Fatima Ezahra
Chrit
,
Peiru
Li
,
Todd
Sulchek
and
Alexander
Alexeev
*
George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332, USA. E-mail: alexander.alexeev@me.gatech.edu
First published on 5th February 2024
Numerous applications in medical diagnostics, cell engineering therapy, and biotechnology require the identification and sorting of cells that express desired molecular surface markers. We developed a microfluidic method for high-throughput and label-free sorting of biological cells by their affinity of molecular surface markers to target ligands. Our approach consists of a microfluidic channel decorated with periodic skewed ridges and coated with adhesive molecules. The periodic ridges form gaps with the opposing channel wall that are smaller than the cell diameter, thereby ensuring cell contact with the adhesive surfaces. Using three-dimensional computer simulations, we examine trajectories of adhesive cells in the ridged microchannels. The simulations reveal that cell trajectories are sensitive to the cell adhesion strength. Thus, the differential cell trajectories can be leveraged for adhesion-based cell separation. We probe the effect of cell elasticity on the adhesion-based sorting and show that cell elasticity can be utilized to enhance the resolution of the sorting. Furthermore, we investigate how the microchannel ridge angle can be tuned to achieve an efficient adhesion-based sorting of cells with different compliance.
Labeling methods (for example: FACS and MACS) can be used for adhesion-based cell sorting and offer high purity with high enrichment. However, these techniques provide a binary output for the receptor expression and are not yet able to fractionate into multiple outputs for finer sensitivity to the molecule of interest. Another major drawback of these methods is the changes they induce by labeling tags or tag removal, such as the risk of tag-induced activation of sorted cells, which restrains their downstream use.5,6 To solve this issue, negative selection of target cells can be used. Yet, even this can contaminate therapy products with magnetic beads coated with foreign antibodies.
To overcome these difficulties, label-free approaches can be employed leveraging differences in cells adhesive properties as the mechanism of separation.7–9 As an example of cell adhesion sorting using micro/nano structures, Kwon et al.10 used patterned microchannels to separate and enrich human breast cancer cells from epithelial cells. Detachment assays were used to measure cells adhesion. Specific adhesion of cells using proteins, peptides, and other molecular mediators can be employed to capture desired cells. Antibody coated surfaces in microfluidic channels were demonstrated to be effective for capturing circulating endothelial progenitor cells (EPCs).11 Hashimoto, Kaji, and Nishizawa12 combined dielectrophoresis with adhesion-based microfluidics to selectively capture neutrophils from a mixture of leukocytes. Zhang et al.13 used a microchannel coated with basement membrane extract to separate heterogeneous breast cancer cell lines based on their adhesive capacities. Yang et al.14 achieved circulating tumor cell (CTCs) capture from blood cells using a TA-functionalized film. Negative selection is also an option for enriching rare target cells.15,16 A tandem microfluidic systems was developed to isolate in a high-throughput, label free manner CTCs that exhibit high compliance and low adhesion.17 Singh et al.18 were able to isolate fully reprogrammed induced human pluripotent stem cells (hPSCs) from heterogeneous reprogramming cultures by capturing target cells on a substrate in a microchannel under shear flow. Microfluidic assay was developed to separate genetically-related yeast strains based on adhesion strength enabling the rapid screening and fractionation of yeast based on adhesive properties.19 However, these methods are not continuous since they capture and arrest cells from the flow. Chang, Lee, and Liepmann20 used microstructured fluidic channels with array of pillars coated with E-selectin to capture and arrest cells from the free flow. Bose et al.21 demonstrated affinity flow fractionation where interactions with asymmetric molecular patterns laterally displace cells in a continuous manner. Similarly, Choi, Karp, and Karnik22 described a separation process called “deterministic cell rolling”, which separates cells based on transient molecular interactions with slanted ridges on the channel floor in a continuous process. However, these approaches are limited to low flow rates to maintain rolling interactions with the adhesive surface and to prevent hydrodynamic forces from detaching the cells without any separation.
A high throughput adhesion based sorting method was recently reported23 that consists of a microchannel decorated with diagonal ridges and coated with adhesion molecules. Flow cytometry data for cells collected at the microchannel outlets showed different cell distributions correlated with the differences in cell adhesiveness indicating sorting dependence on adhesion levels. A unique aspect of this approach is the ability to operate at high throughput. The experiments demonstrated flow of cells at rates up to 0.2 m s−1, which is over two orders of magnitude higher than previous methods. Results indicated that sorting did not activate the cells due to the brief interaction time of cells with adhesion ligands beneficial for downstream cell analysis. We note that ridged microfluidic channels have been previously used to sort cells and microparticles by size and elasticity.24–27
Computational modeling has been used to probe adhesive behavior of cells, especially in the context of cell rolling.28–30 These models are usually based on theoretical frameworks which describe adhesive interactions using chemical reaction kinetics and relate the kinetic rates to the applied force. Bell31 and Dembo et al.32 adapted this kinetic theory and proposed an exponential constitutive law between the dissociation rate and the force. Evans, Berk, and Leung33 and Evans and Ritchie34 employed a power law instead of the exponential law used in the Bell model.31 Jadhav, Eggleton, and Konstantopoulos35 developed a three-dimensional computational model to predict receptor-mediated rolling of deformable cells in shear flow. Hammer and Apte36 simulated the rolling and adhesion of selectin-mediated neutrophil using a kinetics framework. Khismatullin and Truskey37 used a modified Dembo model and evaluated the effect of cell deformability and viscoelasticity on receptor-mediated leukocyte adhesion to ligand coated surface. Interactions of compliant adhesive capsules and cells in patterned microfluidic channels have been studied using computer simulations with applications to the sorting and separation.38–41
In this work, we first examine the rolling of deformable cells represented by a fluid-filled elastic particle on a flat substrate patterned with diagonal adhesive stripes. We show that the cells follow different trajectories based on the adhesiveness and explain the sorting mechanism. We further show that this separation method is limited to low flow rates due to the hydrodynamic lift detaching cells from the substrate. We then demonstrate high-throughput sorting of adhesive cells in ridged microfluidic channels. The periodic diagonal ridges compress cells propelled by the flow, thereby ensuring cell contact with adhesive surfaces even at high flow rates. We systematically investigate the effects of cell adhesion and elasticity on cell trajectories in the ridged microchannel and demonstrate that cells with different adhesiveness follow different trajectories. The simulations reveal that cell elasticity can be used to enhance sorting of adhesive cells. Furthermore, we show that the resolution of the adhesion-based cell sorting can be increased by adjusting the angle of the microchannel ridges. Together these studies suggest ways to improve the sensitivity of separation to adhesion in the limit of transient interactions characterized by fast timescales.
LBM uses time integration of the discretized Boltzmann equation fi(r + ciΔt, t + Δt) = fi(r, t) + Δi[f(r, Δt)] for a distribution function fi(r, t) describing the mass density of fluid particles at a lattice node r and time t propagating along a lattice direction i with a constant velocity ci.46,47 Here, Δt is the time step. The collision operator Δi describes the change in fi due to collisions at lattice nodes. In three dimensions, we use a cubic lattice D3Q19 with 19 discrete velocities.
The compliant cell is modelled as a fluid-filled elastic membrane. LSM uses a triangulated network of harmonic springs that connect regularly spaced mass nodes on the cell surface to represent an elastic membrane.48,49 The stretching elastic energy associated with a node at a position ri is where rij = |ri − rj| and reqij are respectively the length and the equilibrium length of a spring between two nodes with positions ri and ri, and ks is the spring constant. Bending rigidity of cell membrane is included by considering the change of dihedral angle of adjacent pairs of triangles. The bending energy is given by , where kb is the bending spring constant, θi is the instantaneous angle between two adjacent triangles with a common edge i, and θeq is the equilibrium angle.
To model adhesion between the cell and microchannel surfaces, we use the Morse potential that has a repulsive and attractive components, depending on the separation distance. The Morse potential associated with two nodes separated by a distance r is given by , where De is the depth of the potential well, Ke is a parameter controlling the width of the potential, and re is the equilibrium distance. The force derived from the Morse potential is given by . Each LSM node on the particle surface can adhesively interact with nodes on the microchannel surfaces located within a cut-off distance rcut = 1.5re.
The fluid and solid models are coupled via the boundary conditions at the fluid–solid interface that utilize the interpolated bounce back method50,51 and the momentum exchange method.52 The computational methodology for fluid structure interactions has been previously extensively validated by simulating flows with different rigid and deformable particles including mesh independence study, and found good agreement with relevant theories and experiments.24,26,53–56
To characterize the flow within the channel, we define a channel Reynolds number , which represents the ratio between the fluid inertia and viscous forces. The Reynolds number is based on the channel height H, the fluid density ρ and viscosity μ, and the characteristic flow velocity in the channel U0 due to a pressure gradient ΔP/L, defined as . The compliant particle is characterized by a capillary number . Here, is the characteristic shear rate, and is the two-dimensional shear modulus of the particle elastic shell, where E is the particle Young's modulus, ts is the shell thickness, and ν is the Poisson's ratio. Thus, the capillary number represents the ratio between viscous stresses on the particle and elastic stiffness of the particle shell. The adhesion level is controlled by varying the dimensionless adhesion strength .
In our simulations the particles are initially positioned at the channel centerline. The particle trajectory is represented using dimensionless particle's center of mass coordinates Cx = cx/d and Cz = cz/d. The particle motion is quantified using the dimensionless particle displacement per ridge where cz1 and cz2 are the particle's center of mass z-coordinate at x = L/3 and x = L, respectively.
Fig. 2a presents the trajectories of the particle's center of mass in the x–z plane for particles with different adhesion strengths β and Ca = 3.5 × 10−3. As the adhesion strength β increases, the deflection δ increases. In Fig. 2b, we show a schematic illustration of the particle trajectory across an adhesive diagonal stripe and the main forces acting on the particle. Before encountering the stripe, the particle moves along the channel centerline due to a drag force FD. When the particle approaches the adhesive stripe, it is pulled towards the centerline of the stripe by an adhesive force FA. The adhesive force deflects the particle trajectory in the negative x direction. When the particle crosses the middle of the stripe, the adhesive force flips its direction and acts to keep the particle on the stripe to minimize the energy.39 The combined action of FA and FD propels the particle along the stripe until it reaches the back edge of the stripe where the action of the adhesive force diminishes. After that the particle moves along the flow direction due to the drag force FD. As a result, the particle experiences a net positive lateral displacement Δz with increasing adhesion.
Fig. 3a shows how the dimensionless particle displacement δ = Δz/d depends on the adhesion strength β for particles with different elasticity Ca. We find that δ increases with increasing β. Furthermore, δ increases with increasing Ca. Thus, softer particles experience larger lateral displacement as they cross the adhesive stripe, which can be explained by a larger contact area between softer particles and the adhesive stripe increasing the free energy change and, therefore, FA.39
Fig. 3b shows δ as function of β for different channel flow rates expressed in terms of the Reynolds number Re. We find that the deflection decreases as Re increases due to a greater contribution of the hydrodynamic drag force FD. Furthermore, the increase of Re above the values reported in Fig. 1b results in the particle failure to interact with the adhesive stripe due to an increased lift force translating the particle away from the substrate. Thus, the method while can be efficient in separating cells is limited to relatively low flow rates57,58 and, therefore, low cell throughput.
Fig. 4a shows the trajectories of the particle's center of mass in the x–z plane for particles with different adhesion strength β. The figure also includes a typical streamline in the fluid flow passing through the microchannel middle plane. We find that the particles follow different trajectories depending on the magnitude of the adhesion β. The less adhesive particles deflect more along the ridge, whereas more adhesive particles more closely follows the flow streamline. Thus, the adhesion facilitates the particle crossing the ridge and, as a result, the negative transverse z-displacement increases with increasing β. Furthermore, we find that when the adhesion strength β exceeds a critical value βcr, the particles are unable to cross the ridge as they get immobilized by the adhesive interaction with the ridge.
The motion of compliant adhesive particles in a ridged microchannel is governed by three main forces, as illustrated in Fig. 4b. The fluid drag force FD acts along the flow direction due to the relative velocity between the fluid and the particle. The elastic force FE arises due to the particle elastic deformation by the ridge and is directed away from the ridge. The adhesive force FA is proportional to the gradient of the adhesive energy and is directed towards the ridge. Thus, the elastic force FE and the adhesive force FA have opposing effect on the particle trajectory and effectively counteract each other. Indeed, when FA is weak, FE causes the particle to move along the ridge leading to particle cross-channel displacement in the positive z direction. On the other hand, when FA is comparable or exceeds FE, it cancels out the effect of FE on the particle trajectory and the particle follows the flow streamline. When FA is too strong, FD is unable to translate an adhered particle and the particle becomes immobilized by the ridge. We note that the idealized particle trajectory shown in Fig. 4b somewhat deviates from the trajectories obtained in the simulations reported in Fig. 4a. This can be related to the non-linearity associated with particle deformation by the ridge. The particle deformation gradually grows until approximately a half of the particle enters in the gap formed by the ridge. After that the elastic resistance due to particle deformation rapidly weakens and the particle can readily traverse the ridge under the action of the hydrodynamic and adhesion forces. Furthermore, diagonal ridges in the microchannel give rise to a circulatory secondary flow that transport fluid in the cross stream direction, thereby deflection particle trajectory in the positive z direction.24–26
In Fig. 5, we plot the particle displacement per ridge δ as function of the adhesion strength β for particles with different elasticity Ca. The figure shows that at higher adhesion levels with β > 0.05, δ only slightly changes with Ca indicating that the motion under the ridge is dominated by the adhesive interactions. However, for weaker adhesion the particle displacement increases as adhesion decreases. The difference in the displacement between particles with weak and high adhesiveness is the most significant for stiffer particles with low Ca. Thus, the sorting resolution increases as particle stiffness increases.
Fig. 5 Particle lateral displacement per ridge δ = Δz/d as a function of the adhesion strength β in a ridged microchannel with Re = 3, λ = 0.75, and ϕ = 45°. |
Fig. 6a and b compare the trajectories of particles with Ca = 3.5 × 10−3 and different adhesiveness in microfluidic channels with ridge angles ϕ = 30° and ϕ = 15°, respectively. Note that the two figures have different scales of the x-axes, so the ridge angles are not shown to scale. We find that the separation between particles with different β is greater using ridges with ϕ = 15° compared to ϕ = 30°. Furthermore, the trajectories of particles with β = 0.03 and β = 0.05 practically coincide when the ridge angle is ϕ = 30°, whereas for ϕ = 15°, the trajectories of such particles are well separated. Thus, the use of lower ridge angle not only increases the separation resolution, but also enables separation of particles with greater adhesiveness.
To explain the effect of ridge angle ϕ on the trajectories of adhesive particles, we consider how relative direction of forces acting on the particle change with ϕ (Fig. 4b). The elastic force FE and the adhesive force FA are oriented normal to the ridge. Thus the direction of FE and FA changes with ϕ, whereas their magnitudes are independent of ϕ. The drag force FD acts along the flow direction independently of ϕ. Since the component of FD acting normal to the ridge balances FE and FA, the magnitude of this component of FD does not change with ϕ. This requires that the magnitude of FD increases with decreasing ϕ, and therefore, the component of FD directed along the ridge increases for smaller ϕ. This, in turn, results in a greater translation of the particles along the ridge enhancing the trajectory separation for particles with different adhesiveness.
Fig. 7 presents the particle displacement per ridge δ in microchannels with ϕ = 15° and ϕ = 30° as a function of the adhesion strength β for particles with Ca = 2.5 × 10−3 and Ca = 3.5 × 10−3. Note that the displacement δ = 2 corresponds to the condition where particles deflect along the ridge all the way until they reach the channel sidewall. Furthermore, the dashed lines in Fig. 7 indicate the conditions where particle motion is arrested at the ridge due to the large adhesion force. For stiffer particles with Ca = 2.5 × 10−3, the microchannel with ϕ = 30° provides significant separation among nonadhesive particles and particles with the adhesiveness up to β = 0.05. The same particles do not separate in the microchannel with ϕ = 15°, where they are deflected by the ridge to the channel side wall. On the one hand, softer particles with Ca = 3.5 × 10−3 and different β display minor differences in δ in channels with ϕ = 30° and, therefore, cannot be separated in such channels. The use of ridges with ϕ = 15° significantly enhances the separation of softer particles. In fact, the channel with ϕ = 15° yields separation of softer particles with Ca = 3.5 × 10−3 comparable to the separation of stiffer particles with Ca = 2.5 × 10−3 in microchannels with ϕ = 30°. This result suggests that adjustment of the ridge angle can be used to achieve an efficient adhesion-based sorting of particles with specific stiffness.
Finally, we assess the impact of higher Re on adhesion-based particle separation in the ridged microchannels. We modify Re by changing the fluid viscosity, in which case Ca does not change with Re. Fig. 8a and b show the trajectories of particles with Ca = 2.5 × 10−3 and different adhesiveness in microfluidic channels with Re = 60 and ridge angles ϕ = 30° and ϕ = 15°, respectively. Note that the two figures are not shown to scale. We find that the separation between particles with different β occurs even at high Re that significantly exceeds the maximum Re that can be used in microchannels with patterned stripes (Fig. 1a). Similarly to lower Re (Fig. 6), ridges with smaller angle ϕ result in greater separation of adhesive particles. Thus, our simulations show that microfluidic channels with diagonal ridges can be effectively used for high-throughput sorting of adhesive particles.
Our simulations identified three modes of adhesive particle transport through the ridged microchannels. In the case of small ridge angle, cell rolling can occur along the diagonal ridges in which a threshold of adhesion eliminates cell sliding. In the case of high adhesion, cells can occlude the ridge leading to extended contact times and irreversible binding. Finally, cell trajectories can be modified by adhesive ridges in a manner that greater displacement differences were observed for stiffer particles with low Ca number.
Our results demonstrate that cell trajectories within ridged microchannels are sensitive to the adhesion strength, which can be used to continuously sort cells based on their adhesiveness by collecting cells at different microchannel outlets. The simulations reveal that particle elasticity is essential for the adhesion-based sorting in ridged microchannels and can be used to enhance the sorting resolution by counteracting the adhesion force on the cells. We also show that the ridge angle can be tuned to achieve efficient adhesion-based sorting for cells with different mechanical properties. Furthermore, our simulations for Re = 60 show that ridged microfluidic channels can be effectively used for high-throughput adhesion-based particle sorting.
It is further interesting to examine the role of adhesion type on cell-ridge interactions. In these studies, we considered interactions represented by the Morse potential in which bond strength develops instantaneously as in the case of electrostatic interactions. In the case of specific bond adhesion, which is contact time dependent such as described by the Bell model, additional parameter space emerges affecting the cell transport that can be explored.
This journal is © The Royal Society of Chemistry 2024 |