Fabricating hierarchical, multiplexed DNA patterns using photolithography

DNA-instructed assembly provides a simple and robust solution to coordinate multiple signaling components by capitalizing on the specificity and strong, rapid binding kinetics of Watson-Crick base pairing (39, 40). The use of programmable DNA provides unparalleled multiplexing capabilities given the vast sequence space of 20-bp DNA oligonucleotides (i.e., 4²⁰ unique sequences). While a previous effort relied on a serial, cantilever-based system with low-patterning throughput and spatial control, our new parallel approach markedly advances the patterning capabilities of DNA-directed assembly and, in turn, the ability to recapitulate complex, heterogeneous signaling environments (41). Specifically, we use patterned photoresist as a physical template to guide the conjugation of unique oligonucleotide strands. The photoresist template can iteratively be stripped and a new one defined to conjugate another set of unique amine-terminated oligonucleotide strands, remarkably without disrupting the previously patterned oligonucleotides. Complex, hierarchical patterns are thus easily assembled. Although our method may appear to be serial because of repeated photolithographic steps to define each DNA component (i.e., layer-by-layer), patterning occurs simultaneously over large areas, and therefore, scaling from a small subset of, for instance, tens of features to millions requires no additional time. Furthermore, given the advanced state of lithography instrumentation in which automated substrate handling and fiducial recognition are commonplace, the need to conduct multiple lithography rounds to fabricate multiplexed DNA patterns is easily achievable.

There are two key steps for achieving this multicomponent, DNA-patterned platform (Fig. 1B). The first involves traditional photolithography, where patterned photoresist serves to (i) expose selective areas of surface aldehyde groups for DNA conjugation, (ii) act as a physical barrier to prevent conjugation to unexposed aldehyde groups, and (iii) preserve protected, unconjugated aldehyde functionality for subsequent, multilayered DNA patterning steps. The second step covalently immobilizes 20-bp oligonucleotides to the glass substrate by reacting the primary amine group at the 5′ end of the DNA with the surface-exposed aldehyde groups. See Materials and Methods for detailed protocol.

Our approach of ultraviolet (UV)–patterning photoresist offers the distinct advantage of defining, with great control and precision, complex spatial patterns across different length scales (i.e., from micrometers to millimeters and, in turn, from subcellular to bulk population) and over large areas (up to thousands of square millimeters) within minutes. High-resolution surface DNA patterns can be generated and visualized by hybridizing a complementary, fluorescent oligonucleotide (Fig. 1C). Unlike direct-write technologies, in which the active patterning step can be time intensive, the essential patterning steps of our platform focuses on a simple 1-min or less UV exposure step and subsequent DNA conjugation. In-depth characterization of DNA-patterning steps has revealed that optimization of both the buffer composition and the combination of condensation time and temperature play critical roles in achieving robust, uniform surface DNA patterns (fig. S1). An additional key advantage of our method is the tunable control that we have over DNA concentrations patterned onto the substrate. For example, by varying the concentration of the oligonucleotide solution dropcast over the photoresist patterns, we achieved a >100-fold range of fluorescent intensities (Fig. 1D). Having such precise control enables investigations of biological events in which variations in signal concentration, such as morphogen gradients during development, occur (42).

To demonstrate the functional utility of our platform, we first used microfabricated DNA patterns to organize an oligonucleotide-labeled (43) NSC population with high spatial precision, demonstrating the potential to recreate complex cell-based tissue structures (Fig. 2A). Next, to determine the minimum DNA concentration necessary for patterning single cells, we tested cell capture efficiencies of 20-μm-diameter spot arrays over a concentration range from 0.5 to 100 μM. A minimum of 5 μM was necessary to capture at least one oligonucleotide-labeled NSC per spot (Fig. 2B). Above this concentration, >90% of patterned spots were occupied. While DNA concentration can influence cell capture, so too can microfabricated pattern feature size. Increasing the diameter of DNA-patterned circle features resulted in a robust and reproducible increase in the number of cells captured per spot (Fig. 2C). Moreover, when the diameter was commensurate with the size of NSCs (~15 μm), single-cell capture was achieved. As we demonstrate later, this capability enables high-throughput clonal analysis.

To fabricate multicomponent DNA patterns, we found that the aforementioned two-step process can be repeated after dissolving the patterned photoresist in acetone. A new photoresist layer can then be applied to define a new spatial pattern that, in turn, guides the conjugation of additional oligonucleotide strands. To validate the robustness and reproducibility of this iterative process, we first demonstrated that the application of three additional photoresist layers does not adversely affect the first DNA-patterned layer’s functionality (i.e., ability to hybridize) (Fig. 3A). Rather, the application of new photoresist layers preserves the integrity of the previously patterned DNA layers while also allowing for the selective exposure of additional aldehyde regions for multicomponent conjugation (fig. S2). Second, we demonstrated that the actual photolithographic steps—particularly, (i) resist baking at 100°C, (ii) resist removal with acetone (an aggressive solvent), and (iii) resist development with a highly alkaline solution (pH ~13 and highly corrosive)—do not compromise the aldehyde groups on the glass substrate, as DNA patterns fabricated from subsequent layers retained high-intensity fluorescent values (Fig. 3B). Extensive characterization established that multilayer patterning can be remarkably extended to at least 10 layers without loss of fidelity (fig. S2C). Moreover, tunability of patterned DNA concentration can be achieved for multiplexed DNA patterns (fig. S3).

While photolithography imparts high spatial control in defining each DNA layer, it also enables tight spatial coordination between layers. Figure 3C(a) and Fig. 3C(b) highlight the successful registration of three complex DNA patterns and the robust functionality of multiplexed surface DNA patterns, respectively. As a demonstration of the biological utility of this key capability, while also emphasizing the broad applicability of our platform to assemble cell types other than NSCs, we used four different cellular components to mimic different stages of early metastasis that could potentially occur in a breast cancer microenvironment (Fig. 3D). Specifically, we patterned malignant breast epithelial MCF-7 cells (representing the “primary tumor”) surrounded by nonmalignant breast epithelial MCF-10A cells (representing “normal tissue”). We also patterned human umbilical vein endothelial cells (HUVECs) within the MCF-10As to represent vasculature. Last, we assembled a cluster of invasive triple-negative breast cells (MDA-MB-231) at the periphery (white arrows). The size and spatial separation of the MDA-MB-231 cluster from the MCF-7s represent different potential scenarios of the initial stages of metastasis, in which a subset of cancer cells has lost their adhesion to the primary tumor site and, with their migratory and invasive properties, have escaped into ultimately the surrounding tissue. Our patterning method provides a foundation to investigate the many cellular components of the highly complex breast cancer microenvironment in a reductionist manner. Repeating arrays of the DNA-patterned co-cultures can be assembled simultaneously with different cell component ratios tuned with high spatial precision.

Multicomponent DNA patterns instruct the presentation of heterogeneous proteins

A key step for demonstrating the unique capabilities of our DNA-directed strategy is using the surface DNA patterns to control the spatial organization of solid-phase ligands. Having such control would enable one to recapitulate the presentation of both extracellular matrix (ECM)–sequestered and cell surface–tethered signals found in specific tissue niches. Our approach of using DNA as a programmable intermediary capture agent ensures that multiple ligands, each labeled with a different complementary oligonucleotide, can be assembled from a single mixed solution flowed across the DNA-patterned surface. To label ligands-of-interest with oligonucleotides, we used the heterobifunctional linker dibenzocyclooctyne (DBCO)–polyethyleneglycol (PEG₄)–maleimide because of the quick and reliable high reaction efficiencies that could be achieved (Fig. 4A). Briefly, ligands were designed to contain a free terminal cysteine to react with the maleimide group on the cross-linker, thereby introducing a DBCO moiety on the ligand that allows for subsequent click chemistry reaction with an azide-modified oligonucleotide label (see Materials and Methods). As a proof of concept, we conjugated an oligonucleotide to recombinant enhanced green fluorescent protein (eGFP). Surface DNA patterns directed the spatial organization of eGFP and successfully maintained robust protein patterns over long-term cell culture (fig. S4A). In addition, neither eGFP lacking an oligonucleotide label nor eGFP containing a noncomplementary label resulted in protein capture (fig. S4B), indicating that complementary DNA sequences were necessary to achieve high specificity of eGFP patterns. To visualize and quantify the relative concentrations of patterned ligands, we included a Cy5 dye at the 3′ end of the oligonucleotide label (Fig. 4B). Last, to highlight that spatial control could be extended to multiple solid-phase cues, we conjugated a second oligo strand to mCherry and demonstrated tunable patterns of mCherry and eGFP (Fig. 4C).

Applying multiplexed DNA patterns to dissect NSC fate competition

As a first biological demonstration of our method’s unique capabilities, we focused on modeling complex signaling scenarios within the adult NSC niche. Stem cell niches are canonical examples of specialized microenvironments that coordinate the behavior (i.e., quiescence, migration, lineage commitment, etc.) of residing stem cells in response to physiological or pathological directives (44–46). Stem cells must continuously decide whether to self-renew or to differentiate into specialized, mature progeny, and the complex balance between these two competing fate choices ensures that a stem cell population can maintain homeostasis and respond to organismal needs. Here, we investigated how adult hippocampal NSCs resolve a competition between opposing fate cues and to what extent spatial organization of signals offers biophysical context to inform this decision. We leveraged our technique’s spatial control over both cells and solid-phase cues to model solid-phase ligand competition between FGF-2 and ephrin-B2 at the single-cell level. While FGF-2 operates as both a soluble (36) and ECM-sequestered cue (47) to promote proliferation and stem cell maintenance, ephrin-B2 is a key signal presented on the surface of neighboring astrocytes that drives neuronal differentiation via physical cell-cell contact, signaling through the EphB4 receptor (ephrin type-B receptor 4) on NSCs (37). We exposed thousands of patterned, spatially segregated, single-NSC cultures to distinct spatial organizations of the competitive ligands and conducted time-lapse experiments to study and correlate the dynamics within these cultures with endpoint cell fate.

To interrogate single-cell behavior with our DNA-based platform, we microfabricated large arrays of 15-μm-diameter DNA spots to direct the capture of oligonucleotide-labeled single NSCs. We then lithographically patterned a cell-resistive, nonbiofouling material, polyacrylamide (PA), to define recurring arrays of 50 × 50 “microislands” (each 141 μm by 141 μm) (fig. S5A) (48). The combination of DNA and PA patterns provided the high-throughput power to organize and track thousands of single-NSC cultures (each culture confined to a single microisland) in parallel over a 5-day differentiation period. To recapitulate the NSC niche ECM while also enabling cell attachment, we introduced laminin and incubated it within the microislands following single-cell patterning (fig. S5B) (see Materials and Methods). Within our DNA-directed, single-cell microisland arrays, NSCs demonstrated expected proliferation and differentiation responses to soluble FGF-2 and differentiation medium, respectively, suggesting that lipid-DNA labeling had negligible biological effects on NSC behavior (fig. S6). This was validated further through in-depth characterization of NSC viability (fig. S7) and fitness (fig. S8), highlighting our platform’s compatibility with a more sensitive cell type (i.e., stem cells) and potential utility for a wide variety of other cell applications—from examining juxtacrine versus paracrine heterogeneous cell-cell interactions at the single-cell level (fig. S9) to recapitulating tissue-like structures (fig. S10).

Having established that photolithographically-defined DNA and PA patterns can support the high-throughput study of thousands of single-NSC cultures over the course of differentiation, we then prepared both niche cues, FGF-2 and ephrin-B2, by labeling each ligand with unique oligonucleotides containing a Cy3 and Cy5 fluorescent dye, respectively, for visualization (fig. S11). Because of the challenge of producing recombinant ephrin-B2 in high yield, we replaced the full-length protein with a mimetic peptide, TNYLFSPNGPIARAW, that exhibits nanomolar binding affinity to its cognate EphB4 receptor (49). When presented as a monomeric, soluble cue, the peptide was insufficient in inducing strong neuronal differentiation (fig. S12A). However, upon surface immobilizing via DNA hybridization, the peptide induced receptor clustering (fig. S12B) and resulted in Tuj1⁺ differentiation in single NSCs (fig. S11D). In contrast to FGF-2, which, in solid-phase form, acted as a potent activator of proliferation, a minimum concentration threshold was necessary to promote peptide-induced Tuj1⁺ differentiation. These results not only highlighted ligand activity across a range of concentrations but also informed our subsequent biological investigation by identifying appropriate patterning conditions to model ligand competition.

Upon validating that FGF-2 and the ephrin-B2 mimetic peptide individually drive divergent cell fates in single NSCs, we used our DNA-based method to assemble and model scenarios in which NSCs are presented with both conflicting cues, thereby emulating the expression of both of these signals by hippocampal astrocytes contacting NSCs (37, 50). By growing the complexity of our in vitro models to better mimic heterogeneous in vivo niches, we are striving to obtain a fundamental understanding of the complex relationship between proliferation and differentiation, a necessary step in realizing NSC-based therapies. We capitalized upon our system’s robust spatial control over both ligands and cells to modulate ligand presentation within single-cell microisland cultures. Two DNA-patterning strategies were thus implemented in parallel (Fig. 5A). The first constrained single NSCs to the center of either an FGF-2 or ephrin-B2 peptide circular region with the second ligand patterned around the microisland periphery. The second strategy positioned single NSCs at the interface between two, half/half ligand patterns with equal access to both solid-phase cues. Thousands of microislands encompassing all three spatial arrangements were assembled simultaneously to ensure sufficient statistical power, as shown in Fig. 5A (right).

Microislands were imaged over the course of a 4-day time lapse (movie S1) to track cell body distributions across the ligand-patterned regions. Time-lapse snapshots of (i) FGF-2–center, (ii) ephrin-B2–center, and (iii) half/half microislands (left) along with their corresponding day 5 immunostaining results (right) are provided in Fig. 5B. Custom computational analysis (fig. S13) (51) provided the capability to map out the dynamic cell-ligand interactions of each microisland culture and track how the initial parent NSC and its subsequent progeny distributed themselves over time in response to the organization of these competing niche cues (Fig. 5C).

We first investigated how spatial modulation of FGF-2 and ephrin-B2 “domains” shaped the interactions of the single-NSC cultures with these two competing niche ligands. For the FGF-2–center microislands, we observed high-average cell occupancy within the FGF-2 domain that persisted over all 4 days [Fig. 6A(a)]. In contrast, ephrin-B2–center microislands exhibited a much wider distribution of average cell occupancies in the ephrin-B2 region on the first day alone [Fig. 6A(b)], and, by the second day, most of the cells no longer substantially occupied ephrin-B2. This observation was further corroborated upon analyzing the half/half microislands, where single NSCs had the freedom to “choose” either protein-patterned region [Fig. 6A(c)]. Here, average cell occupancy within FGF-2 remained far greater than within ephrin-B2 or at the interface of both domains. The overall strong spatial bias toward FGF-2–patterned regions in all cases may have been driven by the ephrin-B2 mimetic peptide acting as a repulsive cue, which would align with previous findings of Eph/ephrin guiding tissue-boundary formation and axon extension through cell repulsion (52).

Having observed that NSCs prefer FGF-2 over ephrin-B2 surfaces, an observation that would not have otherwise been achieved given the limitations of other patterning techniques, we then immunostained clonal cultures to investigate whether this cell occupancy bias translated to cell fate decisions. A comparison of the three different ligand spatial presentations to FGF-2– and ephrin-B2–only microislands (Fig. 6B) revealed that one niche signal unexpectedly did not exclusively dominate over the other with regard to cell fate. The three different ligand-competition scenarios resulted in proliferation rates that were significantly higher than the ephrin-B2–only condition yet significantly lower than the FGF-2–only microislands. With regard to differentiation, we anticipated that the NSC’s preference to reside in the FGF-2 domains would result in a more proliferative, stem-like state. However, no significant differences in proliferation rate or Tuj1⁺ differentiation were detected between the three spatial organizations. Moreover, an unexpectedly wide distribution of neuronal differentiation proportions was observed, including microisland subpopulations spanning both extremes of 100 and 0% Tuj1⁺ neuronal differentiation. The observed mix of proliferation and differentiation (fig. S14A) strongly indicates that the cells were integrating both signals and that the added presence of either cue is insufficient to instruct or completely alter cell fate decisions, despite NSC’s spatial preference toward occupying FGF-2. The simultaneous sensing of both cues highlights the inherent complexity and heterogeneity of NSC behavior as survival, proliferation, and differentiation are balanced to yield a desired number of new neurons while maintaining a stem cell reservoir. Further in-depth single-cell analysis is required to identify specific contributing factors that give rise to this heterogeneity, and our platform uniquely provides the resolution to do so.

We next dissected the dynamics of individual microislands by tracking the changes in average cell body occupancy within the FGF-2–patterned region over time and subsequently grouped microislands according to end fate (fig. S14B). However, rather than observing a minimum and/or maximum occupancy threshold that predicts endpoint cell fate, microislands with similar FGF-2 occupancy patterns were spread across all three neuronal differentiation categories [“Low (0%),” “Medium (0 to 100%),” and “High (100%)”]. More remarkably, we identified an unexpected subset of single-NSC cultures that underwent differentiation despite having nearly 100% FGF-2 occupancy. A closer examination of these specific cultures (Fig. 6C) revealed a potential source for this paradox: Neurites were extending across into the ephrin-B2 region, such that cells could potentially sense and probe both protein patterns throughout differentiation. While it currently remains unclear whether transient sampling of the ephrin-B2 differentiation cue by a short neurite is sufficient to drive a long-term, cell fate decision, future work focused on tracking neurite dynamics and analyzing temporal aspects of cell occupancy within each ligand-patterned region could elucidate further how single NSCs sense and integrate conflicting instructive cues.