The extraordinary growth of the electronics industry over the past four decades has been fueled by the scaling of transistor features defined by lithographic patterning. In recent years, semiconductor lithography techniques have begun to see use in biological and medical applications ranging from DNA microarrays1,2
and protein chips3,4
to devices for measuring cellular mechanics.5-8
By and large, most of these devices comprise features in the micron scale – approximately where semiconductor technology was in the mid-1980's. Current state-of-theart lithographic processes are capable of achieving ~ 30 nm resolution.9
Among the candidates for future generation lithography technologies is nanoimprint lithography (NIL),10,11
which is a high throughput patterning technique in which a pattern is formed in a thin polymer film that has been cast on a substrate by molding it to a relief image in a rigid template (mask). The pattern is then transferred from the polymer by a variety of thin film deposition and/or etching techniques. There is no theoretical limitation to the resolution of the features imprinted by NIL;12
the practical limit is determined by the size of the features on the NIL template, which is typically patterned by electron beam lithography. We have recently developed a process based on NIL and self-aligned pattern transfer which reduces the imprinted feature size and is capable of creating metallic structures below 5 nm.13
We have also developed a facile surface chemistry which allows us to functionalize these structures with a broad array of biomolecular species with a high degree of selectivity.14
Using these techniques, we have fabricated biomimetic surfaces upon which we can control the precise placement of individual biomolecules. We report here how these surfaces can be used to study the role of geometric organization of extracellular matrix (ECM) binding ligands in controlling cell adhesion and spreading.
Cells interact with the extracelluar matrix (ECM) via integrins, which are transmembrane receptors linking the ECM to macromolecular complexes that bind to the actin cytoskeleton, forming adhesive contacts.15
Integrin-mediated interactions with matrix control cell adhesion and migration and play a central role in many developmental and functional processes in multicellular organisms, including differentiation, wound healing and metastasis.16,17
They are also important for drug design and tissue engineering. Adhesive contacts are stabilized by the binding of talin to integrin cytoplasmic domains, followed by reinforcement and activation of other proteins such as paxillin, vinculin and α-actinin.18
Different integrins bind to different ECM ligands, although the RGD (Arg-Gly-Asp) peptide sequence has been recognized as the main adhesive recognition site within the cell-binding region of fibronectin.16,19,20
Establishing the minimal requirements for adhesion has been a goal of researchers for many years. Several groups have studied cell spreading and focal contact formation as a function of matrix molecule density.21-27
Using this approach, Massia and Hubbell found that a minimum of 6 RGD ligands per micron (corresponding to a spacing of 440 nm) were sufficient to support cell spreading.25
More recent experiments, in which the spacing of RGD ligands could be precisely controlled by micellar diblock copolymer self-assembly,28,29
indicate that cell spreading and viability are negatively affected when the spacing between integrin binding sites exceeds ~ 60 -70 nm,28,30
which is approximately the distance between talin1 globular head binding domains.31
A similar spatial dependence has been observed for attachment and differentiation of neurons 32
as well as for controlling apoptosis.33
Likewise, force-dependent measurements have shown that reliable ECM-cytoskeleton bonds could be formed only with multiple fibronectin 7-10 domains clustered within 40 – 60 nm.34,35
In addition, cell adhesion and migration is enhanced by nanoscale clustering of RGD domains.36-39
An important question raised by these studies is whether or not there exists a nanometer-scale adhesive unit, defined in terms of distance and number, that supports spreading.
To determine how ligand spacing, density and number modulate ECM-cytoskeleton interactions, we used NIL to fabricate biochips () in which these factors were varied independently. Our NIL process offer significant advantages over the type of self-assembly techniques used previously in cell spreading assays,28,30,32,33
due to its ability to form arbitrary, heterogeneous patterns of any possible geometry with a nearly unlimited number of degrees of freedom. Thus, each biochip contained several different patterns to test the effects of a variety of parameters on cell behavior simultaneously, minimizing sample-to-sample variability and making experimental results more reliable. Cell spreading assays confirmed that beyond the importance of integrin spacing, integrin cluster size was critical, independent of cluster density.
Figure 1 a Schematic view of a biochip containing arrays of sub-10 nm functionalized nanodots, arranged in dimers, trimers and extended hexagons with various inter-dot spacings. Each pattern extends over an area of 200 microns. b, Schematic fabrication process (more ...)
Fabrication of the nanoscale bioarray chips in this work is shown schematically in . Briefly, NIL templates, either from diamond-like carbon (DLC)40-42
or hydrogen silsesquioxane (HSQ) on silicon,43
were patterned by electron beam lithography. Pattern transfer to glass cover-slips was done by NIL into a poly-methyl methacrylate (PMMA) film, followed by residual PMMA removal and metal evaporation through an angle-evaporated hard mask,44
lift-off and thermal annealing to obtain spherical13
with diameters as small as 5 nm or less.13
show high resolution SEM images of such patterns. The AuPd nanodot patterns were functionalized14
with a mixed monolayer of thiolated polyethylene glycol (PEG) and thiolated PEG-biotin,46
with the surrounding glass passivated against nonspecific protein binding by a PEG-silane, followed by attachment of streptavidin to the biotin groups (). This facilitates the binding of a broad range of molecular species. We confirmed this selective binding using a double-stranded (ds) DNA oligomer to which biotin was attached at the 5′ end. show fluorescent images of an array, functionalized first with Alexa-fluor488-labeled streptavidin, and subsequently with Cy3-labeled dsDNA on the same pattern, indicating the high fidelity of the process.
Figure 2 a Biofunctionalization scheme of AuPd nanodots. b. Fluorescence imaging of a dimer array functionalized with Alexa-fluor 488-labeled neutravidin. c. Fluorescence imaging of a the same array after the conjugation of Cy3-labelled dsDNA ( 570 nm) to neutravidin. (more ...)
For our cell spreading assays, the nanodots were functionalized with biotinylated cyclic RGDfK, known for its affinity to αV
through the streptavidin bridge. The unpatterned glass surface was passivated by a self-assembled monolayer of PEG in order to minimize non-specific binding of proteins to the surface.49
While more than one streptavidin molecule (~ 4 - 5 nm50
) may bind to a single nanodot (and more than one biotinylated RGDfk can bind to each streptavidin), each dot can accommodate only a single integrin, since the integrin head is ~ 8 – 12 nm,51-54
providing arrays of single integrin molecule binding sites arranged in geometries defined by NIL.
All the arrays were 200 μm × 200 μm in size, providing sufficient area for the spreading of 20 - 40 3T3 fibroblast cells used in this study. In addition to the nanodots arrays, each chip contained a large area covered with a planar AuPd film with a saturated density of binding ligands. This served as a reference point for cell adhesion and spreading.
An initial test of the effect of spacing on cell spreading was done using nanodot arrays of dimers and trimers (, and Fig. S2
), in which the inter-dot spacing was varied in the range of 50 nm to 100 nm from array to array, while the spacing between the dimers (or trimers) in each array was kept constant to maintain the same total density of binding sites for all arrays. In addition, extended hexagonal arrays (), in which the inter-dot spacing and the total RGD density were varied in a dependent fashion, were used for direct comparison to work done on arrays formed by micellar diblock copolymer self-assembly.28,30,55
NIH 3T3 mouse fibroblast cells were plated on the arrays, and their behavior in terms of motility and spreading was monitored simultaneously on different patterned areas. In general, the cell behavior on dimer and trimer arrays was characterized by high motility and poor adhesion and spreading. ( show the cells on arrays of dimers, trimers, and hexagons, all with an inter-dot spacing of 60 nm, 1.5 hours after plating.) Many of the cells migrated from the dimer and trimer patterned areas to marking scribes on the glass a few tens of microns away from the edges of the arrays (). In contrast, cells plated on the arrays of extended hexagons with inter-dot spacings 60 nm and below exhibited a high degree of adhesion and spreading (), and the cells spread only within the 200 μm squares containing the nanodots. Spreading efficiency curves, which measure the percentage of spread cells as a function of time, were plotted for each array. A cell was determined to be spread56
if (a) the cell transformed from a rounded shape to a clearly visible flat shape with microspikes and filopodia adhering to the surface, and with possible further increase in cell area; and
(b) the cell did not return to its original unspread form at any time during the experiment, which typically ran ~ 1.5 - 3 hours. show spreading efficiency curves for arrays with various dot arrangements but with the same spacing between the neighboring dots, 50 nm and 100 nm, respectively. Spreading curves for planar AuPd are included for comparison. All the curves have a sigmoidal form, with a maximal value usually achieved in less than 1 hour. While the fastest and most complete (~ 100%) spreading occurred, as expected, on the planar AuPd, the extended hexagonal array with 50 nm inter-dot spacing appeared to provide nearly as good an environment for cell adhesion and spreading. On the other hand, the dimer and trimer arrays with the same inter-dot spacing (50 nm) exhibited a much lower percentage of spread cells, ~ 20 - 30 %, at the end of experiment. For extended hexagonal arrays with inter-dot spacings of 80 nm and 100 nm, cell behavior was similar to that observed on arrays of dimers and trimers ( and in Fig. S5
). The surface area of the cells spread on extended hexagonal arrays with interdot spacing ≤ 60 nm was, on average, twice as high as the other arrays (Fig. S6
). Because the density of RGD in the dimer and trimer patterns was much lower than in the extended hexagonal arrays, it was possible that density and not spacing was critical in determining the cell spreading efficiency for these specific types of pattern.
Figure 3 a-c. 3T3 cells spread on arrays of dimers, trimers and extended hexagons respectively, with b = 60 nm. Area enclosed by boxes is 200 μm × 200 μm. d. Cell spreading efficiency for different arrays with b = 50 nm. e. Cell spreading (more ...)
In order to distinguish more clearly between the roles of density and spacing in integrin binding, we designed a different pattern in which the density was allowed to vary, but the inter-dot spacing was kept constant. The pattern consisted of small heptagonal clusters with a constant inter-dot spacing of 60 nm. The distance, a
, between clusters was varied in order to achieve a range of global densities (). As shown in , we found virtually no variation in cell spreading efficiency with available binding site density over a range from 216 dots/μm2
() down to 51 dots/μm2
(), which corresponds to the density of binding sites in the dimer configuration that had shown poor spreading. Thus, the overall density of dots was not critical, indicating that perhaps the cluster size was playing a key role in the observed cell spreading efficiency. Such a concept has been previously proposed.34,35,38,57-59
In fact, an analytical model developed by Irvine et al.58
predicts strong ligand spatial distribution effects with a saturation in cluster size in the range of 3 - 4. Until now, however, it has been difficult to create a biomimetic environment in which the configuration of individual liganded binding sites could be controlled with high precision. Indeed, Arnold et al.,60
used electron beam lithography to “subtract” areas from hexagonal close packed arrays formed by block copolymer micelle lithography to create clusters as small as 6 integrin binding sites; however they presented no data for smaller clusters.
Figure 4 a-d. SEM of arrays of nanodots, arranged in heptagonal clusters. e,f. 3T3 cells spread on arrays of heptagonal clusters with a = 180 nm ( global density of 216 dots/mm2) and a = 370 nm ( global density of 51 dots/mm2), respectively. g. Average values (more ...)
To determine experimentally the minimum cluster size that supports spreading, we patterned arrays of clusters of 2 – 7 liganded binding sites (). The spacing between the dots in the clusters was set at 60 nm for all the arrays, and the cluster configuration was designed to be a fragment of an extended hexagonal array. The distance, a, between the clusters was varied in order to maintain a constant global density of 50 dots/μm2 in every array, i.e., a very low density of heptagonal dot clusters (7 clusters/μm2, ). When cells were plated on chips containing all these arrays, a striking increase in spreading efficiency was observed between cluster sizes of 3 and 4 (). The average of multiple experiments, , confirmed that there was a dramatic difference between cluster sizes of 3 or less and 4 or more at a spacing of 60 nm. show SEM images at the edge of a fixed and dried cell. Adhesive contacts can clearly be seen at the sites of the nanodot clusters.
Figure 5 a-d. SEM of arrays of nanodots, arranged in clusters of 4, 5, 6, and 7, respectively. e. An example of spreading curves obtained from a biochip with arrays of clusters of 2 -7 dots. f. Average values of spreading efficiency for the tested arrays with (more ...)
The resolution and patterning versatility of electron beam and nanoimprint lithography facilitate the production of many different matrix arrays, thereby enabling the determination of a basic adhesion unit that can support cell spreading and adhesion. Our results suggest that such a cell spreading unit involves clustering of at least 4 liganded integrins within ~ 60 nm. Furthermore, there is no evidence that cluster size influences adhesion above the threshold of 4 (). The spacing between the clusters in these experiments, which was nearly 400 nm in the case of the heptagonal arrays, suggests that they are most likely inducing independent multimolecular complexes in the cells that support the spreading process. Integrin-ligand clusters would naturally have greater adhesion strength than would single molecules. High (nN) contractile forces are generated on the ECM-integrin-cytoskeleton linkages even at early times in the spreading process that could dissociate integrin-RGD bonds.61
It is therefore reasonable to suggest that an increase in adhesion strength should correlate with an increase in spreading. Roca-Cusachs et al.35
found that clustering can increase individual ligand adhesion strength by 7-fold, either by the recruitment of a stabilizing protein complex on the cytoplasmic end or by increasing lateral integrin interactions. In either case, clustering would reduce diffusion of integrins and their ligands after bond breakage during the spreading process, thereby promoting their reattachment and enhancing adhesion. Clustering is an important part of the adhesion and spreading process, and studies of talin-depleted cells have shown that longer term spreading requires talin binding.62
Notably, talin is a dimeric, integrin-binding protein that has 4 potential integrin binding sites.31
Thus, we suggest that there is a discrete multimolecular adhesion complex involved in spreading and adhesion that requires the juxtaposition of at least 4 RGD-liganded integrins within 60 nm, and involves talin.
In earlier studies with constructs of fibronectin type-III domains formed by biochemical assembly into dimers, trimers and pentamers with 20 - 30 nm spacers,57
the minimal cluster size for protein edge binding and bead reinforcement was found to be a trimer.34
Our results on immobilized ligands, on the other hand, point to a minimum cluster of 4. In contrast to previous approaches, the unprecedented spatial control we have achieved over the organization of the integrin-binding RGD domains allows us to unequivocally rule out aggregation effects that can affect the size of the minimal adhesion unit. It may be that the full fibronectin domains, along with the flexibility of the linkers, could enable the multimolecular complex formation with fewer ligands per cluster.
New nanofabrication techniques combined with site-specific surface biochemical functionalization such as the one we present here enable unprecedented control over the arrangement of individual integrin molecules. We have used this strategy to identify a cell spreading adhesive unit that involves the clustering of at least 4 liganded integrins within ~ 60 nm. This finding agrees well with the established role of the cytoskeleton protein talin in cell adhesion. This approach can be generalized to enable a wide range of new studies on cellular systems and on other biomolecular interactions at the single-molecule level. They can also be used to develop new strategies in customized tissue scaffolds, wound healing bandages and cell-type-specific diagnostic and therapeutic tools.