breast tumors. In culture, these cells form geometric epithelial structures and display high levels of E-Cadherin and other epithelial markers (Fig 6A,B). Consistent with the known effects of hypoxia, we observed a significant decrease in E-Cadherin levels in ischemic cells (Fig 6A). We compared these results with similar experiments where we co-cultured the same tumor cells with BMDMs, ensuring we maintained the same total cell number. In these co-cultures, ischemic tumor cells dramatically reduced their epithelial morphology and the drop in E-Cadherin was much stronger and it occurred closer to the opening (Fig 6B,C). These morphological and molecular changes suggest that tumor cells are undergoing epithelial to mesenchymal transition (EMT) – a key step in the development of most carcinomas which represent ∼80% of all solid tumors, (Weinberg, 2013). Consistent with our observations, HIF1α directly promotes EMT (Dongre and Weinberg, 2018; Nieto, 2013) and tumor-associated macrophages can also drive EMT in tumors (Condeelis and Pollard, 2006; Linde et al., 2018). Our results suggest that ischemia and signals from ischemic macrophages synergize to trigger EMT in tumor cells (Fig 6D). While the molecular underpinnings of this synergy still need to be elucidated, these results illustrate how the MEMIC can be used to study the interplay between different cellular and molecular factors in the tumor microenvironment.
DISCUSSION
In vivo animal tumor models are – and will continue to be – crucial in cancer research. However, they are expensive and difficult to scale up. Additionally, it is virtually impossible to accurately control and measure multiple experimental parameters in these experiments. At the other extreme of the experimental spectrum, conventional in vitro cell cultures allow the control of most experimental parameters, but this simplicity comes at the cost of losing the context of the intricacy and heterogeneity of tumors. The microphysiological culture system described here – named MEMIC – seeks to bridge these extremes by re-creating key aspects of the complexity of tumors under tunable and controllable conditions.
Cells in the MEMIC form concentration gradients of oxygen and nutrients equivalent to those formed in tumors in vivo. These different levels of nutrients induce changes in major signaling pathways that in turn change cell behaviors and phenotypes in predictable spatial patterns. As an example, we showed how concentration gradients of nutrients in the MEMIC lead to a heterogeneous microenvironment that ranges from well-nurtured conditions – where cells rapidly proliferate – to increasingly more ischemic environments where cells show cell cycle arrest. This heterogeneous distribution of resources and the resulting regions with high and low cell proliferation are well documented in the analysis of histological tumor samples from patients (Weinberg, 2013).
The simplicity and affordability of the MEMIC makes it compatible with mid- to high-throughput experiments and screens. Most chemotherapy drugs target rapidly proliferating cells, senescent and non-proliferating tumor cells are spared and leading to lower tumor regression and relapse (Gordon and Nelson, 2012; Oshima et al., 2020). Thus the MEMIC could be particularly useful for screening drugs that more efficiently target ischemic regions of tumors, where non-proliferating cells are abundant.
Since nutrient levels change gradually with distance from their source, cellular changes in the MEMIC follow predictable spatial patterns. We and others have previously shown that similar spatial patterns emerge in vivo following the distribution of the vascular network (Carmona-Fontaine et al., 2017b; Gatenby and Gillies, 2008; Hobson-Gutierrez and Carmona-Fontaine, 2018b; Thomlinson, 1977b). The idea that nutrient and oxygen availability influence cell behavior and the tumor microenvironment is not new. However, this effect has been hard to model experimentally. With the combined power of in vivo observation and experiments in the MEMIC, we can argue that metabolite gradients do not only change cell phenotypes, but they organize and pattern the tumor microenvironment.
The spatial patterns that nutrient gradients impose in the tumor microenvironment also modulate cell-cell interactions. We showed here that ischemic macrophages reduce epithelial features in neighboring tumor cells. This inhibition of epithelial features is much stronger than the direct effects of nutrient deprivation alone. We thus propose that signals from ischemic macrophages synergize with the effects of hypoxia and nutrient deprivation to induce a stronger EMT in tumor cells (Fig. 6D). Overall, we think that the MEMIC provides a platform to model complex multicellular and heterogeneous in vivo conditions. In addition to having most of the advantages of conventional in vitro cultures, experiments in the MEMIC can be designed to incorporate high molecular and cellular complexity. This complexity can be increased gradually to carefully tease apart key factors shaping the tumor microenvironment. We believe that the process of deconstructing tumors into their basic units, and then carefully reassembling them in the MEMIC, will help us to better understand and control the tumor microenvironment.
MATERIALS AND METHODS
Cell culture
C6-HRE-GFP were a generous gift from Dr. Inna Serganova (Memorial Sloan Kettering Cancer Center). PyMT cells were derived by Daniela Quail (McGill University). MDA-MB-231 (HTB-26), HEK293T, and DLD-1 (CCL-221) cells were purchased from ATCC. C6-HRE-GFP cells, MDA-MB-231 cells, DLD-1 cells, LN-99, and HEK293T cells were grown in High Glucose DMEM (Gibco,11965-092) supplemented with 10% Fetal Bovine Serum (FBS; Sigma-Aldrich, F0926). RPE cells were grown in High Glucose DMEM-F12 medium (Gibco,11039) containing 10% FBS. All cells were incubated at 5% CO2 and 37°C.
Animal protocols and differentiation of BMDMs
Bone marrow-derived macrophages (BMDMs) were extracted from C57BL/6 mice using standard protocols (Gocheva et al., 2010). Following euthanization, femurs and tibiae were harvested from both legs of the mice under sterile conditions. After careful cleaning of the tissue, the bone marrow was flushed using a 25-gauge needle and passed through a 40 μm strainer. The bone marrow flush was cultured in low attachment culture dishes (VWR, 25384-342) in High Glucose DMEM supplemented with 10% FBS, 10 ng/mL Recombinant Mouse CSF-1 (R&D Systems, 416-ML), and 1x Antibiotic-Antimycotic (Gibco, 15240) for 7 days. Media was replaced every other day to induce macrophage differentiation. All protocols involving animal work were approved by NYU’s University Animal Welfare Committee (UAWC protocols 17-1496 and 19-1515).
Construction of genetic fluorescent reporters
To create a hypoxia reporter for live imaging, an existing GFP-based HIF1α reporter (Vordermark et al., 2001), Addgene, 46926) was subcloned into a lentiviral delivery plasmid using a Gibson assembly-based modular assembly platform (GMAP) (Akama-Garren et al., 2016). The 5xHRE-GFP region and a PGK-driven puromycin selection cassette from the pMSCV-Peredox-mCherry-NLS plasmid (Addgene, 32385) were amplified using primers containing overhangs with the homology sites for GMAP cloning and inserted into a lentiviral vector (LV 1-5, Addgene, 68411). This lentiviral backbone was a gift from Dr. Tyler Jacks (Massachusetts Institute of Technology) and Dr. Thales Papagiannakopoulos (New York University). The DNA construct for this new hypoxia reporter is available in Addgene.
To create a fluorescent membrane label, we used the CAAX sequence method to target a fluorescent protein to the plasma membrane (Hancock et al., 1989). mCherry was PCR amplified from pMSCV-Peredox-mCherry-NLS (Addgene, 32385) using primers containing overhangs with the homology sites for GMAP cloning. The reverse primer inserted the CAAX polybasic sequence. The CMV promoter and enhancer sequence to drive CAAX-mCherry expression was amplified from pLenti-CMV-Puro-DEST plasmid (Addgene, 17452). The two fragments were then assembled into one using overlapping PCR and inserted into LV 1-5 using the GMAP method. The DNA construct for this new fluorescent membrane reporter is available in Addgene.
Both resulting constructs were delivered to destination cells through lentiviral transduction. Briefly, lentiviruses were packaged in HEK293T cells through co-transfection with 2μg of VSV-G, 4μg of Delta8.9 viral vectors and 6μg of 5xHRE-GFP or mCherry-CAAX vector using Lipofectamine 3000 reagent (Thermo Scientific, L3000) in Opti-MEM (Gibco, 31985). The vectors mentioned above were kindly provided by Dr. Nicholas P. Gauthier (Dana Farber Cancer Institute). On day 2, Opti-MEM media was replaced with fresh DMEM supplemented with FBS. On days 3 and 5 after transfection, the virus-containing medium was collected, filtered through 0.45μm filters, and immediately stored at −80°C until infection. For lentiviral infection, 1mL of freshly thawed virus media mixed with 1mL of culture media along with 10μg/mL polybrene was plated onto destination cells in a 6-well plate at 40-60% confluency. For the 5xHRE-GFP infection, infected cells were selected using 1μg/mL puromycin for 4 days starting 48 hours after transduction. For the mCherry-CAAX infection, infected cells were selected using fluorescence-activated cell sorting. To create the cell line with both mCherry-CAAX and 5xHRE-GFP, we used the lentiviral particles containing the mCherry-CAAX insert to infect a 5xHRE-GFP cell line and then selected for using both selection methods.
Printing and Manufacturing the MEMIC
The main framework of the MEMIC is 3D printed using fused filament fabrication, which allows for on-demand supply as well as easy design modifications. For a detailed, step-by-step MEMIC fabrication protocol, please refer to the supplementary information attached to this manuscript and to supplementary videos S1 and S2. We designed the framework using CAD software and printed them using Ultimaker 3D printers and black PLA filament (Ultimaker) – a biocompatible, non-autofluorescent, and biodegradable material. Other printers and other materials may be used as well, depending on printer availability and experimental design. The printer settings, as well as the 3D design, that we use can be found in the supplementary information of this paper. We showcase a fused-filament-fabricated MEMIC version in this paper, as we believe that the combination of the widening availability of fused filament printers in research facilities and their precision would make this MEMIC the best option for most users. We also describe other options in the supplementary information.
Once a 3D-printed framework was retrieved from the printer, glass coverslips (No. 1, 18×18 mm; VWR) were glued to it using a UV-curable adhesive (NOA68, Norland Products) to create 12 individual chambers. We provide a detailed video on how to use the NOA68 and how to glue coverslips onto the MEMIC (Video S2). To create the bottom of the chambers, the coverslips were glued so that one completely covers each opening. The glue must completely seal the glass surrounding the opening to prevent leakage from the bottom of the chamber but should not seep into the surface of the chamber where cells will be seeded, as this limits space for cell adhesion. Once the adhesive was carefully applied, it was cured for 15 minutes under a long-wave UV lamp (DR-5401, MelodySusie). To complete construction of the inner chamber, glass coverslips were glued to the inside of each well with one edge against the inner ridge. This created the top layer of the gradient chamber and an opening to the media reservoir. Some chambers were left open without a top glass layer so that no gradients form, which are used as controls. The top layer adhesive was cured for 15 minutes under a long-wave UV lamp. For a fully detailed protocol on manufacturing and illustrative images, please see the supplementary information.
To decouple the nutrient and hypoxia components of the ischemic gradient in the MEMIC, in some experiments the glass coverslip was replaced with PDMS coverslips. The PDMS coverslips were prepared using the Dow Sylgard 184 Silicone Encapsulant Kit. To prepare PDMS coverslips, 1 part of the curing agent was thoroughly mixed with 10 parts of the silicone elastomer base by weight and centrifuged at 1300 RCF for 2-5 minutes to remove any air bubbles. A petri dish was then coated with about 6mL of the mixture by gently swirling it. To remove trapped air bubbles, the dish was placed into a vacuum desiccator for approximately 5 minutes. Curing was done by placing the PDMS coated dish into an oven at 75°C for at least 2 hours. The layer of PDMS was then cut into 18×18mm squares, matching the dimensions of the glass coverslip. Since PDMS is hydrophobic and may pose problems while gluing to the MEMIC framework, the PDMS coverslips were treated with a plasma gun. Then, the PDMS coverslips were glued with NOA68 optical adhesive as described for the top glass coverslips above.
Seeding cells into the MEMIC
To prepare the MEMIC for cell culture, the framework was first sterilized in a short-wave UV lamp (CM-2009, Meishida). Then, the inner chambers and wells of the MEMIC were washed twice with sterile DPBS (Gibco, 14040117). To improve cell attachment, the chambers were filled with a 50μg/mL solution of Poly-D-Lysine (Sigma-Aldrich, P6407) in DPBS and incubated at 37°C overnight, then washed twice with DPBS. Afterwards, the chambers were washed with culture media. Cells to be seeded were detached from their dish using 0.05% Trypsin-EDTA (Gibco, 25300-062). A cell suspension containing 106 cells/mL was prepared and the chambers were filled with 100μL of the suspension. To open wells serving as controls, 500μL of the cell suspension and 2.5mL fresh media was added. As the glass portion occupies about a fifth of the total surface of the well bottom, the resulting cell density on the glass portion should be similar to the closed wells. To allow cell attachment, the framework was placed into an incubator for approximately an hour. Once cells settled to the bottom of the chamber, the reservoir of the closed chambers was filled with 3mL fresh media. The framework was placed back into the incubator to start gradient formation.
Immunofluorescence
After 24 hours of gradient formation, the media was aspirated from the wells and then carefully aspirated from the chambers using a pipette to prevent cell loss. The cells were fixed by filling the chambers with 2% (v/v) paraformaldehyde solution, obtained by diluting 4% paraformaldehyde (Affymetrix, 19943) 1:1 in PBS (Fisher Scientific, BP399) and incubating at room temperature for 5 minutes. The chambers were then gently washed twice with PBS. To permeabilize the cells for staining, the chambers were filled with a 0.1% (v/v) solution of Triton-X (Sigma-Aldrich, T8787) in PBS and incubated at room temperature for 5 minutes. The chambers were gently washed twice with PBS and filled with 100μL of blocking solution - 2.5% (w/v) bovine serum albumin (Sigma-Aldrich, A9418) in PBS - and incubated for 1 hour at room temperature. The chambers were gently washed twice with PBS. A primary antibody dilution was prepared in the blocking solution with the antibodies diluted 1:200 for HIF1α (BD Bioscience, 610958), 1:50 for phosphor-S6 (CST, 2211), 1:200 for E-cadherin (BD Bioscience, 610182), 1:200 for phospho-H3 (CST, 3377), and 1:200 for CD68 (Serotec, MCA1957) and was added to the chambers. The framework was placed inside a humidified chamber at 4°C. The primary antibody solution was removed. The chambers were washed 3 times with PBS and incubated at room temperature for 5 minutes. The chambers were then filled with a secondary antibody (1:500 dilution) and nuclear stain solution (Hoechst 33342, 1:2000) and incubated at room temperature for 1 hour in the dark. The chambers were finally washed with PBS three times. To preserve cells until imaging, we filled the chambers with a 0.02% (w/v) solution of sodium azide (Sigma-Aldrich, S2002) in PBS.
Microscopy
We imaged MEMICs using an inverted fluorescent microscope (Eclipse Ti2-E, Nikon) equipped with a spinning disk confocal unit (X-Light V2, CrestOptics). For live imaging, we used a stage top incubator (Tokai Hit, INUCG2-KRi) fed with a mixed gas source (5% CO2 in compressed air, Airgas, X02AI95C2000117). Entire chambers were imaged in different fluorescent channels using tiling microscopy and then exported as stitched 16-bit TIF files.
Image Cytometry Analysis
Microscopy images were processed using our MATLAB-based scripts. A full tutorial on how to use our image cytometry code for users without image analysis experience is available with this paper (see SI for further details). This analytic approach can be used with MEMIC images, and also with any other fluorescent images of cells or cell clusters where fluorescence intensity and/or spatial information are of interest. Briefly, the software intakes images of multiple replicates or conditions, with a single TIFF image per channel per chamber, and outputs data on per cell fluorescence intensity, cell area on the image, and cell position on the image. It also produces plots of fluorescence intensity versus distance for each channel. The image analysis process requires that at least one channel displays a uniform stain across, such as a nuclear stain for a cell monolayer.
ADDITIONAL INFORMATION
Supplementary information is available for this paper. Please visit http://carmofon.org/publications/newmemic/ to download updated designs of the MEMIC and other information relevant to this work. Correspondence and requests for materials should be addressed to C.C.-F.
ACKNOWLEDGEMENTS
We thank all members of the Carmofon Laboratory for feedback and comments on the manuscript. This work was supported by awards from the National Cancer Institute of NIH to C.C.-F. (R00 CA191021 and DP2 CA250005). C.C.-F. is a Pew Biomedical Fellow and additional support for this work was provided by the The Pew Charitable Trust (00034121) and the Center for Genomics & Systems Biology at New York University.