Engineered basement membranes: from in vivo considerations to cell-based assays

Improvements in the physiological relevance of cell-based assays have been enabled by the development of various interdisciplinary methods. However, due to their complexity, in vivo structures such as basement membranes (BMs), which regulate the phenotype of adherent cells, are still difficult to mimic in vitro. The reconstruction of a physiologically relevant BM is crucially important to develop cell-based assays with the capacity for drug screening and disease modelling. Here, we review the biophysical and biochemical properties of BMs in vivo and their interactions with neighbouring cells. We discuss the current methods used to mimic BM functions in cell-based assays according to the type of targeted applications. In doing so, we examine the advantages and limitations of each method as well as exploring approaches to improve the physiological relevance of engineered or cell-derived BMs in vitro.


I. Introduction
Since the invention of the petri dish in 1887 by Julius Richard Petri 1 , there have been many improvements in cell culture approaches in order to reconstruct biologically and physiologically relevant cell-based assays, using interdisciplinary methods. These improvements have been achieved by the development of new technologies such as: (i) stem cells with both mesenchymal stem cells (MSCs) 2 and induced pluripotent stem cells (iPSCs) 3 (ii) microfluidics to control the cellular phenotype with chemical gradients 4,5 or shear stresses 6,7 and (iii) biomaterials, which regulate the microenvironmental cues. 8,9 However, these cell-based assays often lack relevant cell-extracellular matrix (ECM) interactions. This is particularly critical with assays aiming to recapitulate endothelial/epithelial tissues where cell phenotypes are regulated by the ECM from adjacent basement membranes (BMs). 10 BMs are thin layers of specialized ECM proteins, which have various fundamental functions: (i) to provide a structural scaffold to support endothelial or epithelial cells, (ii) to act as a barrier between tissue compartments and (iii) to maintain neighbouring cell phenotypes. [11][12][13] These functions depend on the biophysical/biochemical properties of BMs, which change according to the localisation within the body. 14 In this review, we discuss the properties of in vivo BMs induced by the core BM components (type IV collagen, laminins, nidogens and heparan sulphate proteoglycans). [15][16][17] We highlight the impact of these properties on cellular phenotypes as well their modification due to disease. [18][19][20] We, subsequently, describe the different methods used to mimic the BM, mainly in coculture systems, according to their properties from polymer membranes to cell-derived ECM. [21][22][23][24][25] Finally, we conclude by comparing these methods according to the targeted applications for cell-based assays, namely drug screening and disease modelling. 26,27 II. In vivo basement membranes In vivo BMs are very complex supramolecular assemblies formed by the interactions between two predominant networks of ECM proteins, type IV collagen (formed by three α chains) and laminin (formed by α, β and γ chains). 11,12 These networks are cross-linked by nidogens and interact with the heparan sulphate proteoglycans (e.g. agrin and perlecan). 28 Moreover, BMs are connected to cell surface receptors such as integrins or dystroglycans. [29][30][31][32] The BM molecular assembly is illustrated in Figure 1. 12,13,33 Besides the major components shown in Figure 1, various other molecules from the ECM are also present within the BM but in lower amounts, which makes it difficult to isolate and identify them. However, much recent progress has been made in this regard thanks to new "omics" technologies (genomics and proteomics) allowing for the characterisation of the matrisome. [34][35][36] Proteomic studies have now reported the complexity of: (i) the renal glomerular ECM composition, where the presence of 144 structural and regulatory proteins has been revealed 37 , (ii) the ECM secreted by bone marrow-derived and adiposederived MSCs 38 , (iii) the ECM from decellularised rat liver 39 and others tissues. 40 This complexity also applies to BMs. 41 a) Biophysical properties An important biophysical characteristic of BMs is their thickness, which varies from 50 to 100nm according to their localisation, as well as their composition. 12,13 The mechanical stability of BMs is thought to be regulated by the type IV collagen network due to its unique structure and its abundance within BMs. This network is formed from a trimer of three α chains that bind together to form a triple helix and the trimers then assemble into a network. The collagenous domains in the triple helix have hydrogen bonds and electrostatic interactions between the key residues (mainly glycine, proline and hydroxyproline) and these are likely to stabilize the molecule. [42][43][44] The two terminal domains, 7S domain at the N-terminus and the non-collagenous NC1 domain at the Cterminus also stabilize the network. This is achieved by the presence of cysteine and lysine residues, which crosslink the chains through disulphide bonds and lysine/hydroxylysine interactions at the N-terminus 45 and by non-covalent bonds mediated by various ions 46,47 and cross-linking formed via post-translational modification and sulphilimine bonds at the C-terminus. 48 50 Moveover, these nonlinear properties are modified in peroxidasin knockout mice due to the reduction of sulphilimine cross-links. It is important to note that the stability of the type IV collagen network can be modified according to type IV collagen isoforms, particularly in the glomerular BM. 51 The independent laminin network also contributes to the BM stability because of its interactions with the neighbouring cells and its cross-linkage to the type IV collagen network with nidogen. Besides these two networks, the flexibility of the BM is also due the presence of perlecan, a heparan sulphate proteoglycan. This flexibility has been demonstrated to help the organ shaping in the drosophila. 52 BM biophysical properties regulate the cell phenotype through various cues: the BM topography and the BM mechanical properties (e.g. stiffness, shear-strain response) 53 , which depend on the BM structure. The BM topography is induced by the BM protein selfassembly, which generates a fibrous scaffold from type IV collagen and laminin networks.
This scaffold exhibits specific patterns in term of size and geometry at the nanometre scale. 54,55 As the previously described biophysical cues, these topographic features also depend on their localisation in the body and on the species as shown by ex vivo measurements. 53 Inspired by the topographic features found in native vascular endothelial BMs, Liliensiek and co-workers generated substrates with: (i) anisotropically ordered ridge and grove structures and (ii) isotropically ordered pores from 0.2µm to 2µm in size. 56 Using different human endothelial cell lines, they demonstrated that all cell lines exhibit a highly pronounced orientation and alignment on anisotropically ridges equal or above 0.8 µm in size. Moreover, one cell line (human umbilical vein endothelial cells) showed a decreased proliferation on the smallest topographic features. Along with this work, various reports have demonstrated that BM topographic features (size, geometry and isotropy) have a strong influence on the phenotypes of the overlying cells. [57][58][59] Besides the BM topography, the BM stiffness or, in other terms, its elastic modulus (i.e. Young's modulus), plays an important role by regulating gene expression in response to mechanical cues by virtue of the links between the cytoskeleton and the ECM components via adhesion receptors. 7,60 It has been demonstrated in vitro that cell fate is influenced by the elastic modulus of the substrate on to which the cells adhere. 8,61 For this reason, cell behaviour is likely to be modulated by the Young's modulus of a BM, which varies from 1kPa to 2.57MPa according to the BM localisation. 53,62 Furthermore, the ECM components forming BMs exhibit nonlinear mechanical properties and a shear modulus ~100Pa. 63 Two particular properties have been described in the literature: (i) a nonlinear stressstiffening observed on biological gels 64,65 and (ii) a nonlinear shear stress-softening observed on decellularised normal and fibrotic rat liver. 66 These examples of nonlinear behaviour are typical of biological gels and cannot be reproduced using artificial polymers such as polyacrylamide. Furthermore, it also has been demonstrated that the BM mechanical properties strongly influence not only cell migration 67 but also BM breaching through mechanical processes. 68,69 Significantly, diseases targeting BM components can destabilize the BM structure (Table 1) and thus alter BM biophysical properties. Figure 1: Structure of the in vivo basement membrane composed of two predominant networks of type IV collagen and laminin, respectively. These networks are cross-linked by nidogen and interact directly or through agrin and perlecan, with the cell receptors (e.g. integrins and dytroglycan).

b) Biochemical properties
Along with these biophysical features, the biochemical properties of the BM are essential for cellular regulation due to the BM components. These properties are involved not only in various functions regulating neighbouring cells, such as cell adhesion and cell phenotype, but also in the diffusion of macromolecules or ions from the external environment and BM remodelling. As illustrated in Figure 1, cell adhesion on BMs is controlled by different types of biochemical interactions between the major cell surface adhesion receptors (e.g. integrins, dystroglycans) and matrix ligands. These adhesion interactions can activate intra-cellular signalling pathways leading to altered cell motility or cell differentiation. 70 Furthermore, it has been shown that laminin-integrin and laminin-dystroglycan binding are required to polarize the epiblast epithelium during early embryogenesis. 32 These interactions between the cell adhesion receptors and the laminin network explain the important role of laminin in numerous signalling pathways (modulation of growth factors, migration and inflammation of immune cells) that change cell phenotype. 29 The role of the laminin α subunits was studied during C. elegans development, showing that mutated α subunits disrupt BMs and lead to ectopic adhesion complex formation with downstream effects on cell-cell adhesion and cell signaling. 71 Furthermore, cell phenotype is also affected by various endogenous factors as BMs act as a reservoir for growth factors and release them according to biochemical stresses. 17 These factors include fibroblast growth factor-2 (FGF-2), vascular endothelial growth factor (VEGF), platelet-derived growth factor (PDGF) and Von Willebrand factor. 17 If laminin exhibits endothelial growth factor (EGF)-like domains, these growth factors are mainly sequestered by the heparan sulphate proteoglycans (HSPGs). For example, perlecan is composed of 5 major domains with their own biochemical activities such as angiogenesis, growth factor and morphogen sequestration, lipid retention and cell surface binding. 72 Agrin is the dominant HSPG of the glomerular BM and plays an important role in neuromuscular junctions. Moreover, it contains domains that interact strongly with various growth factors. 37,73,74 Besides these GF-like domains, the HSPGs have also another function in the biochemical properties of BM due to their sulphated lateral chains. As a result of their negatively charged groups, they control the interaction with cationic proteins and positively charged ions (diffusion of molecules and ions through the BM).
The loss of these lateral chains can impair various biochemical functions of the HSPGs, the filtration properties of the BM and be linked to diseases. 75,76 Another important biochemical property arises from localized remodelling of the BM, which occurs during development, via controlled degradation to allow immune cells trafficking from the endothelium to the epithelium, and also during extravasation of metastatic cancer cells. 28,77,78 This disruption and remodelling are triggered by two families of metalloproteinases, which cleave the ECM components: (i) matrix metalloproteinase (MMP) and (ii) a disintegrin and metalloproteinase with thrombospondin motifs (ADAMTS). 79 Some fragments resulting from this degradation have been shown to have specific biochemical activities within the BM such as being antiangiogenic (e.g. endorepellin) 17,72 or with other organs through the blood flow. [80][81][82] Furthermore, these biochemical properties depend strongly on the BM protein isoforms present within a particular tissue. The presence of the incorrect isoform or defects in the isoform structure can alter the BM properties as detailed in the  Table 1: Effects of the mutations in the basement membrane core proteins due to diseases, on the basement membrane biophysical and biochemical properties.
Although the conditions mentioned in Table 1 are all rare diseases a range of therapies are currently in clinical use. In the case of Alport syndrome (mutations affecting type IV collagen 3, 4, 5 chains), angiotensin-converting-enzyme inhibitors are indicated from the point at which patients have persistent leakage of protein into the urine. 44 However, these drugs only slow the progression of the disease. In a recent review, Nystrom et al.
proposed the development of cell therapy to restore permanently the damaged BM proteins in order to move forward to clinical applications. 90 The efficiency of such treatments could be tested within in vitro models before being translated to in vivo.

III. In vitro basement membranes
As described above, in vivo BMs are highly complex. Initial attempts to mimic these structures, were based on the use of simple polymer membranes but the complexity of in vitro membranes has been increasing over recent years. [91][92][93][94] a) Polymer membranes Due to their simplicity to fabricate by virtue of injection or soft-lithography, polymer membranes are widely used to mimic the in vivo BM within cell-based assays, particularly to test the BM permeability of the blood-brain barrier 95,96 or the glomerulus to drugs. 97 Different polymers: (i) polycarbonate, (ii) polyester, (iii) polyethylene terephthalate (PET) or (iv) polytetrafluoroethylene (PTFE) are used to generate these membranes and different porosities are commercially available to suit the targeted applications. These membranes are attached to an insert in order to be compatible with cell culture well plates and to be used for permeability assay for drug screening as shown in Figure 2a. (1-10μm) 101 needs to be great enough to be handled easily during the fabrication/integration steps. However, the thickness used is considerably higher than the thickness of in vivo BMs, and this may inhibit communications between the cells cultured on either side of the membrane. Cell-cell communication is also affected by the polymer membrane topography, which is limited by the pores generated during the polymer membrane fabrication. The porosity limits the use of this device in migration assays, as the polymer membrane cannot be degraded by the trafficking cells such as lymphocytes. Concerning the stiffness of the porous membrane systems, they generally present a Young's modulus (2.3GPa) three orders of magnitude higher than the highest in vivo BM Young's modulus (~2.57MPa). 62 This is also likely to influence cell phenotype. flibroblasts on the polymer membrane in order to generate fibroblast-derived ECM coating before removing the fibroblasts. However, the decellularisation process seems to damage the secreted laminin network and so alters the biochemical properties of the coating.  modulus range (1kPa-6MPa) and a highly porous structure, which is not fibre-like. 112,113 These matrices are useful to promote cell adhesion and phenotype in tissue engineering for transplantation.
A polysaccharide matrix was also recently used by Tibbe et al. to generate a self-standing chitosan membrane within a microfluidic device in order to reproduce the basement membrane. 114 Here, they formed a chitosan membrane by playing on the pH according to properties. 127 They demonstrated not only that the composition of these gels especially the nidogen concentration, is very different in each gel according to the manufacturer process but also that the gel composition affects its topography and its Young's modulus.
Furthermore, they showed that the cell migration (human promyelocytic leukemia cell line HL-60) in the gel depends on its composition. BM protein gels like Matrigel are actually one of the best in reproducing the BM in cell-based assays as it is composed of only biomolecules but like the other methods described above it suffers from a couple of disadvantages. Firstly, due to its tumorigenic origin, its components and their concentration cannot reproduce accurately the desired BM and they display batch-to-batch variation. Secondly due to its physical state, it is difficult to integrate them in advanced cell-based assays such as organs-on-chip without polymer membranes. 10 To overcome this integration issue, Takezawa and co-authors proposed a novel scaffold of type I collagen vitrigel, which is maintained by a silk fibre network or a nylon membrane ring 128 , in order to reconstruct the BM of various organs such as the kidney 129 , the liver 130 and the eyes. 131 Toh et al. reconstructed a collagen barrier to mimic the basement membrane within a microfluidic device. 132 They studied the migration of breast cancer cells (MX-1) through collagen barrier although without further characterisation.  133 This membrane whose fabrication process is described in However, they did not investigate if this ECM blend exhibit a stress-stiffening property as type I collagen display one but not Matrigel. 63 To generate an Organ-on-Chip device, they peeled off their membrane from the PDMS stab to insert it within a microfluidic chamber.
They showed, using a permeability assay that the main difference between the ECM-made (ii) The membrane thickness (~20μm) is higher than those in vivo.
(iii) Due its thickness, the membrane needs to be very carefully peeled off from the PDMS stub and inserted within the microfluidic device. coculture). 136 The expression of various BM components clearly shows the importance of cross talk between cells for BM formation. However, the addition of this layer can change the BM biophysical properties and to date there is no proof of BM remodelling. The layerby-layer coculture has two major advantages: (i) its compatibility with bioprinting, which increases the complexity of the culture system by generating a specific pattern with various cell types [145][146][147][148] and (ii) the possibility to use cell sheet engineering to generate a complex sheet of cells thanks to a thermo-sensible polymer. 149,150 This cell sheet method can be used to integrate the cocultured cells in an advanced cell-based assay.
Nevertheless, its integration within a microfluidic device can be quite challenging in order to keep the structure of the cell layer and its sterility.
IV. Applications of in vitro basement membranes As we described in the section II, in vivo BMs are generated by a complex self-assembly of various biomolecules, creating a dynamic 3-D matrix. Due to the difficulty in mimicking this dynamic in vivo behaviour in in vitro systems, the current models only simulate some of the specific functions of the BM. The limitations of these in vitro BMs are mainly due to material issues. 10 In order to perform a given function related to the BM within an advanced coculture cell-based assay, a trade-off has to be made between the material needed for the function and the ease of integrating the material into the required system.
This trade-off will be discussed in this section, particular for coculture cell-based assays applied to drug screening and disease modelling.    97 They demonstrated that without cells the uncoated membrane is more permeable than a collagen-coated membrane and that GEnCs seems to provide the largest contribution to albumin retention. This is probably due to the fenestrae and glycocalyx formed by the endothelial cell. 151 They also investigated the effect of drugs targeting the podocytes or the GEnCs on albumin permeability. However, their system was restricted to a static analysis, as it did not integrate a microfluidic circuit. To improve the capability of these assays, two different research groups, one in the USA and the other in China developed microfluidic devices based on porous polymer membrane-mimicking BM. 26,100 Musah et al. demonstrated that their iPSC-derived podocytes cocultured with GEnCs enhanced the albumin retention compared to a culture of GEnCs only. 100 This result contradicts the findings in the static culture discussed above. The reason for this difference might be due to differences in the distribution of type IV collagen, which is mainly generated by GEnCs in static conditions and by podocytes when mechanical strain is applied. 100 Zhou et al. cultured rodent GEnCs and podocytes within a microfluidic device. They confirmed that the coculture improves the permeability resistance as assessed using labelled inulin, bovine serum albumin and IgG. Their data showed that the contribution of GEnCs to the permeability resistance is more important than of podocytes but in this system, there is no information concerning the BM components secreted by the cultured cells.
An explanation of these discrepancies in the literature might be due the stiffness of the polymer membrane mimicking the BM and show the importance of the choice of material.
For drug screening, in particular permeability assays, the model requires the generation of two different compartments separated by the BM-mimicking material. This separation depends on a material that can be easily handled and has a high mechanical strength. For these reasons, porous polymer membranes are widely employed in permeability assays.
Although yet to be used, electrospun scaffolds might be a good candidate for this type of assay due to their topography and the possibility of engineering their biochemical properties. Although their integration was quite limited, the best candidate is probably an ECM protein gel-based membrane due to their biochemical properties and composition.

b) Disease modelling
Various disease processes result in the modification of BM biophysical and/or biochemical properties. These modifications can be due to BM thickening, the modification of the BM microenvironment (e.g. hypertension, diabetes), BM invasion by cancer cells or the mutation of a BM component gene (see table 1). All these parameters make in vitro disease models strongly dependent not only on the materials used to mimic the BM but also on the coculture cell-based assay design. Diseases that are due to changes in the BM microenvironment are the easier to model, particularly in coculture cell-based assays integrating microfluidics. Zhou et al. developed a device to mimic hypertensive nephropathy, assuming that higher blood pressure induces higher blood flow rate. 26 They demonstrated that high flow rate (15µL/min) increases the protein permeability and damages the cell phenotype of both podocytes and GEnCs. Although these interesting results were confirmed by in vivo measurements, their device is limited by the nonphysiologically relevant properties of their BM-mimicking polycarbonate membrane.
Indeed, polycarbonate displays a Young's modulus of 2.5GPa 152 , which is three order of magnitude higher than that seen in vivo. Furthermore, they did not calculate the value of the shear stress applied to the cells. The BM extract used to coat the polycarbonate membrane, is from EHS tumour, which can contain irrelevant biochemical cues.
To study cancer progression from endothelial-mesenchymal transition (EMT) to the cancer cell adhesion, invasion and vascularisation, various in vitro cancer models have been developed. 77,78,140,153,154 Nevertheless, the generation of a good advanced cell-based assay to model cancer still remains challenging as it depends on both biophysical and biochemical properties of BMs. Kamm's group at the Massachusetts Institute of Technology, has been developing advanced cell-based assays to study cancer progression.
They demonstrated that cancer progression is affected by various biophysical and biochemical cues. [155][156][157][158] However, the relevance of their device is limited because of the use of type I collagen to reproduce the BM instead of type IV collagen and laminin.
To study the effect of BM components, Walter et al. demonstrated using a type IV collagencoated polyacrylamide gel that physical defects within the type IV collagen network triggers EMT, causes MMP-9 inhibition and changes BM stiffness due to the lack of type IV collagen degradation suggesting the beginning of a fibrosis like behaviour. 159 They also generated a type IV collagen-coated type I collagen gel and showed that cells were able to invade the type I collagen gel after undergoing through EMT due to the defects within the type IV collagen network. This work shows the importance of the BM biophysical properties in mechanotransduction/mechanoactivation for cancer research. The authors acknowledge the limitations of their model due to the lack of BM biochemical cues, especially those from the laminin network. However, they showed the importance of MMP inhibition/activation in such cancer models.
The major role of the proteinases suggests that models using polymer membranes or electrospun scaffolds are not a good choice, as they generally cannot be degraded by proteinases. Furthermore, another important consequence of the proteinase activity is the generation of peptidic fragments called matricryptins, which can exhibit various bioactivities either physiological or pathological. [160][161][162][163] To study these effects, the use of only ECM components is required to mimic the BM behaviour. For example, Horejs et al. Commercially available ECM components from BM extracts have their own limitations because they are specific isoforms of BM components. They can have different biochemical activities due to the generation of different fragments and the remodelling also depends on the surface properties of the material, which support the ECM components. 165,166 For this reason, the rational for the used ECM components mainly depends on the targeted objectives. In the case of in vitro model to mimic diseases due to BM component mutation, its generation is still very difficult because of the limited availability of these mutated components. The development of iPSC technology using cells from patients in order to generate specific BM components is a promising new method to model disease within cell-based assays.

V. Conclusions and future prospects
In vivo basement membranes are complex biological structures, which performs various functions by virtue of their specific components. Furthermore, they display a dynamic behaviour, which is not yet fully understood. For this reason, all the BM functions are not only difficult to recapitulate from both a biophysical and biochemical perspective, but are challenging to integrate in cell-based assays. These assays are generally limited by the low functionalities of the BM-mimicking materials that are used. Currently, the best BM mimics seem to be the functionalized electrospun scaffold and the extracellular matrix membrane, such as those introduced by Rossi et al. 24 and Mondrinos et al. 27 , respectively.
These both reproduce the biophysical properties of in vivo BMs, except the nonlinear mechanical properties. On the electrospun scaffold developed by Rossi and co-workers 24 , the biochemical properties could be tuned by changing the nature of the peptide used.
Furthermore, the relative robustness of the scaffolds make them easier to integrate within microfluidic assays. In contrast, extracellular matrix membranes are more difficult to integrate into devices due to their fragile nature, but their offer advantageous in applications such as BM invasion studies.
Future improvements of engineered BMs for coculture cell-based assays will mainly depend on the targeted applications. Applications, in which BM composition is important, such as genetic disorders of BM, will require the development of recombinant BM components and/or the generation of differentiated relevant cells from patient-derived iPSCs. Applications related to the blood pressure will need a BM-mimicking material, which exhibits nonlinear mechanical properties. For this purpose, the integration of hydrogels, which display stress-stiffening effect 167,168 , can be a good solution. However, their biochemical properties will have to be improved by integrating specific BM peptides, for example. Further models integrating both biophysical and biochemical BM properties will not only help to study the contributions of each BM components and their correlation but also to get a better understanding of the drug diffusion through BMs and BM-related diseases. This will enable the development of new drugs and/or the identification of new biomarkers.

Conflicts of interest
There are no conflicts of interest to declare.