FLUID DYNAMICS AND INERTIAL FOCUSING IN SPIRAL MICROCHANNELS FOR CELL SORTING
A dissertation submitted to the Graduate School of the University of Cincinnati in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
in the Department of Electrical Engineering and Computing Systems of the College of Engineering and Applied Science
2016
by Nivedita Nivedita
B.Tech., Mody University of Technology and Science, India, 2009
Committee Chair: Ian Papautsky, Ph.D.
ABSTRACT
Effective cell sorting is critical to sample preparation, especially in the fields of therapeutics and diagnostics. The advent of lab-on-chip systems has increased the demand for integrable, low cost, highly efficient and continuous cell sorters. Inertial microfluidic devices, especially spiral sorting channels, have the potential to address all of these requirements. These devices are usually designed on the assumption that there are two counter rotating vortices which balance the inertial lift forces to facilitate sorting. However, this assumption does not fully explain the shift in focusing of the cells or particles on increase in the flow rate. In fact, the fluid flow dynamics in spiral microchannels for high Reynolds number and high Dean number is not fully understood. In this work, for the first time we show experimentally spatial development of multiple Dean vortices across the microchannel cross-section with change in aspect ratio and
Dean number. These experiments lead to an explanation of the uncharacteristic focusing behavior of cells and particles in spiral microchannels at high Dean number. Based on this work, spiral devices were designed for specific sorting applications including plasma extraction, blood cell sorting, and isolation of sub-populations of cancer stem cells. The modified and integrable inertial spiral devices offer >90% sorting efficiency in the case of blood cell sorting, along with high throughput of >106 cells/min. Additionally, this work conclusively shows that effects of fluidic shear on cell viability, functionality and stem-ness are negligible when using these devices for sorting cancer stem cells.
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude towards a number of individuals and organizations for their support during the course of my dissertation research.
Foremost, I would like to thank my advisor, Dr. Ian Papautsky whose constant support, guidance and encouragement, led to the successful completion of this work. He provided a nurturing environment which supported both my professional and personal development. His expert guidance and organized approach to research motivated me to innovate and provided me with the skill set needed to advance my career in the interdisciplinary field of bio-microfluidics.
In addition, I would like to thank the members of my dissertation committee for their insight and suggestions. A special thanks to Dr. Susan Kasper for her constant encouragement and skillful guidance in the aspects of cell biology and cell culture. I would also like to thank Dr. Ligrani for his expert guidance in the field of fluid flow dynamics.
This work would not have been possible without the help and support of a number of other individuals. I would like to express my gratitude towards Jeff Simkins and Ron Flennikin, the staff of OCMI and ERC Clean room facility, for their support and patience, especially the training and technical expertise required for the fabrication of the devices required in this work.
For the training and help in cell culture and assays, I would like to thank Dr. P.
Vummidigiridhar, the former postdoctoral fellow in Dr. Kasper’s lab. Additionally, I would like to express my appreciation for the staff and secretaries at the Department of Electrical
Engineering and Computing Systems, for their helpful and congenial attitude. Many thanks to my fellow colleagues in BioMicroSystems Lab, who provided me with both technical and personal support whenever needed. A special thanks to Prithviraj Mukherjee for his friendship and support that only a brother can provide.
This work was financially supported by various sources. I would like to acknowledge the financial support of University of Cincinnati Research Council fellowship, NSF I/UCRC
CADMIM (Center for Advanced Design and Manufacturing of Integrated Microfluidics), and
DARPA N/MEMS S&T Fundamentals Program by SPAWAR to the Micro/nano Fluidics
Fundaments Focus (MF3) Center.
Finally and most importantly, I would like to thank my parents Shobha and Ranvir Singh
Antil for their love, support and encouragement. Your constant support has given me the strength to achieve my dreams.
TABLE OF CONTENTS
LIST OF FIGURES ...... viii
LIST OF SYMBOLS ...... xiii
CHAPTER 1 INTRODUCTION ...... 1
Microfluidics for cell sorting ...... 2 Inertial microfluidics ...... 6 Motivation ...... 9 Scope of work ...... 10 Chapter summary ...... 12
CHAPTER 2 FLOW IN SPIRAL MICROCHANNELS ...... 14
Introduction ...... 14 Methods...... 17 Evolution of primary Dean vortices ...... 21 Evolution of secondary Dean vortices ...... 28 Critical Dean number ...... 38 Effects on particle focusing ...... 39 Summary ...... 46
CHAPTER 3 PLASMA AND BLOOD CELL SORTING ...... 47
Introduction ...... 47 Device principle and optimization ...... 48 Experimental methods ...... 55 Blood dilution ...... 59 Blood component sorting ...... 62 Summary ...... 66
CHAPTER 4 ISOLATION OF PCA STEM CELL SUBPOPULATIONS ...... 68
Introduction ...... 68 Sorting protocol and methods ...... 70 Assessment of cell viability and proliferation after separation ...... 73 Assessment of cell functionality after separation ...... 77
vi Doublet and single-cell sorting of HPET cells ...... 83 Summary ...... 85
CHAPTER 5 CONCLUSIONS ...... 86
Summary ...... 86 Future work ...... 89
REFERENCES ...... 91
vii LIST OF FIGURES
Figure Page
1. Particle focusing due to inertial and Dean migration. (a) Focusing of particles in a circular tube. (b) Schematic showing the focusing of particles in 4 positions along the channel walls in a square micro-channel. (c) Schematic showing the stage 2 equilibrium position in a rectangular straight channel. (d) Single focusing position of particles/cells in a curved rectangular channel due to the presence of Dean vortices ...... 7
2. Dean flow dynamics at low and high De. (a) Schematic illustrating two counter rotating vortices in a curved rectangular channel at lower De. (b) Schematic illustrating flow behavior at high De causing the formation of multiple vortices ...... 16
3. Geometric configurations for fluid flow dynamic study. Schematic layout of the spiral device with the table listing cross-sectional dimensions of the three geometries and the respective radii of curvature (convex radius of curvature, R)...... 18
4. Soft-lithography process for device fabrication. (a) Schematic of the soft- lithography process of master fabrication and PDMS casting. (b) Image of the PerMX-3050 master used for device fabrication...... 19
5. Numerical simulation of a single loop of Archimedean spiral. (a) Top-view of the single loop of spiral model simulated using Star CCM+ at 1 m/s. (b) Parabolic velocity profile obtained at the straight channel near the inlet of the spiral loop (section-1). (c) Plot of the velocity profile near the middle of the loop (section-2) ...... 22
6. Simulated cross-sectional images of the scalar velocity profile and corresponding images of the Dean flow vectors. The single loop was simulated for De = 0 (straight channel) and De = 343.09 ...... 23
7. Effect of increase in De on velocity profile. (a) Plot of the shift of center of maximum velocity from the center of the channel (xs) relative to half width of the channel (W0.5) as De is increased. (b) The plot of the shift in the secondary flow regime ...... 24
8. Onset of instability as a function of De. (a) Image of the spiral device used for studying the flow behavior in spiral microchannels. (b) Fluorescent image of the inlet showing the three inlet system and the 1/3rd confinement of the dye. (c) Plot of the Angle of instability (θi) as a function of De and the three geometries g1, g2 and g3.The rate of change of the angle of instability defines the rate of the onset
viii of instability in the curved channel The amplitude difference, d, between the plots is dependent on the width of the channel. (d), (e) Fluorescent images taken by inverted microscope using fluorescein for contrast. The white arrows indicate the position where multiple streaks were observed indicating pinching of the dye between multiple vortices (the images were taken at the two loops-1 and 2). θi is the angle of instability that quantifies the migration of the dye towards the opposite wall ...... 26
9. Schematic of the 1/3rd confinement of dye used to visualize cross-sectional streamline development...... 29
10. Confocal images of the cross-section of the rectangular spiral microchannel with geometry g1. The images were taken at regular intervals over two loops. These images show a gradual development of secondary vortices in the second loop of the spiral device whereas in the first loop only primary vortices were observed...... 31
11. Confocal images of the cross-section of spiral microchannel with geometry g2. These images show visualization of cross-sectional flow as it develops with increase in De...... 33
12. Confocal images of the cross-section of rectangular spiral microchannel with geometry g3. These images show the development of cross-sectional flow as a function of increase in De...... 35
13. Plot of Area as a function of De for the rectangular spiral microchannel geometries: g1, g2 and g3. The plot in blue shows the area of the channel covered by one of the two primary Dean vortices and the plot in red shows the area covered by one of the two secondary Dean vortices ...... 37
14. Plot of the critical dean number as a function of the aspect ratio of the rectangular channel. This plot provides a threshold for operation in the flow regime with primary Dean vortices ...... 39
15. Effect of secondary Dean vortices on particle entrapment. (a) Schematic of the process of entrapment of particles in the additional vortices with the insets of the corresponding particle focusing positions and the behavior of RBCs. (b) Intensity plot of the focused stream of 10 µm diameter particles at low De and trapping at high De. (c) Stacked confocal image of the events involving trapping of particles near the outer channel wall/concave wall. (d) Schematic of the positions where the confocal images were taken to determine the position of trapped 10 µm particles. (e) Intensity plot across half height and width of the channel to determine the movement of particles in the secondary vortex ...... 41
16 Effect of increase in flow rate on particle focusing. Intensity scans across the channel width for 10 µm (a) and 15 µm (b) particles in all the three stages. The insets show the pseudo-colored fluorescent images of the positions of the focused streams of particles in the channel in their respective stages...... 44
ix 17. Effect of secondary Dean vortices on particle mixture. Intensity scans for the mixture of 10 μm and 15 μm particles at 0.9 mL/min (a) and 2.3 mL/min (b) ...... 45
18. Effect of primary Dean vortices and inertial lift forces on particle focusing. (a) Schematic illustrating the effect of curvature on the focusing positions. Larger particles focus in a single position closer to the inner channel wall and an appropriate outlet system can enhance the collection of the particles/cells sorted according to their sizes. (b) Intensity plot of 20 µm diameter particles across the width of the channel at the end of each loop in the spiral (loop1 being the inner- most loop and loop 4 being the outer –most). The two inset figures show the fluorescent images of the 20 µm polystyrene particles at the inner most loop (loop1) of the spiral device (500 µm×110 µm) and at loop4 focused in a single stream near the inner channel wall...... 50
19. Numerical optimization of spiral device design. (a) Optimization results for Design1 (250 µm×75 µm). Dean number decreases as b (0.159, 0.0795) increases, although there is not much change in the order of the curve, except for the amplitude. The residual for each case is denoted by the dotted plot of the respective color. The point of zero residual gives the approximate focusing length required for forming a spiral with initial radius, a = 2 mm and the given dimensions of the rectangular channel. (b) Optimization results for Design 2 (500 µm × 110 µm) with b the same as the one taken for the previous device . The change in cross-section changes the hydraulic diameter, thereby changing the order of the exponential decrease in De with respect to increase in downstream length...... 52
20. Optimization of flow parameters for plasma extraction. (a) Image of design1. (b) Bright-field image of the outlet system. (c) A mixture of 7.32 µm, 10 µm, 15 µm and 20 µm diameter fluorescently labeled particles is injected at the flow rate of ~1 mL/min in the device. (d) Almost all the particles focus in a broad stream near the inner channel wall eluting in outlet1. (e) Flow cytometer results for the particles collected at the outlets...... 57
21. Optimization of flow parameters for blood cell sorting. (a) Image of design2. (b) Bright field image of the outlet system. (c) At 1.8 mL/min flow rate (F1), two focused streams are observed, the narrow stream in the middle of the channel is formed by 7.32 μm particles and the broad stream near the inner channel wall is the composite of three streams of 10, 15 and 20 μm particles. Particles in the range of 10-20 μm are obtained out of the first outlet, and the 7.32 μm particles are obtained from the second and third outlets. (d) Fluorescent image of the focused streams of all three particles, 10 μm, 15 μm and 20 μm in diameter, at the flow rate of 2.2 mL/min (F2). (e) Normalized focusing position of particles (x is the distance of the focused stream from the inner channel wall, and w is the width of the channel) as function of De ...... 58
22. Effect of blood dilution on sorting efficiency. (a) Bright Field images of blood cells focused in the outermost loop of an Archimedean spiral (250 µm × 75 µm)
x with 10, 50, 100 and 200 fold dilution. (b) A log-plot of normalised width of the focused stream (δ is the width of the focused stream and w is the width of the channel) with the dashed curve defining the plot for design 1and the solid curve defining the plot for design 2. (c) Plot of separation efficiency of extraction of plasma in 250 µm×75 µm spiral device as a function of dilution of the whole blood ...... 61
23. Plasma extraction from diluted blood sample. (a) Bright-field image of blood cells distributed throughout the width of the channel at the inlet of design1. Arrows indicate the white blood cells. (b) All the cells, including platelets elute in the 1st and 2nd outlet. (c) No cells are obtained, only plasma from outlet 3. (d) Bright field image of the cells focused in a broad stream near the inner-channel wall, thereby eluting in the first and second outlet. Samples collected from each outlet were centrifuged and then stained and observed...... 63
24. Blood cell sorting from diluted human blood. (a) Bright field images of the stained samples after they were collected from each outlet of the design2 and centrifuged. Inlet has all the cells present. Arrows indicate the white blood cells. (b) Outlets 2 and 3 have RBCs and platelets (c) Outlet1 has majority of WBCs (Neutrophils, Eosinophil, and Monocyte), some platelets and very little RBCs. and (d) outlet 4 has only diluted plasma and platelets ...... 64
25. Hemocytometer results with normalized cell count. Cells were counted at each outlet showing ~ 90% efficiency of separation of RBCs from WBCs ...... 65
26. Sorting of HPET cells using FACS. (a) Bright field image of the sample of HPET cells sent to CCHMC for FACS sorting. (b) Size estimation of the set of cells. (c) Zero viability obtained at the outlet of the FACS sorter...... 70
27. Schematic showing the protocol of sorting of cells in the spiral device from trypsinizing and suspending the cells to cell count after sorting...... 72
28. Plot of viability of HPET, LNCaP and DU-145 cells determined over a range of flow rates...... 74
29. Sorting of PCa cells using spiral sorter. (a) The pseudo-colored image of the focused larger cells near the inner wall of the spiral microchannel and eluting in outlet 1. (b) Normalized cell count of the sorted cells ...... 76
30. Effect of sorting on rate of proliferation. (a) Proliferation assay data over a period of 120 h for LNCaP cells. (b) Proliferation assay data over a period of 72 h for HPET cells ...... 77
31. Effect of inertial sorting on androgen-mediated transcription. (a) Luciferase expression in LNCaP cells treated with DHT and Casodex (CSX). (b) Lucifersae expression in HPET cells treated with pSVoAR, DHT and OHF ...... 79
xi 32. Schematic representing the concept behind sphere formation assay. In suspension, cells with stem cell like properties form spheres, whereas cells without any stem cell like properties die ...... 80
33. Evaluation of stem-ness of sorted HPET cells. (a) Confocal image of the HPET cells plated on 10cm Primaria plate coated with matrigel, before sorting. Sphere formation assay is performed on the un-sorted (control) and sorted (sub-population 1(outlet1) and sub-population1 (outlet 2&3)) HPET cells. After 14 days, the cells in each case were imaged and counted. Control/un-sorted cells had large (b), medium (c) and small (d) spheres. (e) Spheres from cells eluting in Outlet 1 had medium sized spheres, (f) Spheres from cells eluting in Outlet 2&3 were smaller. (g) The spheres in each case were counted using Image J and plotted. Error bars were calculated using n = 3...... 81
34. Evaluation of stem-ness of sorted HuSLCs. (a) Confocal image of the HuSLCs plated on 10 cm Primaria plate coated with matrigel, before sorting. Sphere formation assay is performed on the un-sorted (control) and sorted (sub-population 1(outlet1) and sub- population1 (outlet 2&3)) HuSLCs. After 14 days, the cells in each case were imaged and counted. (b) Spheres that formed in Control were very dense and large in size as seen in the zoomed-in image (c). (d) Spheres from cells eluting in Outlet 1 had medium sized spheres, (f) Spheres from cells eluting in Outlet 2&3 were smaller. (g) The spheres in each case were counted using Image J and plotted. Error bars were calculated using n = 3...... 82
35. Sorting of doublet and single HPET cells. (a) Image of the spiral microfluidic device used to sort the doublets from single HPET cells. (b) Fluorescent image of the focused streams of HPET cells near the inner channel wall (as indicated by the arrows)-FITC filter. (c) Bright-field image of the unsorted cells at the inlet of the device (20X). (d) Bright-field image of the larger cells-doublets and triplets that eluted in Outlet 1. (e) Bright-field image of the smaller single cells eluting in outlets 2 and 3(f) ...... 84
xii LIST OF SYMBOLS
ap = particle diameter
AR = Aspect ratio b = s/2π
CL = Co-efficient of lift d = Amplitude (De) difference for angle of instability
D = Capillary diameter
De = Dean number
DeC = Critical Dean number
Dh = Hydraulic diameter
Dei = Dean number of the inner-most loop
Deop = Optimized Dean number
FCF = Centrifugal force
FCP = Centripetal force
FD = Force due to Dean drag
FL = Net lift-force
Fs = Shear-induced inertial lift force
FW = Wall-induced lift force
FΩ = Rotation-induced lift force
G = Shear rate
H = Channel height
Lm = Dean migration length
xiii N2 = Nitrogen (molecular formula)
O2 = Oxygen (molecular formula)
Q = Volumetric flow rate r = Inner-most channel radius (in Archimedean spiral)
R = Radius of curvature of the convex surface (inner-wall) of the curved channel
Re = Reynold’s number
S = Center-to-center spacing of spiral channels
U = Average channel velocity
UD = Average Dean velocity
W = Width of the channel
W0.5 = half width of the channel x = Downstream length of the spiral xs = shift of the center of maximum velocity from center of the channel width
ν = Kinematic viscosity
θ = Angle between each point where the radius of the loop, R, is calculated
θi = Angle of instability
µ = Viscosity of the fluid
xiv CHAPTER 1
INTRODUCTION
Sorting is a fundamental step in preparation and analysis of cellular samples in diagnostics, therapeutics and clinical applications.1-3 It involves isolation of a target cell population from a complex heterogeneous mixture. On macroscale, this is done most commonly using conventional cell separation techniques such as centrifugation, fluorescence-activated cell sorting (FACS) and magnetic-activated cell sorting (MACS).1,2,4 Centrifugation uses centrifugal force to sediment components of a heterogeneous mixture.2,3 Although widely used due to ease of operation, centrifugation cannot be used to sort complex cell mixtures and often has poor washing effect of sediment. On the other hand, FACS is an automated, robust approach that can detect a wide range of fluorescently labeled cell signatures like cell size, cell surface markers and density.1,4,5 Another commonly used technique is MACS which uses magnetic labels to assist sorting and is popular for rapid processing and throughput.6,7 The advantages of these conventional techniques are detracted by their limitations, including high cost of operation, large footprint, need for trained technicians, and can lead to low viability of fragile cells like cancer stem cells (CSCs). In recent years there has been a growing focus on microfluidic techniques for cell separation since they provide new separation modalities and possibility of integration in point-of-care systems. Additionally, microfluidics offers cell separation with smaller sample and reagent volumes, high sorting efficiency, and reduced analysis time and low cost of fabrication.6,8,9
1 Microfluidics for cell sorting
Microfluidic sorting techniques can be broadly categorized into active and passive, based on their principle of operation.7,9-12 Active techniques require an external force field to enable sorting and can be used for both label-based and label- free sorting regimes. They include the use of electric, magnetic, optical and acoustic fields for cell separation. On the other hand, passive techniques rely on pressure driven flows, channel geometry and inherent hydrodynamic forces.
Following sections discuss both active and passive microfluidic techniques in further detail.
Electrokinetic mechanisms like electrophoresis, dielectrophoresis (DEP) and electroosmotic flow, utilize the effects of applied electric field on cells and droplets containing cells leading to cell/droplet migration and consequential sorting.7,13-15 Electrophoresis uses direct current (DC) to facilitate cell migration, whereas DEP uses alternating current (AC) for the same.
DEP enhances cell migration by polarizing the cell. Along with directly sorting cells, DEP can be used to sort emulsified droplets containing single cells. This allows the cells to remain safe and intact for post-sorting analysis. Fluorescent labeling of cells allows determination of cell type and identification for further sorting when encapsulated in droplets. Bead-based labeling has been used by various groups to enhance the electric force exerted on cells according to their size.
Hu et al.13 and Cheng et al.19 designed a DEP device that used beads bound to target cells for the purpose of sorting and filtering. DEP is also used for label-free sorting. Huang et al.17,18 demonstrated the concept of using DEP to sort cells depending on their polarizability. This concept has been further integrated with other approaches like gravitational, centrifugal and magnetic field to enhance separation. Unlike DEP, electrosmotic flow sorts the cells by migration of the solvent containing the charged ions/cells.7-14 Although very effective, DC electric fields can cause Faradaic reactions leading to generation of bubbles and harmful
2 compounds which can affect the viability and functionality of cells. Additionally, the approaches using electric fields can get easily influenced by surface charges and pH and can in turn affect the cell physiology.7,14,17,19
Another active technique is magnetophoresis which uses the differential effect of magnetic field on cells labeled with magnetic beads or cells which are magnetically responsive.7,16,20 The bead-based approach, although quite efficient in cell sorting, does not yield high throughput and contaminates the sample due to labeling.21 Along with time consumption during sample preparation, there are potential problems in removing the magnetic beads after the sample has been sorted. To overcome this disadvantage, label-free magnetophoresis has been used to sort cells according to their natural response to the applied magnetic field, for example, red blood cells have iron content in methemoglobin and iron is magnetically responsive.7,20,22
Hence, this technique has been used to magnetically sort cells without using any labels. These label free approaches are only suitable for blood cells with iron component and cannot be used for cells without any non-magnetic component without bead-based labeling.
Recently, acoustofluidics or acoustic microfluidics has emerged as a new area of flow cytometry and cell sorting. Use of acoustic field resonating within a microchannel has been used to sort cells providing both label-based and label-free sorting mechanisms.23,24 Jakobsson et al.23 used bulk acoustic waves to sort fluorescently labeled cells which when detected by the camera led to transduction of the ultrasound waves, thereby modifying and controlling the trajectory of the identified cells. Like DEP, acoustic standing waves have been used to sort larger cells from smaller ones since larger the cell, higher is the acoustophoretic force. Lee et al.25 have demonstrated the label-free technique for the use of acoustic field to trap particles in vortices generated by resonance of the air liquid interface within a microchannel. This sorting mechanism
3 although quite selective lacks in throughput and requires additional equipment along with extensive device and sample preparation.
The final set of active techniques is based on optical methods. For example, optical trapping has been used to sort cells based on their refractive index and size.7,26,27 This method is usually referred to as the optical tweezer or optic switch.28 Although optic-based methods are more effective for fluorescent labelled cells, Lau et al.27 showed the use of laser tweezers based on automated Raman-activated cell sorter that functions without fluorescent labels. A major limitation of optical methods is the processing time which leads to lower throughput and narrow application range. In fact, most of the active techniques either face a trade-off between sorting efficiency and throughput or rely on labeling to improve the performance of the sorting technique. On the other hand, passive methods of separation offer a variety of sorting techniques without the use of labels while offering high efficiency of sorting along with high throughput.
Passive separation techniques have been widely explored for cell sorting as they do not use external forces or labeling, and utilize pressure driven flows and channel geometry to enhance separation.6,7,9,10,24 Passive microfluidic techniques offer a number of advantages over the active techniques including integrable with a lab-on-chip analysis systems, label-free, high throughput, no power consumption since there is no external force field involved, and low cost along with being disposable depending on the material used for fabrication. Among the passive techniques, the most commonly used for sorting and selective isolation of cells is deterministic lateral displacement (DLD) which uses laminar flow and pillar structures for asymmetric division of flow which aides in size and deformability based cell separation.7,29 Huang et al.29 used an array of posts to sort larger cells from smaller cells since the smaller cells traverse the posts more easily than larger cells. Use of pillars and weir for sorting has also been utilized for
4 circulating tumor cell isolation since the pillars can be coated with cell specific biomarkers to facilitate the isolation.7,30 The major disadvantage of this technique is lower throughput and high complexity of design and fabrication.
Another passive technique called pinched flow fractionation (PFF) uses pinching of the sample flow with the sheath of buffer to pinch the cells against the wall. This pinching causes the cells to separate according to size when the channel is broadened.7,30-34 In addition to PFF, hydrodynamic filtration, introduced by Yamada et al.34, uses multiple branched outlets to sequentially drain the fluid from the microchannel. This facilitates size-based of sorting of cells with the smaller cells eluting in the proximal outlets. This technique has been quite successful in the enrichment of blood cells, but there is a trade-off between high efficiency of separation and throughput in most of these devices. Another passive technique which is popular for blood component separation is centrifugal microfluidics. This technique combines centrifugation with microfluidic channels. This concept is more popular with immunoassays, but can be used for blood component separation.1
Along with size based cell sorting, passive microfluidic devices have been used for sorting based on other cellular properties like deformability and surface biomarkers. For example margination is a phenomenon where leukocytes which are less deformable than red blood cells
(RBCs), get displaced to the wall of the microchannel and RBCs migrate towards the axial center. Hou et al.35 used margination to sort malaria infected RBCs which are stiffer than healthy
RBCs with an efficiency ranging from 75% to 90%. This technique is limited to blood cells and cannot be used where hemodynamic effects are absent.
Microfluidic techniques have been used in conjugation with immunoselection, where beads coated with cell specific antibodies/cell surface markers are introduced along with cell
5 mixture to sort the target cells. This method can be lead to both positive and negative selection depending on the surface antigen expression. This technique is very selective and sensitive cell types. For example, Toner et al.36 used microposts coated with the anti-EpCAM antibody to immobilize circulating tumor cells (CTCs) of epithelial origin with 50% purity and >90% selectivity. However, EpCAM is not expressed by all CTCs, particularly CTCs which are not of epithelial origin.37,38 Hyun et al.39 introduced a microfluidic chip with herringbone structure with immobilized anti-CD45 antibody to capture leukocytes while releasing the CTCs into the outlet.
The major challenge for immunoselection is when same surface antigen is expressed by multiple cell types or if the antigens for the target cells have not yet been discovered.7,9,13 Inertial microfluidics is an emerging area of microfluidics that relies on the carrier liquid and hydrodynamic forces acting on cells (or particles) to separate them based on size.7,40-48 It has the potential to address the aforementioned shortcomings and offer the ability to perform separation in a continuous, flow-through manner at low cost and high efficiency.
Inertial microfluidics
Inertial migration was first observed by Segré and Silberberg in 1960s who experimented with neutrally buoyant particles in capillaries and observed a narrow annulus formation at ∼0.2D from walls of a capillary of diameter D (Figure 1a).40,44,45,48,49 This migration behavior is believed to be caused by the balance of lift forces arising from the curvature of the parabolic velocity profile (the shear-induced inertial lift, FS) and the interaction between particles and the channel wall (the wall-induced lift, FW). With advent of microfluidics, observations of the same phenomenon were confirmed and identified that for particles of diameter ap in a channel of
4 diameter D, the net lift force scales as FL CLap , where CL is the lift-co-efficient. This strong
6
Figure 1. Particle focusing due to inertial and Dean migration. (a) Focusing of particles in a circular tube. (b) Schematic showing the focusing of particles in 4 positions along the channel walls in a square micro-channel. (c) Schematic showing the stage 2 equilibrium position in a rectangular straight channel. (d) Single focusing position of particles/cells in a curved rectangular channel due to the presence of Dean vortices
dependence of the lift forces on size offers a powerful ability to perform separation in a continuous, flow-through manner at low cost and high efficiency.40,45-47,50
While the balance of the two lift forces can successfully explain particle focusing in a capillary, square and rectangular microchannels present a more complex situation due to radial asymmetry. Since the shear-induced lift causes particles to migrate away from the channel center, down the shear gradient toward the channel wall, one would expect particles to equilibrate along the perimeter of the channel (including corners) to achieve force balance.
However, work in microfluidic channels has identified four distinct focusing positions centered at each face in square microchannels (Figure 1b). The absence of particles in corners suggests that additional lateral migration effects take place near channel walls that cause the particle migration toward wall centers. Indeed, a rotation-induced lift force (FΩ) was proposed in the early work by Saffman et al.51 and recently confirmed by Zhou et al.50 While negligible away from the wall, FΩ is significant at the channel wall and explains the asymmetric equilibrium positions in rectangular microchannels. The introduction of a rotation-induced lift force leads to a more complex model of inertial migration at finite Re. Both shear and wall induced lift forces
7 dominate particle migration toward channel walls. Once the initial equilibrium is reached, near channel walls particle motion is dominated by the rotation-induced lift force. As a result, particles migrate to the center points of walls (Figure 1c). This two-stage model of inertial focusing developed by Zhou et al.50 is generally applicable to rectangular microchannels of any aspect ratio at finite Re and can be used to aid design of inertial microfluidic systems.
Taking advantage of both stages of inertial focusing, Zhou et al.50,7 demonstrated a microfluidic system for complete separation of cells by manipulating the inertial equilibrium positions through modulation of the microchannel aspect ratio (MARCS system). This system offers high purity (>90%) and ultrahigh separation efficiency (>99%) with operational flows
~100 mL/min. The inertial forces in straight channels can also be manipulated by expanding the microchannel geometry, leading to the formation of microvortices which can selectively isolate and trap particles or cells from a mixture. Isolation and enrichment of both extremely low concentration (~1 particle/mL, enrichment factor of 105) and extraordinarily high selectivity
(1:105) were successfully demonstrated in trapping particles/cells but with a low throughput.
Nevertheless, despite success in the development of straight microfluidic channels for cell sorting, a number of challenges remain. Multiple focusing positions present in a straight microchannel make it difficult to extract sorted cells or particles when mixtures contain more than two different types.
For sorting complex cell mixtures with high throughput and efficiency, it is desirable to reduce the number of focusing positions for each cell types to one such that a planar outlet design can extract cells easily without compromising their viability. Spiral microfluidic channels can accomplish this by introducing Dean drag (FD) arising from secondary flows and channel curvature to balance lift forces (Figure 1d).42,52-55 The result of this new force balance is a single
8 focusing position near the inner channel wall. This concept has been used by Seo et al.32 to sort particles in a Fermat spiral and by Kuntaegowdanahalli et al.42 to separate neural cells in an
Archimedean spiral (~80% efficiency). Sun et al.56 showed sorting of spiked HeLa cells and 20× diluted blood cells in a double/fermat spiral with a ~90% efficiency, but only ~80% recovery rate of sorted HeLa cells. Spiral channels with trapezoidal cross section has also been used by Han et al.55 for >90% white blood cells (WBCs) enrichment, but the trapezoidal cross-section has complicated fabrication process which requires the master for PDMS (polydimethylsiloxane) casting be milled rather than patterned on a wafer.
Motivation
Curved microchannels have been widely used in the field of microfluidics, particularly for on-chip reagent mixing and cell sorting. However, the dynamics of fluid flow in curved microchannels, especially spiral microchannels, is not completely understood. The current model of spiral sorting devices is based on the assumption of two counter rotating vortices which facilitates a single focusing position. This model does not adequately explain the uncharacteristic focusing behavior of cells/particles in spiral microchannels beyond a certain flow rate. In fact, most of the fluid dynamics studies on this subject to date are limited by the lack of experimental data to confirm numerical predictions and the range of flow parameters used (low Reynolds number (Re < 100) and low Dean number (De < 15)). Hence, there is a need for the study of flow behavior in spiral microchannel and its effects on the sorting mechanism. This work investigates the fluid flow regime in spiral microchannels and proposes an alternate explanation of particle focusing regime at higher flow rates.
Although conventional cell sorting techniques (FACS, MACS) offer high selectivity and throughput, they often yield low viability due to high shear when sorting cancer cells. In fact,
9 mechanical/shear stress has been a known factor for inducing both functional and morphological changes in cells.57 Hence, it is imperative that microfluidic cell sorting devices are designed and optimized to ensure that the sorting mechanism does not alter morphological and functional properties of the cells. The devices in this work, have been designed to sort sub-populations of
CSCs while maintaining high viability and keeping the functionality of the cells intact. They combine hydrodynamic forces in a rectangular channel with Secondary flow/ Dean vortices caused in curved channels to provide a high throughput cell sorting system which can provide low enough shear rate to ensure the viability and functionality of the sorted cells. In addition to two counter rotating vortices, previous spiral sorting device design has been based on the assumption that De is constant for the purpose of determining the focusing length and device dimensions.42 This is not quite accurate since in an Archimedean spiral, the radius of curvature varies with every angle and De varies as a function of the radius. In this work, devices have been optimized for sorting by taking into account the variability of De across the length of the device.
Scope of work
In this work, it is hypothesized that secondary flow in a curved rectangular microchannel develops spatially into multiple vortices with increase in De and these secondary flows can be further manipulated to sort multiple cell lines as well as isolate sub-populations within a single cell line. Three specific aims were investigated to further examine this hypothesis. The first aim investigated the presence and development of secondary flows in low aspect ratio spiral microchannels. The development of these vortices with change in aspect ratio and Dean number was defined. Since, there has been little experimental study of fluid flow in spiral microchannels, there is little evidence of the flow regime and its effects on particle focusing for curved microchannels. This work provided evidence of presence of multiple vortices in spiral
10 microchannels and defined the parametric range of the operation of the device for cell sorting. It developed a better understanding of secondary flows in spiral microchannels and formulated an accurate model which can predict the optimal design parameters for spiral sorting devices. This model not only described the flow behavior in curved channels, but also aided in design of sorting devices with more efficient and precise size-based separation of cells. The concept of two counter-rotating vortices, along with the parametric range provided by this fluid dynamic study, was then used to optimize and design spiral microfluidic devices for two specific cell sorting applications.
Second aim focused on the first application of sorting blood cell components and isolation of cell-free plasma. Numerical as well as empirical model were used to inform device design. The devices were designed by taking into account change in De as a function of change in radius of curvature, thereby reducing the footprint of the devices by 1/5th of the previous design. Additionally, the outlet system was simplified to deal with heterogeneous cell samples, along with providing lower shear at the outlet. The optimized device was then used to sort the components of blood using human male blood as the sample.
The third aim focused on the second application which involved isolation of sub- population of CSCs. Cancer stem cells, being in different state of differentiation, can be sorted from tissue biopsies according to their size. This work focused on optimization of the spiral design to obtain a standard for sorting cell-lines to aid in the study of growth and metastasis of prostate cancer. The approach and devices developed in this work, provide an easy-to-use, label- free and disposable cell sorting tool for cell biologists.
11 Chapter summaries
Following this introduction, Chapter 2 discusses the fluid flow dynamics in low aspect ratio rectangular Archimedean spiral microchannels. Both numerical and experimental methods were used to investigate the flow development in three different aspect ratios of spiral microchannel cross-section. Critical Dean number that defines the transition from primary to secondary Dean vortices is introduced and plotted for each of the three geometries. The effects of secondary Dean vortices on particle focusing regime is discussed along with determining the parametric range for size-based cells sorting.
Chapter 3 will discuss optimization of the spiral sorting device for a smaller footprint with high efficiency and throughput of blood cell sorting. These devices were used for plasma isolation and sorting white blood cells and red blood cells. Numerical and empirical design equations are provided for spiral design. The sorting device is small enough to be fabricated using roll-to-roll processing, providing a possible mass-production alternative to soft lithography. Furthermore, this device can be integrated with other lab-on-chip devices for further cell isolation and sample preparation.
In Chapter 4, the optimized spiral devices were used to sort the sub-populations of prostate CSCs. For this application effects of shear on cell viability, proliferation and functionality is discussed. Sphere formation assay evaluating the stem-ness of the sorted sub- populations is then discussed and analyzed. Human prostate epithelial-TERT (HPET) cells and human stem cell like cells (HuSLCs) are used to represent the prostate cancer stem cell lines.
The bulk of the prostate cancer cells is represented by the use of LNCaP cells (prostate cancer cells metastasized to lymph nodes).
12 Finally, Chapter 5 will summarize and conclude this work, along with a discussion of the remaining challenges and potential future directions.
13 CHAPTER 2
FLOW IN SPIRAL MICROCHANNELS
Introduction
Curved microchannels have been gaining prominence in the field of microfluidics for various applications such as mixing63,69,70, cell sorting53-55, enhancing reaction in immunoassay chips58, heat transfer (thermal management)59 and for size reduction by compacting large length in a small area58. The flow in curved channels is considerably more complex than straight channels owing to the presence of Dean vortices in addition to inherent hydrodynamic forces.
Although widely studied at macro-scale, the study of the flow dynamics in curved microchannels is fairly recent.
The first conclusive research to analyze flow in curved macrochannels was done by W.R.
Dean in 1927.60 He showed that in curved channels/pipes, the plane Poiseuille flow is disturbed by the presence of centrifugal instability (FCF) causing the flow to be unstable, resulting in vortices above a critical value of a non-dimensional number, De.61, 62 Dean defined De as a control parameter for the secondary flow which directly represents the Dean force or force due to secondary flow in curved channels.
퐷 퐷푒 = 푅푒√ ℎ (1) 2푅
14 where R is the radius of the curvature of the convex surface (inner-wall) of the curved channel, and Dh is the hydraulic diameter (Dh = 4A/P, where A is the area of the channel cross section and
P is the perimeter of the cross section). Re is Reynolds number and is given by
푈퐷 푅푒 = ℎ (2) 휈
where U is the average channel velocity and ν is kinematic viscosity. Hence, the strength of the secondary flow is strongly dependent on the dimensions of the channel and the radius of curvature. This change in velocity gradient causes manifestation of secondary flow in a rectangular channel in the form of two counter-rotating vortices, known as the primary Dean vortices or the main secondary flow vortices.61,63-65
Additionally, for macroscale, Eustice et al.66 provided experimental proof of the presence of secondary flows by injecting dye into the water flowing in curved pipe and Taylor et al.67 found that the instability arises in the form of an array of longitudinal vortices. This led to further development of longitudinal vortices at higher De in classical fluid mechanics system of curved pipes. In fact, Cheng et al.62 provided a numerical study of the fully developed laminar flow in curved rectangular channels and reported on experimental study using flow visualization to prove the presence and significance of Dean vortices in rectangular channels. They also reported that in addition to the two primary secondary flow vortices, there is further formation of secondary vortices which only form after a certain threshold of De is crossed (Figure 2). A similar experimental investigation of a fully developed laminar flow in curved rectangular macrochannels was reported by Sugiyama et al.68. They also showed the development of these vortices with change in De and aspect ratio of the channel.
Although, the presence of secondary flows in curved microchannels has been utilized for multitude applications in microfluidics, it has always been assumed that the flow has two counter
15
Figure 2. Dean flow dynamics at low and high De. (a) Schematic illustrating two counter rotating vortices in a curved rectangular channel at lower De. (b) Schematic illustrating flow behavior at high De causing the formation of multiple vortices
rotating vortices similar in amplitude. This assumption of two counter-rotating vortices is partially supported by various fluid mixing studies which although quite detailed, are primarily simulation studies and are usually confined to lower Reynolds numbers (Re < 20). For example, the detailed studies of fluid mixing in curved microchannels for 1 < Re < 10 by Liegler et al.69 and convective vortex based mixing in spiral microchannels by Sudarshan et al.70, do not show presence of more than two vortices in the channels. However, at higher De and Re the assumption of two-counter rotating vortices falters since the focusing positions of cells vary from the predicted ones. We observed this with a diluted blood sample when the RBCs focused closer to the outer channel wall/concave wall instead of focusing closer to the inner channel wall/convex wall (as predicted by the assumption) in a spiral sorting device. This can be attributed to the anomaly in the assumption of two counter rotating vortices and lack of physical understanding from systematic experimental investigation of the flow in curved microchannels.
Hence, it is significant to examine the origin of the instability in the microchannel, especially the low aspect ratio channels since the sorting mechanism works within a particular range of De.
16
In this work, presence of multiple secondary flow vortices (secondary Dean vortices) in low-aspect ratio rectangular spiral microchannels is confirmed experimentally and computationally. A non-dimensional parameter, critical Dean number is introduced to describe the threshold for the onset of secondary Dean vortices. This work offers insight into the phenomenon of the development of the secondary flows in spiral microchannels and improves the understanding of the concept of particle focusing in the spiral devices. Furthermore, the concept presented can assist in manipulation of the interaction of multiple vortices and other inherent fluid forces to achieve higher efficiency and selectivity in cell sorting. This work also paves way into understanding the reason behind the need for a certain range of De for cell focusing in sorting techniques.
Methods
Geometries
The Archimedean-type spiral channel design offers a gradual development of the secondary flow vortices. Alternative spiral types, such as Fibonacci-type which are most commonly found in nature, cause dramatic changes in curvature and substantial reduction in
Dean flow strength downstream. Although Fermat spirals (i.e., in-and-out spirals) have been used in the past primarily due to convenience of having input and output ports in-line, these designs offer weaker downstream progression of Dean flows and are generally larger. Thus, the
Archimedean spirals are the most reasonable solution.
Archimedean spirals can be designed by calculating radius of curvature at each downstream point. The design equation relates the radius of curvature to the initial channel radius and spacing between channels in each loop. It is expressed as
17
푅 = 푟 + 푏휃 = 푟 + 푠휃/2휋 (3)
where r is the inner-most channel radius, b = s/2π, s is the center-to-center spacing of spiral
channels, and 휃 is the angle between each point where radius of the loop R is calculated. To
investigate flow characteristics and development of the secondary flow, three configurations of
the spiral with three different aspect ratios (AR) were used. The details of each geometry were as
follows: g1 with AR = 0.6 (250 µm × 150 µm), g2 with AR = 0.4 (250 µm × 100 µm) and g3
with AR = 0.2 (500 µm × 100 µm). The table in Figure 3 lists the radius of curvature at various
positions for the three different designs, specifically for the two loops at which the confocal
images were taken. A three inlet system was used to provide 1/3rd confinement of the dye at the
input so that higher contrast was achieved with more time before the dye mixes downstream. A
Figure 3. Geometric configurations for fluid flow dynamic study. Schematic layout of the spiral device with the table listing cross-sectional dimensions of the three geometries and the respective radii of curvature (convex radius of curvature, R).
18 single outlet was located at the center of each spiral. Each device was designed with the inner- most radius as 2 mm.
Microfabrication
Devices were fabricated in PDMS (Sylgard 184, Dow Corning) using the standard soft lithography process (Figure 4). The 100 µm high masters were patterned in SU-8-2075 negative photoresist (Microchem Corp.). SU-8 was coated on a 3 inch Si wafer cleaned with BOE
(buffered oxide etch). Post leveling, the coated wafer was baked at 65ºC for 10 min and at 95ºC for 20 min. After 15 min of cooling, the wafer was coated with glycerin and exposed for 45 s to
UV at the power of 5 mW/cm2 using a long pass filter (I-line 365 nm). Post exposure, the wafer was washed with DI water and dried using a N2 gun and baked at 65ºC for 7 min and at 95ºC for
12 min. Standard Microchem SU-8 developer (propylene glycol monomethyl ether acetate) was used to develop the master. The development took 15-20 min, especially when the developer was gently sprayed onto the wafer, followed by a rinse with DI water and N2 dry. Following development, O2 plasma (CS-1701 RIE, March Plasma System, CA) was used to clean the surface for any residue
Figure 4. Soft-lithography process for device fabrication. (a) Schematic of the soft- lithography process of master fabrication and PDMS casting. (b) Image of the PerMX-3050 master used for device fabrication. 19
For a 150 µm high master, a dry film negative resist-PerMX 3050 (Dupont) was used.
Three layers, each with a thickness of 50 µm was used to fabricate the master. The film was laminated onto the 3 inch Si wafer using a rolling laminator laminator at 85ºC with <1 RPM. It is very important to make sure that no bubbles form during layering. The lithography process used was very similar to SU-8 master- fabrication process as described earlier. The pre-exposure bake was set for 10 min at 95ºC. The exposure time was 300 s at 5 mW/cm2. Followed by a post- exposure bake at 95ºC for 3 min and cool down, the photoresist was developed by spraying
PGMEA (Propylene glycol monomethyl ether acetate) for 20 min. The master was ready after the O2 plasma cleaning (Figure 4b).
A mixture of PDMS base and curing agent (10:1 ratio) was poured on the master treated with Sigmacote® (Signma Aldrich); after degassing PDMS was cured for 4 h on a 60°C hotplate.
The cured PDMS devices were peeled off, and inlet/outlet ports were punched with a 14 gauge syringe needle. PDMS was bonded to standard glass slide using a hand-held plasma surface treater (BD-20AC, Electro-Technic Products, Inc.). Alternatively, CS-1701 RIE, March Plasma
System was used for bonding.
Device operation and imaging
Fluorescein (1 µM) solution was used to provide contrast for visualizing the stream lines.
A syringe connected to a device with 1/16” peek tubing (Upchurch Scientific) and proper fittings
(Upchurch Scientific) provided input flow using a syringe pump (Legota 180, KD scientific).
The flow behavior was first visualized using an inverted epi-fluorescence microscope (IX71,
Olympus Inc.) equipped with a 12-bit high-speed CCD camera (Retiga EXi, QImaging). To image the cross-section, confocal microscopy using Zeiss LSM710 LIVE Duo Confocal
Microscope was performed with the similar experimental set-up. Cross-sectional images were
20 taken at each loop of the spiral starting from the outer-most loop (closest to the inlets) at the interval of 60º and analyzed using Zen lite software along with Image J.
To evaluate the effects of secondary flow on particle focusing and trapping, 10 µm diameter fluorescent polystyrene particles were used at a volume fraction of 0.1%. To ensure that the particles are neutrally buoyant, the solution for particle suspension was prepared with 0.06% saline and a drop of Tween-20 (Fisher Scientific) in filtered deionized (DI) water. For the purpose of determining the behavior of a particle mixture at higher De, a mixture of 10 and 15
µm diameter fluorescent polystyrene particles was suspended at the volume fraction of 0.2%. For streak velocimetry images, an inverted epi-fluorescence microscope (IX71, Olympus Inc.) equipped with a 12-bit high-speed CCD camera (Retiga EXi, QImaging) was used. At least 50 images were stacked using Image J for determining position of the streaks across the width of the channel.
Evolution of primary Dean vortices
Computational analysis
As discussed in previous section, particularly Figure 2, in curved channels, the laminar
60-63 Poiseuille flow is subjected to centrifugal force (FCF). This external force disturbs the parabolic profile of the laminar flow and causes the maximum point of velocity distribution to shift from the center of the channel towards the concave wall of the channel.60,63,64 This shift causes a sharp velocity gradient to develop near the concave wall between the point of maximum velocity and concave wall where the velocity is zero. This leads to increase in pressure gradient near the concave wall. The local velocity near the channel walls is not sufficient to provide complete balance of the pressure gradient. This gives rise to what is known as Dean instability
21
which in turn leads to recirculation of fluid from center of the channel towards the outer channel
wall and back towards the center leading to the development of secondary flow from the concave
wall to the convex wall of the channel.
Numerical simulation (Star-CCM+) in Figure 5a show the velocity gradient at the inlet
(section 1) and outlet (section 2) of a single segment of spiral microchannel (250 µm×150 µm)
with the inner radius of 2 mm (v (average velocity) = 1 m/s). The shift in velocity gradient is also
evident from the velocity profile plots from the simulations. Figure 5b shows the parabolic
velocity profile of the straight section near the inlet of the simulated loop, with the center of
velocity profile at the center of the channel width. As the flow downstream is affected by the
centrifugal force due to the curvature, the center of velocity shifts towards the outer channel
Figure 5. Numerical simulation of a single loop of Archimedean spiral. (a) Top-view of the single loop of spiral model simulated using Star CCM+ at 1m/s. (b) Parabolic velocity profile obtained at the straight channel near the inlet of the spiral loop (section-1). (c) Plot of the velocity profile near the middle of the loop (section-2).
22
wall/concave wall (Figure 5c). The pressure and velocity gradient imbalance caused by Dean
instability results in vortices above a critical De value.
Figure 6 shows profiles of the velocity gradient across the cross-section of the curved
rectangular channel with increasing De, along with the corresponding simulated Dean vectors.
For continuity, all of the panels are presented here, even those with secondary Dean vortices,
which will be discussed in the next section. At De = 0 (straight section), maximum velocity is at
the center of the channel cross-section, leading to no secondary flow and hence, no cross-
sectional vectors. At De = 17.2, the center of maximum velocity (red region) shifts towards the
concave wall leading to imbalance in pressure, thereby causing recirculation of flow from the
Figure 6. Simulated cross-sectional images of the scalar velocity profile and corresponding images of the Dean flow vectors. The single loop was simulated for De=0 (straight channel) and De=343.09.
23 center of the channel towards the wall, leading to formation of counter rotating vortices called primary Dean vortices. As De is increased, the region of maximum velocity not only shifts closer to the concave wall, but increases in area.
The secondary flow behavior can be further described using the shift of the center of maximum velocity from the center of the channel (xS). Figure 7a shows the plot of this shift relative to half width of the channel (W0.5). There is a linear increase in the shift till it reaches the secondary vortex regime where the primary vortices start to expand and sharpen and ultimately bifurcate into four vortices. The center of maximum velocity shift gradually moves further closer to the concave wall (Figure 7), further increasing the pressure gradient and causing the formation of multiple vortices. These results are comparable with the secondary flow behavior at macro- scale where multiple Dean vortices are observed for high De. In fact, it has been reported that in macrochannels the additional vortices develop beyond a critical Dean number (DeC) depending on the aspect ratio of the channel.61-65 Although the simulations computationally confirm the presence of multiple vortices in micro-channels, we had observed the shift in focusing position
Figure 7. Effect of increase in De on velocity profile. (a) Plot of the shift of center of maximum velocity from the center of the channel (xs) relative to half width of the channel (W0.5) as De is increased. (b) The plot of the shift in the secondary flow regime
24 of RBCs at much lower De than indicated by the simulations. Hence, it was imperative to visualize the flow behavior in spiral microchannels and confirm the results obtained from the simulations.
Experimental analysis
In curved macrochannels, a wide variety of methods have been used to study the fluid behavior including dye and smoke contrast and perturbation methods. For example, Dean used perturbation methods to study the behavior of Newtonian fluids in curved pipes and Ligrani et al.64 and Sugiyama et al.68 used smoke contrast in macroscale curved rectangular channels to study Dean flows. These methods are incredibly challenging to perform on the microscale and are thus not used. Hence, numerical studies along with simulation have been the most common methods for the study of secondary flows in the curved microchannels.63,70 In this work, we have used both planar/inverted imaging and confocal imaging using contrast provided by fluorescent dye, to visualize the flow in spiral microchannels, as discussed below.
The onset of instability can be defined by the point at which De is sufficient enough to cause the formation of two counter rotating vortices (primary Dean vortices).61-63,65 To determine
De and flow conditions which lead to the onset and rate of instability, we used inverted microscopy to investigate the effect of introduction of curvature and flow rate on the mixing of dye in spiral rectangular channel (Figure 8a, b). The onset of instability caused the laminar flow to be disturbed and the dye started to migrate towards the wall opposite to the introduction of dye. This migration was believed to be produced by the onset and development of primary Dean vortices.
The rate at which the dye migrated determined the rate of onset of instability, and this was quantified by measuring the angle of instability (θi). This angle of instability was measured
25
Figure 8. Onset of instability as a function of De. (a) Image of the spiral device used for studying the flow behavior in spiral microchannels. (b) Fluorescent image of the inlet showing the three inlet system and the 1/3rd confinement of the dye. (c) Plot of the Angle of instability (θi) as a function of De and the three geometries g1, g2 and g3.The rate of change of the angle of instability defines the rate of the onset of instability in the curved channel The amplitude difference, d, between the plots is dependent on the width of the channel. (d), (e) Fluorescent images taken by inverted microscope using fluorescein for contrast. The white arrows indicate the position where multiple streaks were observed indicating pinching of the dye between multiple vortices (the images were taken at the two loops-1 and 2). θi is the angle of instability that quantifies the migration of the dye towards the opposite wall.
26 at the beginning of the curvature of the spiral channel, with respect to a line which is tangent to the centerline and is located along the middle of the channel. At Q = 0.1 mL/min, De was not sufficiently high enough (De< 3) to cause the instability to set in and lead to formation of primary Dean vortices. As the flow rate increased, manifestation of the instability was observed.
As this happened, it was apparent that the onset of instability and the rate of this onset are highly dependent upon AR of the channel, and the initial angle of instability θi. The variation of the radius of curvature was maintained the same for each of the channels at approximately 4 mm, to compare the rate of change of θi, with a common value of R, as the Dean number was varied. It was observed that higher the AR, steeper is the rate of change of θi and higher is the rate of the onset of instability. Here, the amplitude of the angle defines the rate at which the dye migrates to the opposite wall of the channel and hence, the rate of the onset of instability.
Figure 8c illustrates the dependence of θi on De and AR of the channel. It is also evident that the slope of the plots (θi/De) is the function of radius of curvature and amplitude difference
(d) of De for the angle is a function of channel width. When the flow rate was increased to Q = 3 mL/min, multiple streaks were observed during planar imaging of flow behavior in spiral microchannels (Figure 8d, e). More than 3 streaks of dye indicated the presence of multiple vortices, in the form of both primary and secondary Dean vortices .These multiple streaks were created as the dye accumulated, and was pinched between vortices, after being recirculated in the channel.
Although this observation indicated possible presence of multiple vortices, but it did not provide definite proof of presence of secondary Dean vortices. To visualize the secondary Dean vortices, it was required that we take cross-sectional images of the channel. Since this phenomenon was observed at higher De (higher flow rate), very specific set of conditions were
27 required to have a confocal image clear enough to determine the number of vortices and flow direction. In the following section, we discuss the confocal images which indicate the evolution of secondary flow from two primary counter rotating Dean vortices to multiple smaller secondary Dean vortices.
Evolution of secondary Dean vortices
Computational analysis
Spatial evolution of the Dean vortices provides insight into the effect of Dean number and change in aspect ratio on the development of secondary flows in spiral microchannels. As discussed in the previous section, computation analysis showed the progression of velocity profile as it develops with increase in De (Figure 6). At De = 171, the maximum of velocity gradient expands near concave wall and bifurcates, leading to further elongation and extension of the cross-sectional vortex vectors. As the velocity gradient increases due to increase in De, the vorticity vectors become sharper and begin to shift towards the concave wall which leads to splitting of the primary Dean vortices. This increase in De, causes further increase in centrifugal force leading to development of additional regions of pressure gradient near the concave wall.
This leads to significant gradient shift in the velocity vectors as well. To balance this increase in pressure gradient, there is formation of additional counter-rotating vortices which recirculate the fluid near the concave wall. These vortices are called secondary Dean vortices or additional secondary flow vortices as seen in Figure 6 panel for De = 257. The De at which secondary Dean vortices are distinguishably present is comparable to the macroscale results from the work done by Cheng et al.62 and Sugiyama et al.68 When confocal analysis was performed for the same AR, we observed the presence of multiple Dean vortices at much lower De, as described below. This
28 drastic difference of De at which presence of multiple vortices was observed can be attributed to the numerical models and convergence criteria used to compute Dean vortices and vorticity in laminar flow in the fluid dynamics software.
Confocal analysis
We tested channels for different aspect ratios (mentioned in the methods section) to determine the effect in the development of the vortices at each De. As mentioned earlier, 1/3rd confinement of the dyed DI water was used to provide contrast for streamline visualization
(Figure 9). The reason for using two miscible liquids is to allow uniform streamline development across the entire width of the channel. If immiscible liquids were used, individual flow development and vortex development would occur in each of the liquid space.71
For spiral geometry g1 (AR = 0.6), the rate of onset of instability and its transition to formation of secondary Dean vortices was observed, including spatial development of associated secondary flows. To obtain this information, confocal microscopy was employed such that images were taken with an interval of 60º along both the outer loop, and the inner loop of the spiral microchannel (Figure 10). A constant Re = 200 was maintained as the data was acquired.
Near the beginning of curvature (De = 28.2), due to the onset of the instability, the dye was
Figure 9. Schematic of the 1/3rd confinement of dye used to visualize cross-sectional streamline development.
29 locally advected by secondary flows which were directed from the convex wall towards the concave wall within the central part of the channel. Such fluid motion occurred simultaneously as fluid was locally advected from the concave wall towards the convex wall near the edges of the channel. The resulting local variations of static pressure caused recirculation of the dye from the center towards the concave wall, and then, back towards the center of the channel leading to the formation of two primary vortices (Figure 10a-d).
In the second loop, De = 29.4 and the concentration and recirculation of the flow within primary Dean vortices was evident near the concave wall (Figure 10e-f). Near this location, secondary Dean vortices started to form. In the second loop, although dye contrast was reduced slightly due to diffusion based mixing, the formation of multiple vortex pairs was apparent.
Figure 10f shows the formation of hook shaped vortices which illustrate the redistribution of the dye containing fluid from the concave wall towards the convex wall, and then, recirculation within the primary Dean vortices, as well as the formation of additional, secondary Dean vortices near the concave wall. These vortices were clearly visible at De = 30.6 (Figure 10h).
Secondary Dean vortices developed rapidly as the flow traveled from the first towards the second loop. By the end of the second loop, the dye did not provide sufficient contrast for visualization of vortices, due to diffusion based mixing. Note that local Dean numbers increase as the flow advects through the spiral micro-channel, because of progressively smaller values of
R, which result in stronger effects of concave curvature, and locally stronger centrifugal instabilities.
30
Figure 10. Confocal images of the cross-section of the rectangular spiral microchannel with geometry g1. The images were taken at regular intervals over two loops. These images show a gradual development of secondary vortices in the second loop of the spiral device whereas in the first loop only primary vortices were observed.
31
For AR = 0.4, due to lower aspect ratio as compared to 0.6, the development of the dean vortices was slightly slower. As stated previously, the images were taken at regular interval along the outer loop and inner loop of the spiral microchannel (Figure 11). Re was increased from 200 to 260 to visualize the multiple vortices in the inner loop. As the laminar flow of the dye and DI water entered the curvature of the outer loop, DI water started moving from concave wall to the convex wall due to secondary instability causing formation of primary Dean vortices.
These vortices were fully visible in Figure 11d where the secondary flow is reinforced in the form of counter rotating vortices in equal amplitude and size (De = 36.0). Further downstream, it was observed that due to increase in pressure gradient near the concave wall, the flow started to recirculate in the area closer to the concave wall (Figure 11e). As the flow continued into the second loop or the inner loop, the recirculation near the inner channel wall became more prominent and developed into two small counter rotating vortices in addition to the primary dean vortices (Figure 11g, h).
For AR = 0.2, the Dean vortices develop at slow rate. The spatial development of the flow was visualized using the cross-sectional confocal images in Figure 12. Images were taken at regular intervals along the outer and inner loops of the spiral device. The onset of instability pinched the fluorescein in the center of the channel leading due to the development of two primary vortices (Figure 12a, b). Since the aspect ratio is quite low, the movement of the dye along the channel and its redistribution was very slow and exact formation of two vortices was not clearly visible, but the pinching of the dye confirmed the presence of the two primary vortices (Figure 12d). As the flow moved into the second/inner loop, the De changed from 30 to
43. The spatial distribution in the second loop illustrates the slight change in the number of vortices and development of secondary instability in the flow (Figure 12f, g).
32
Figure 11. Confocal images of the cross-section of spiral microchannel with geometry g2. These images show visualization of cross-sectional flow as it develops with increase in De.
33
At around De = 42.0, the primary instability was unable to maintain the balance of flow due to increase in pressure near the concave wall and resulted in secondary instability which in turn caused the formation of secondary vortices. In each case, the development of additional vortices is dependent on a critical De, beyond which the basic two major vortices start recirculating to form multiple vortices. In the previous section, we simulated g1 geometry, but the multiple vortices were observed at higher De (De>205).
In case of confocal imaging, the visualization shows that multiple vortices are observed at much lower De (De>30), especially for higher aspect ratios. Since, flow visualization does not exactly provide the direction of vortices, based on the simulation results shown in Figure 6, we can assume that primary Dean vortices and secondary Dean vortices rotate in the same direction.
This is in agreement with the numerical simulations of two phase stratified flow done by Picardo et al.71, especially the flow regime described by sandwich-principle circulatory flow. At macroscale, most of the work done to study Dean vortices is contour study62 and dye visualization68 which does not clearly define the direction of secondary Dean vortices.
To determine the size of the secondary Dean vortices and their trend of development, the portion of the channel cross-sectional area of the vortices is presented as it varies with De in
Figure 13 for each of the three spiral geometries. The plots describe the trend only in one half of the channel, since the vortices form in pairs and the other half is a mirror image. Thus, the maximum vortex area corresponds to 50%. The blue data points show the area of the channel covered by one of the two primary Dean vortices, and the red data points show the area of the channel covered by one of the two secondary Dean vortices. The area of the vortices was measured using the image analysis software ImageJ. The area covered by the fluorescein was
34
Figure 12. Confocal images of the cross-section of rectangular spiral microchannel with geometry g3. These images show the development of cross-sectional flow as a function of increase in De.
35 quantified using Image J and the absolute values were then divided by the area of the entire half of the channel to get the percent-coverage of the vortices. For AR = 0.6, (Figure 13a) the rapid change of area associated with the primary vortices stabilized at De ~ 29 at approximately 45%.
The remaining 5% of the channel cross-section was covered by the residual flow. At De>29.6, the secondary vortices started to develop and the area occupied by the primary vortices reduced to approximately 25% of the channel. At De ~ 31.25, both the primary and secondary Dean vortices occupied nearly the same area of approximately 15% each. Similar trends of development were present for channels with lower aspect ratios. For g2 (AR = 0.4) (250 µm×100
µm), the change of area associated with the Dean vortices was comparatively lower as is indicated by the gradient of increase in the area of the vortices (Figure 13b).
The flow patterns changed in the same manner as for case g1, and a pair of secondary
Dean vortices was observed for De>DeC. Here, these vortices were somewhat narrow, compared to the primary and secondary vortices associated with channel g1. As such, the area covered by the vortices and recirculation of vortices is governed by AR and De as the flow moves downstream. The data for the g2 channel show that the area variation of primary Dean vortices has a lower slope (compared to g1 data). The size of the primary vortex then becomes approximately constant for 35 De>36. For most experimental conditions, the area covered by secondary Dean vortices was much smaller than the area associated with the primary Dean vortices. For channel g3, the AR is quite low (0.2), and the movement of the dye along the channel and its redistribution was very slow, and thus, exact formation of two vortices was not always clearly visible. However, pinching of the dye confirmed the presence of the two primary Dean vortices. At De ~ 38, the 36 Figure 13. Plot of Area as a function of De for the rectangular spiral microchannel geometries: g1, g2 and g3. The plot in blue shows the area of the channel covered by one of the two primary Dean vortices and the plot in red shows the area covered by one of the two secondary Dean vortices. 37 primary instability vortices were unable to persist in the same form as they advected downstream, due to local increase in static pressure near the concave wall. This resulted in the secondary instability, which, in turn, caused the formation of secondary Dean vortices. In this case, the secondary Dean vortices were much smaller than the primary Dean vortices. The secondary Dean vortices were also smaller in area when compared with the previous aspect ratio channels, providing additional verification that the strength and shape of the vortices is highly dependent upon AR (Figure 13c) Critical Dean number The flow in spiral microchannels transitions from primary Dean vortices to secondary Dean vortices over a certain threshold of the Dean number. This threshold Dean number is defined as the critical De (Dec). For De >Dec, the perturbation of the primary Dean vortices begins, followed by the development of the secondary Dean vortices. At this point, the pressure gradient between the high velocity area near the concave wall and the concave wall has increased to the point that the primary Dean vortices are unable to maintain the balance of pressure across the width of the channel. To balance this additional pressure near the concave wall, the primary vortices split near the concave wall to recirculate the fluid near the concave wall leading to the formation of secondary Dean vortices. Dec denotes the condition associated with a decrease in the area covered by the primary vortices, and increase in the area covered by the secondary vortices. Dec was determined from the results of the experiments in flow visualization for each of the ARs of 0.2, 0.4 and 0.6. For De>Dec, we observed a steady increase in the area covered by secondary vortices till it stabilized to cover the area either equal to or lesser than the primary vortices as discussed in 38 the previous sections . We also observed that Dec has a strong inverse relationship with AR (Figure 14). This order of dependence is consistent with the classical studies done by Sugiyama et 68 al. For AR < 1, their absolute values of critical De were different (l00-300), but the DeC in his work was also indirectly proportional to AR. They concluded that the Dec reaches a minimum value when AR nears 1, which agrees with the results presented herein. The presence of multiple vortices beyond the Dec affects the focusing of particles and cells in spiral microchannel as discussed in the following section. Effects on particle focusing The presence of multiple vortices for De>Dec has important effects upon the focusing of particles and cells in spiral microchannels. Of particular importance are the secondary flows Figure 14. Plot of the critical dean number as a function of the aspect ratio of the rectangular channel. This plot provides a threshold for operation in the flow regime with primary Dean vortices. 39 which are associated with secondary Dean vortices, as they affect particle focusing. Their overall effect is to shift lateral positions of the particles from the convex wall towards the concave wall of the channel. This phenomenon was observed using diluted blood samples. Within such curved channels at lower flow rates, RBCs initially focus closer to the convex wall, as previously reported by the present investigators, as well as by other investigations. The cells then shifted towards the concave wall as flow rate increased above 2 mL/min. This observation, coupled with confirmation of the presence of secondary Dean vortices near the concave wall, suggested that the cells became entrained within secondary Dean vortices at the higher flow rate values. To investigate the dynamics of the entrapment, a solution of 10 µm diameter fluorescent polystyrene particles was introduced into a spiral device with AR = 0.6 (g1) at low De = 17 and high De = 37. It was observed that the particles get trapped in the additional vortices and hence focus near the concave wall (Figure 15). This trapping of particles is evident from the fluctuation of the width of the focused stream. At lower De, the particles focus near the inner channel wall and the full width at half maxima (FWHM) of the intensity scan across the channel was ~21 µm. As the flow rate was increased, De increased to 37 and the particles start focusing closer to the concave wall. This focusing position was in the form of a fluctuating band. The focusing regime shifts between a narrow focused stream to a broad band. The FWHM of the broad band was 42 µm indicating trapping of the particles in the secondary vortices. To confirm the process of trapping, confocal images were taken at high De and the trapping events were quantified by measuring the intensity of the trapped particles in one of the additional vortices. Figure 15c shows the stacked image along downstream (x-axis) so that multiple events of trapping can be overlapped to determine the approximate position of the trapped particles. Since the flow rate was very high (~3 mL/min), only intensity spikes were obtained rather than 40 Figure 15. Effect of secondary Dean vortices on particle entrapment. (a) Schematic of the process of entrapment of particles in the additional vortices with the insets of the corresponding particle focusing positions and the behavior of RBCs. (b) Intensity plot of the focused stream of 10 µm diameter particles at low De and trapping at high De. (c) Stacked confocal image of the events involving trapping of particles near the outer channel wall/concave wall. (d) Schematic of the positions where the confocal images were taken to determine the position of trapped 10µm particles. (e) Intensity plot across half height and width of the channel to determine the movement of particles in the secondary vortex. 41 complete fluorescent signal. The intensity was measured across the channel at five positions in the bottom half of the channel, spanning one of the lower secondary Dean vortices (Figure 15d). The intensity peaks at each position describe the subsequent motion of the particles which get trapped within the vortex being measured (Figure 15e). The intensity peaks show that the particles get trapped in the vortex and recirculate. In fact, the distance of the position of intensity peaks from the concave wall is comparable to the one obtained from the streak velocimetry images (~150-230 µm). From this observation, we can also conclude that the particles get trapped in the secondary Dean vortices. As such, particle positions reorganized as they were advected by secondary vortex motion. It can also be concluded that this trapping is not a very stable regime of particle focusing, as was evident from the following experiments. Overall, the presence and evolution of multiple vortices within the channel beyond a critical De suggests an explanation for the focusing behavior of the cells/ particles within the present investigation. This work confirms the presence of multiple Dean vortices in low-aspect ratio, rectangular spiral microchannels, which is consistent with observations at specific Dean numbers in many curved macroscale channels. These secondary vortices also provide justification for the entrapment of particles, and uncharacteristic focusing of particles, which are observed at higher De. The effects of multiple vortices on particle focusing have been observed in the form of a shift in the equilibrium position of particles from the inner/convex channel wall towards the outer/concave channel wall. This shift happens in three stages. The first stage involves evolution of primary vortices that encompass the whole channel at De 42 As flow rate increases and De approaches DeC, a transitional stage occurs which causes particles to focus in two positions near the center of the channel (a vertical overlap). The third stage is when particles focus near the outer channel wall as they get trapped in the secondary vortices stage (De>DeC). We observed these stages for both 10 µm and 15 µm particles as they were introduced in the spiral microchannel device. The rate at which these stages are reached is highly dependent on particle size since FD α ap (particle diameter). This is evident from the intensity scan for each stage for both 10 µm and 15 µm particles (Figure 16). A flip of focusing positions was observed when the individual streak velocimetry positions of the focused streams of particles for each of the particle sizes was overlapped. An alternate explanation of this behavior is provided by Toner et al.40, 72 who denoted the shift of equilibrium positions towards the concave wall to interactions between increased wall- lift forces and Dean forces. This explanation is generally applicable when particles are located closer to the center of the channel. However, it does not fully explain shifts in particle focusing which occur in close proximity to the concave wall. The transitional stage is a very complex regime to study, especially since streak velocimetry does not provide the cross-sectional positions of the focused particles. For a mixture containing both 15 and 10 µm diameter particles, at Q = 0.9 mL/min, the 15 µm particles focus closer to the inner channel wall, followed by a stream of focused 10 µm particles as is predicted by the previous model of two counter-rotating vortices (Figure 17a). As expected, the larger particle focusses closer to the inner channel wall followed by the smaller sized particle. As the De is increased beyond the critical De, the development of the secondary flow into secondary Dean vortices starts affecting the particles. At Q = 2.3 mL/min, the 10 µm particles are still in 43 Figure 16. Effect of increase in flow rate on particle focusing. Intensity scans across the channel width for 10 µm (a) and 15 µm (b) particles in all the three stages. The insets show the pseudo-colored fluorescent images of the positions of the focused streams of particles in the channel in their respective stages. 44 their transition stage and focus in the center of the channel when observed using an inverted microscope, but the 15 µm particles focus closer to the outer channel wall (Figure 17b). Since both the transition position and the trapping of particles are unstable positions, there was high amount of fluctuation leading to a very small separation distance between the focused streams. Hence, for a sorting regime, De>Dec yields unstable focusing positions near the concave wall. Note, that the line scans for Figure 17 are taken at a different position in the loop as compared to the line scans in Figure 16. Hence, there is a slight shift in the focusing positions for each of the particles. The effect of multiple vortices on cell focusing was already demonstrated using 500× diluted blood introduced into the channel. Beyond DeC, cells experience overwhelming Dean drag that pushes cells into the secondary vortices leading to their entrapment. Hence, RBCs initially focused closer to the inner channel wall, shift toward the outer wall with increased flow rate as shown earlier in insets in Figure 15a. Figure 17. Effect of secondary Dean vortices on particle mixture. Intensity scans for the mixture of 10 µm and 15 µm particles at 0.9 mL/min (a) and 2.3 mL/min (b). 45 Summary The present work confirms the presence of multiple Dean vortices in low-aspect-ratio, rectangular spiral microchannels, which is consistent with observations at specific Dean numbers in many curved macro-scale channels. The sequential development of secondary Dean vortices depends upon the gradient of the Dean number in the spiral micro-channel, as well as the radius of curvature, and the aspect ratio of the channel. The observed vortex development is consistent with the macroscale investigation which is reported by Sugiyama et al.68 In both macro-scale and microscale investigations, a dimensionless parameter, critical De, provides a threshold for the formation of secondary Dean vortices. These secondary vortices also provide justification for the entrapment of particles, and uncharacteristic focusing of particles, which are observed at higher De. Even though flow streamline determination is challenging using cross-sectional visualization images from spiral microchannels, images from the present investigation demonstrate the presence of the additional secondary Dean vortices within these channels. This development directly affects particle focusing within spiral/curved microfluidic devices. As such, the assumption of only two Dean vortex pairs is no longer valid for all experimental conditions. The present study of the effects of secondary flows and Dean instabilities thus provides new insight for manipulation of multiple vortices and other inherent fluid forces to achieve higher efficiency in fluid mixing, particle separation, and heat transfer. This work also paves way into understanding the reason behind the need for a certain range of De for cell focusing in sorting techniques, thereby providing a working window for successful cell sorting in spiral microchannels. 46 CHAPTER 3 PLASMA EXTRACTION AND BLOOD CELL SORTING Introduction Enrichment and separation of blood is often the first step in blood analysis, as each component of blood provides critical information for both diagnostics and therapeutics.12,73 For example, diseases like anemia, hemolysis, thalassemia, spherocytosis are diagnosed by determining physiological and quantitative changes in RBCs levels. WBCs play a significant role in the immune system and abnormal levels of WBCs or any form of deformation of WBCs indicate immune disorder, infection or blood cancer. It is not surprising that the most common tests in clinical diagnostics involve sorting blood components, such as CBC (complete blood count), clotting test, blood-chemistry test and blood enzyme test. In addition to clinical diagnostics, sorting of blood cells is also important in cell-research and therapeutics. Centrifugation is the “gold standard” and is the most common method of blood cell separation. While the approach is fairly simple and is successful in separating blood components, it can lead to contamination of the sorted levels during extraction and may cause lysing of blood cells.10,74 In addition to compromised purity, it cannot be easily intergraded with on-chip blood analysis systems. Another common approach to sorting blood cells is flow cytometry75,76, which is while effective, but is time consuming and can require multiple passes for full analysis. Expensive instrumentation and the need of skilled personnel make this approach 47 less desirable. Microfluidics has the potential to offer techniques for sorting of blood components with designs that are sufficiently compact for integration into point-of-care systems for clinical diagnostics.8,12,73 This work describes optimized spiral inertial microfluidic devices which utilize geometry induced, hydrodynamic forces for size based sorting along with providing reduced size and improved performance for separation of blood cells from plasma, as well as separation of WBCs and RBCs. Previous spiral designs77 had a focusing length of 40 cm with an inner radius of 1cm which resulted in a large device footprint (~3 in2). These devices were designed with the assumption that De is constant over the length of the spiral, which is not the case with spirals as there is a continuous change in radius of curvature that leads to progressive decrease in De. The optimized spiral designs are not only 10× smaller in size as compared to our previous design, but also provide higher efficiency (>95%) and throughput (1-2 mL/min with ≥0.1% hct). The reduced form factor and the low-aspect ratio nature make these devices more amenable to integration with lab-on-chip systems and to fabrication by high-throughput techniques such as roll-to-roll processing. Device principle and optimization Sorting cells in rectangular microchannels is dependent on the intricate balance of hydrodynamic forces. Shear gradient arising from parabolic velocity profile of Poiseuille flow induces inertial lift forces (FS) that cause cells or neutrally buoyant microparticles to migrate towards channel walls. As they migrate closer to channel walls, they are repelled away from walls towards the channel center due to wall induced lift forces (FW). As discussed in the introduction, a cell suspended in a spiral microchannel experiences a transverse Dean drag force due the two major Dean vortices and the inertial lift forces due to the 48 rectangular cross section. Assuming Stokes drag, the Dean force FD experienced by the cells or particles while travelling through a spiral channel is given by −4 1.63 퐹퐷 = 3πµ푈퐷ap = 5.4 × 10 휋µ퐷푒 푎푝 (4) -4 1.63 where 푈퐷 is the average Dean velocity (UD = 1.8×10 De ), ap is the diameter of the particle and µ is the viscosity of the fluid. Since most of the spiral devices use low aspect ratio channels, the rotational lift force can be neglected and it can be assumed that the inertial focusing stays in transition between first and second stage of the two stage migration model in straight channels with the net lift force (FL) defined as 2 4 퐹퐿 = 휌퐺 퐶퐿푎푝 (5) where ap is the cell (or particle) diameter, CL is the lift co-efficient, and G is the shear rate that is dependent on flow velocity and characteristic length. The scaling of FL is highly dependent on whether the lift coefficient is constant or scaling with particle/cell size. Recently, Zhou et al.42 -2 2 have experimentally proven that in high aspect ratio channels, CL α ap which leads to FLα ap . But, in case of low aspect ratio spiral channels, CL can be considered a constant since the Dean vortices push the particles/cells towards the walls, especially the inner channel wall since the flow generates from the center towards the walls causing the recirculation. The ratio of the inertial lift and Dean drag forces is strongly dependent on cell size, 3 scaling as FL/FD α CL ap . Thus, the magnitude of the Dean force provides mechanisms to adjust the focusing position and the focusing length for the particles. The two major counter-rotating Dean vortices in a spiral microchannel can then be used to manipulate the focusing position of the cells. These vortices displace the cells from the multiple focusing positions due to the imbalance between the net Dean force and the net lift force. The result is that cells focus in a single stream near the inner channel wall depending on their size, with the largest focusing 49 closest to the inner wall (Figure 18a). This size dependence of the ratio of inertial lift forces and Dean drag forms the basis of the design of spiral devices for size-based separation of particles and cells. Figure 18b illustrates focusing in a 500µm × 100 µm Archimedean spiral microchannel. The fluorescent intensity line-scan indicates the progressive focusing of a sample containing 20 µm diameter polystyrene particles. At 900 µL/min flow rate, it was observed that the particles focus in a tighter stream, almost a single-particle stream by the time the particles reach loop 4. Figure 18. Effect of primary Dean vortices and inertial lift forces on particle focusing. (a) Schematic illustrating the effect of curvature on the focusing positions. Larger particles focus in a single position closer to the inner channel wall and an appropriate outlet system can enhance the collection of the particles/cells sorted according to their sizes. (b) Intensity plot of 20 µm diameter particles across the width of the channel at the end of each loop in the spiral (loop1 being the inner-most loop and loop4 being the outer –most). The two inset figures show the fluorescent images of the 20 µm polystyrene particles at the inner most loop (loop1) of the spiral device (500 µm×110 µm) and at loop4 focused in a single stream near the inner channel wall. 50 The calculated FWHM at loop 1 was ~340 µm which indicates distribution of the particles almost across the entire width of the channel. As particles move downstream, they experience Dean drag along with the inertial lift forces causing progressive focusing closer to the inner channel wall. By loop 4, the FWHM reduces to ~24 µm indicating focusing in a single stream, one particle after the other. As particles focus in a single stream near the inner channel wall, with the largest being closest to it, an outlet system can be used to collect the separated particles. We first optimized the spiral design for a single central input port. With an input port centrally located, as the radius of channel curvature increases with downstream progression the magnitude of Dean flows gradually reduces. The tight curvature at the channel input then dictates the maximum De experienced by the cells in the channel. In this work, the inner-most radius of curvature R = 2 mm which is 5× smaller than previous designs, permitting increase in Dean number to De = 17, depending of input flow conditions. Decreasing R further could increase strength of Dean flows, and ultimately would permit separation of smaller cells and particles, but the size is limited by the need to accommodate the input port. Since Archimedean spiral design allows for a more uniform gradient of change in De. This design equation is used to determine the radius of curvature at each θ = π/4 and to calculate the focusing length downstream as the De changes with change in radius of curvature. To focus particles in a single stream, it is necessary that the lift forces balance the secondary forces from Dean vortices acting on the particle. This permits us to determine 1.84 focusing conditions by equating FL and FD, leading to 퐷푒 α 푎푝 . Figure 19 illustrates numerical optimization results for two sets of device dimensions. There is an exponential decrease in the Dean number with respect to the increase in downstream length in a spiral and 51 Figure 19. Numerical optimization of spiral device design. (a) Optimization results for Design1 (250 µm×75 µm). Dean number decreases as b (0.159, 0.0795) increases, although there is not much change in the order of the curve, except for the amplitude. The residual for each case is denoted by the dotted plot of the respective color. The point of zero residual gives the approximate focusing length required for forming a spiral with initial radius, a = 2 mm and the given dimensions of the rectangular channel. (b) Optimization results for Design 2 (500 µm × 110 µm) with b the same as the one taken for the previous device . The change in cross-section changes the hydraulic diameter, thereby changing the order of the exponential decrease in De with respect to increase in downstream length. 52 the point of zero residual provides the approximate focusing length for the particles satisfying the condition, ap/Dh > 0.07. The focusing length for each point in the loop curvature was calculated as De decreased from center (input) to the outermost loop (outlet). This calculation was based on the data calculated for Re = 100, leading to a general optimization for the particles satisfying the aforementioned conditions. The design equations can be formulated based on the numerical design model and empirical data of the optimization. The optimal length of focusing of particles satisfying the ap/Dh condition was then found to be dependent on the gradient of the change in De along the length of the spiral. For each of the two designs, we obtained the following differential equations describing the dependence of the variation of the De number on the length downstream. 휕퐷푒 −2.39 −1.136 = [ ] [푥 ] (6) 휕푥 −3.55 푥−1.171 where x is the downstream length of the spiral. Eq. 6 was derived from the data of Figure 19. We used MS Excel to fit data, analyze regression co-efficients (2.39, 3.55) and determined mathematical expressions describing the data trends for the 2 different designs. A common design equation was then empirically formulated to suit the two numerical cases. The single design equation which describes downstream length of the spiral (x) required to focus the cells or particles for an optimized Dean number Deop 퐷 퐿 − ℎ 푚 퐿 퐷푒표푝~3.5 푥 푚 (7) 퐷ℎ where Lm is the Dean migration length. Dean migration length (Lm) is the maximum distance the particle has to travel to reach closer to the inner channel wall and reach the equilibrium position. 3푊 퐿 = 푊 + 퐻 + (8) 푚 4 53 where H is the height of the channel and W is the width of the channel. The constant, 3.5, and focusing length, x, are in mm. Cells of larger size require a shorter downstream length to focus as compared to cells of smaller size. The optimized De can be calculated using the following equation. 푊 푎푝 퐷푒표푝~3.3 (퐷푒푖 ) (9) 퐷ℎ 퐷ℎ where, ap/Dh ratio is taken to be 0.07 and Dei is the Dean number for the inner-most diameter which in this case is usually <15 (to ensure primary Dean vortices to maintain two counter rotating vortices to ensure particle focusing regime) resulting in the expression Dei (ap/Dh) = 1. The resulting optimized De equation can be reduced to 푊 퐷푒표푝~3.3 (10) 퐷ℎ The focusing length of the particles in a spiral microchannel can be calculated from equation (9) by adjusting the optimized Dean number. Any increase in spacing between the channels decreases the Dean number, thereby increasing the focusing length for the particle. If s (spacing between the adjacent channel center-line) is kept constant, change in the width of the channel accordingly modifies the rate of the exponential decay in De with respect to the downstream length. The focusing length is the point at which the residual is determined to be either equal to or less than zero. Residual is the difference between the focusing length calculated with the assumption that the radius is constant at that particular point and the circumference/total length of the spiral arc using the same radius. Since both the constant-radius focusing length and the circumference/arc change in length at each point in the spiral due to the progression in the radius of curvature, the point at which the difference between the two is zero determines the actual focusing length. For separation of blood cells from plasma, we used the 250 µm × 75 µm device 54 with a spacing of 250 µm between the loops and focusing length of ~6 cm. Although the platelets do not quite meet the ap/Dh criterion for focusing, yet, we selected this specification for the device as it should at least result in partial focusing of the platelets. For separation of WBCs and RBCs, a broader channel with a longer focusing range was needed to focus cells into distinct streams, within the length of the spiral. To achieve that, we used 500 µm × 110 µm devices with a spacing of 500 µm between the loops (more convenient for soft-lithography process) and the focusing length of ~8 cm. The focusing length determines the total length of the channel when unwound from its spiral coil into a straight channel. This length also determines the number of loops a spiral device will need to have in order to accommodate the required length for cell-focusing. Hence, focusing length is the key factor in determining the total size of the device if the size of the inlets and outlets are fixed. Compared to previous Archimedean spiral devices, we have reduced the focusing length ~5× and reduced the device area 10×. This substantial decrease in size allows the ease of fabrication of these devices using fabrication techniques other than soft lithography (PDMS devices). The reduced size of these devices makes them amicable to fabrication by large-scale manufacturing methods such as roll-to-roll processing78. Indeed, we have successfully fabricated these devices in PET and PMMA in collaboration with VTT Technical Research Centre of Finland using is approach. Ultimately, this could lead to inexpensive, disposable devices. Experimental methods Fabrication Microchannels were fabricated in polydimethysiloxane (PDMS, Sylgard 184, Dow Corning) using the conventional soft lithography process. Masters 75-110 µm thick were formed 55 in SU-8 photoresist (2075, Microchem Corp.). A mixture of PDMS base and curing agent (10:1 ratio) was cast on masters and cured for 4 h on a 60 °C hotplate. Cured PDMS devices were peeled off, and inlet/outlet ports were punched with a 14 gauge syringe needle. PDMS was bonded to standard glass slide using a hand-held plasma surface treater (BD-20AC, Electro- Technic Products Inc.). Two fabricated designs included a three outlet system for with a 250 µm × 75 µm channel cross-section (design 1) and a four outlet system with a 500 µm ×110 µm channel cross-section (design 2). Particle and blood experiment protocols To demonstrate device performance, fluorescently-labeled polystyrene particles 7.32, 10, 15 and 20 µm in diameter (Bangs Laboratories) were used. A particulate mixture at 0.1% volume fraction was used at flow rates <3 mL/min. We first loaded a syringe with particle solution and connected it to the device by using a 1/16” peek tubing (Upchurch Scientific) with proper fittings (Upchurch Scientific). We drove the syringe with a syringe pump (NE-1000, New Era Pump Systems, Inc.). For fluorescent particle stream images, we used an inverted epi- fluorescence microscope (IX71, Olympus Inc.) equipped with a 12-bit high-speed CCD camera (Retiga EXi, QImaging). We stacked ~100 images and added pseudo-colors to form fluorescent particle-stream pictures in ImageJ. For experiments with blood, we first performed a dilution test with 0.9% saline solution. Whole blood from anonymous female donors (Hoxworth Blood Center) was diluted 2× to 2000×. Aliquots of dilute blood were run through devices. Samples collected at the outlets were centrifuged at 1,000rpm for 10min and stained with Wright-Geimsa stain to identify cell types collected. The stained cells were counted using a hemocytometer (Hausser Scientific). 56 Device characterization The developed devices were characterized using neutrally-buoyant fluorescently-labeled polystyrene particles (Polysciences, Inc.). Since the sorting mechanism is dependent on the size of the cell/particles in the flow stream, we used particles with diameters similar to the blood cells. The flow parameters were optimized by injecting polystyrene particles ~7.32, 10, 15 and 20 µm in diameter (1.6×105 particles/mL) at flow rates 1-3 mL/min. The Dean number was calculated using the input Re and Dh values (width and height for Dh calculation were measured post fabrication) since the microchannel cross-section is constant. The radius of the curvature, R was measured from the center to the inner-most wall at every 30º interval on each of the loops. When particle mixture was injected at 1 mL/min in design1 device, all particles focused into a broad band near the inner channel wall. At the outlet bifurcation, they all followed the flow, eluting in outlet1 (Figure 20). As the size of RBCs and WBCs is in the range of sizes of particles, it was concluded that this device not only provides the confirmation of the hypothesis for Figure 20. Optimization of flow parameters for plasma extraction. (a) Image of design1. (b) Bright-field image of the outlet system. (c) A mixture of 7.32 µm, 10 µm, 15 µm and 20 µm diameter fluorescently labeled particles is injected at the flow rate of ~1 mL/min in the device. (d) Almost all the particles focus in a broad stream near the inner channel wall eluting in outlet1. (e) Flow cytometer results for the particles collected at the outlets. 57 isolation of blood cells from plasma, but also shows that this separation may be accomplished with a high level of efficiency (~100%). Next, we characterized design2 devices for sorting RBCs and WBCs. The devices based on design2 focused particles into two streams at a constant flow rate of 1.8 mL/min. The particles in the broad stream (10, 15, 20 µm) eluted in the first outlet and the particles from the narrow stream, closer to the center of the channel, eluted in the second and third outlet (Figure 21). Further increase in the flow rate to 2.2 mL/min disrupted the focusing of 7.32 µm diameter particles, but further focused the particles in the broad stream into three distinct streams of 10, 15 and 20 µm diameter particles with 20 µm diameter particles focused closer to the inner channel wall. These results further emphasize that the equilibrium position of the focused cells is strongly dependent on their size, as well as flow properties and channel geometry. These results also show that by maintaining proper flow conditions and a four outlet Figure 21. Optimization of flow parameters for blood cell sorting. (a) Image of design2. (b) Bright field image of the outlet system. (c) At 1.8mL/min flow rate (F1), two focused streams are observed, the narrow stream in the middle of the channel is formed by 7.32 µm particles and the broad stream near the inner channel wall is the composite of three streams of 10, 15 and 20 µm particles. Particles in the range of 10-20 µm are obtained out of the first outlet, and the 7.32 µm particles are obtained from the second and third outlets. (d) Fluorescent image of the focused streams of all three particles, 10 µm, 15 µm and 20 µm in diameter, at the flow rate of 2.2 mL/min (F2). (e) Normalized focusing position of particles (x is the distance of the focused stream from the inner channel wall, and w is the width of the channel) as function of De. 58 pattern, RBCs, which are ~7 µm in diameter, should elute in the second and third outlets and WBcs, which are in the range of 10-20 µm, should elute in the first outlet of the device. After determination of optimal conditions and definitive flow parameters, the devices were tested with blood. Approximately 10 mL of male whole blood sample (45% hematocrit/hct) was provided by Hoxworth Blood Center (University of Cincinnati Academic Health Center). Since blood is a viscous, non-Newtonian fluid, behavior of blood cells does not follow to that particles as determined during the optimization experiments of devices. Hence, to work with blood, it was necessary to first evaluate the effect of blood dilution on separation efficiency in our microchannels based on hydrodynamic forces acting on the cell–components of blood. Blood dilution Blood rheology affects the functionality and efficiency of all passive microfluidic sorting. Blood is a non-Newtonian fluid and the viscosity is highly dependent on hematocrit, plasma protein concentration, platelet count and leukocyte count.12,73,74,79 The plasma concentration also affects the cell to cell interaction in blood. In passive microfluidic devices which are based on hydrodynamic forces acting in a Newtonian, poiseuille flow, blood is required to be diluted for the forces to be strong enough to cause cell-separation.12 Spiral microchannels not only depend on hydrodynamic forces, but also on dean vortices to effectively sort cells according to their size. Although the shear induced lift force and wall induced lift force are dependent on the parabolic velocity profile and Newtonian nature of the fluid to cause cell-separation, dean vortices are known to act in highly viscous fluids with vortex formation. This differential effect of blood rheology on the dean vortices and hydrodynamic forces affects the sorting efficiency and throughput in spiral microchannels. 59 Although the secondary flows manifest themselves in viscous fluids, their amplitude and point of instability and balance with inertial lift forces is not as well defined when optimizing the devices using particle experiments. Also, the neutral buoyancy of cells and cell to cell interaction poses a significant contention to the sorting ability of the optimized devices. To test the particle behavior as a function of blood-dilution, whole blood was diluted 10 to 700x times (4.5% hct to 0.07%) with 0.9% saline solution, and each sample was used in both devices. The width of the focusing streams of cells was observed at each dilution (Figure 22a). To determine the sorting efficiency, the samples collected at the outlets (in 2 mL vials) were centrifuged at 1,000 rpm for 10 min. The centrifuged samples were stained with Wright-Geimsa stain to identify the type of cells collected. The staining was done as per the standard procedure of the use of the WG-stain. The stained cells were then counted using hemocytometer. At 10-20× dilution, the sheer concentration of cells overwhelms the focusing position of cells. It is clear that higher the dilution better is the focusing of the cells along with precise extraction of the separated cells. Figure 22a shows the effect of dilution on focusing of the cells in a 250 µm × 75 µm spiral channel. The plot in Figure 22 describes the width of the focused cells near the inner channel wall for each dilution at 1 mL/min flow rate proving the assumption that lower dilution provides more precise focusing. The higher dilution causes blood to behave like Newtonian fluid and it also prevents cell to cell interaction providing better focusing (Figure 22b). The separation efficiency is shown to increase with dilution especially in the case of separation of RBCs and WBCs, as stronger inertial force is required for precise size-based separation. Although at lower dilution, RBC-RBC interaction is still observed in the form of cells bouncing off of each other especially due to cell concentration, the spacing between the focused streams is not disrupted and the efficiency of sorting is not compromised. Although the lateral 60 Figure 22. Effect of blood dilution on sorting efficiency. (a) Bright Field images of blood cells focused in the outermost loop of an Archimedean spiral (250 µm×75 µm) with 10, 50, 100 and 200 fold dilution. (b) A log-plot of normalised width of the focused stream (δ is the width of the focused stream and w is the width of the channel) with the dashed curve defining the plot for design1and the solid curve defining the plot for design 2. (c) Plot of separation efficiency of extraction of plasma in 250 µm×75 µm spiral device as a function of dilution of the whole blood. 61 force generated by cell deformability of RBCs may cause the broadening and pulsation of the focused band of RBCs since deformability may cause cells to elongate within the flow leading to change in focusing positions depending on whether the cells are focusing along the shorter axis or longer axis. These results are in agreement with the work done by Toner et al.80, who showed decrease or shift in focusing of polystyrene particles, WBCs and PC-3 cells with change in the volume fraction in the blood sample from 1 (45% hct) to 0.07 (3.15% hct). Although plasma extraction is possible at low dilution, higher efficiency of separation in case of RBCs (Figure 22c) and WBCs is only possible at higher dilution (~500×). Blood component sorting Plasma extraction Following the determination of the flow parameters required for the separation and appropriate dilution, the 100× diluted blood (0.45% hct) was injected into the devices with design1 at a constant flow rate of ~1 mL/min. It was observed that no cells eluted in outlet three, showing that the third outlet provided us with cell-free plasma. The first and second outlets however, had all the RBCs, WBCs and platelets eluting from it. Figure 23a-c show stained cells at the inlet of the spiral device, cells collected at the 1st and 2nd outlet and plasma collected at the 3rd outlet. Figure 23d shows the focusing of cells in a broad stream near the inner channel wall at the outer-most loop of the device. Although it was observed that cells bounce off of each other, their net movement was along the focused stream and this did not compromise the sorting efficiency. We surmise that this cell to cell interaction is more relevant in case of lower dilution due to sheer overwhelming of cells at the focusing positions. In this case however, the focused streams of cells with different sizes are close together such that they form a broad, visible band 62 of focused cells, all eluting in the same outlet. The cell to cell interaction is not strong enough to modify or disrupt the focusing of cells near the inner channel wall. This device provides ~100% cell-free plasma from the 100× diluted blood (0.45% hct). It is possible to achieve higher efficiency at lower dilution (~50×) by suitable modification of the outlet system. At 0.9% hct, we achieved ~90%±3% efficiency, which is equivalent to plasma extraction at the rate of 20µL/min of whole blood (45% hct). The efficiency and throughput is significantly higher than other plasma extraction techniques like the passive microfluidic device used for plasma extraction by Sollier et al.98 with 17% extraction yield with 1:20 dilution, electro-osmotic flow controlled device used by Jiang et al.79 which yielded 26% efficiency and two phase plug flowing through disposable tubing technique by Sun et al.56 with 10 × diluted blood yielding efficiency of ~64 % in 1 µL plug. Even the most common technique of centrifugation is extremely prone to Figure 23. Plasma extraction from diluted blood sample. (a) Bright-field image of blood cells distributed throughout the width of the channel at the inlet of design1. Arrows indicate the white blood cells. (b) All the cells, including platelets elute in the 1st and 2nd outlet. (c) No cells are obtained, only plasma from outlet 3. (d) Bright field image of the cells focused in a broad stream near the inner-channel wall, thereby eluting in the first and second outlet. Samples collected from each outlet were centrifuged and then stained and observed. 63 contamination while extraction of plasma from whole blood. Blood cell sorting To sort the RBCs and WBCs, the 500× diluted blood sample (~0.1% hct) was run through design2 devices at 1.8mL/min to ensure lower cell to cell interaction and higher effects of inertial lift forces and dean vortices. It was observed that the WBCs focused closer to the inner channel wall owing to their larger size and RBCs focused closer to center of the channel, allowing the collection of the cells by the use of the four outlet system. The cells at each outlet were collected, centrifuged, stained and counted using a hemocytometer. Cells collected from the first outlet had 95%±2.2% of WBCs and 6% ± 2.4%of RBCs, with a few platelets (Figure 24b). This corroborated the results from the particle experiments where the 10, 15 and 20 µm diameter particles eluted in the first outlet. WBCs being in the range of 10-20 µm focused closer Figure 24. Blood cell sorting from diluted human blood. (a) Bright field images of the stained samples after they were collected from each outlet of the design2 and centrifuged. Inlet has all the cells present. Arrows indicate the white blood cells. (b) Outlets 2 and 3 have RBCs and platelets (c) Outlet1 has majority of WBCs (Neutrophils, Eosinophil, and Monocyte), some platelets and very little RBCs. and (d) outlet 4 has only diluted plasma and platelets. 64 to the inner channel wall and eluted in the first outlet. The sample collected from the second and third outlet had 94%±2.5% RBCs (Figure 24c). The fourth outlet had only platelets floating in diluted pale yellow liquid or plasma. The hemocytometer counts confirmed the separation efficiency of ~95% ±3%, with a high throughput (~107 RBCs/mL and ~105 WBCs/mL), as shown in Figure 25. This translates to a throughput of ~3.6 µL/min of whole blood (45% hct) which is comparable to the recent blood sorting work by Han et al.55 where they showed >90% efficiency of enrichment of WBCs. Although platelets could not be separated, there is an almost complete separation of leukocytes and erythrocytes. This is significantly higher than the leukocyte extraction by DEP microseparator14,15 which had efficiency of 92%, but throughput of 0.8 µL/min, hydrodynamic filtration with 29× enrichment of leukocytes with a throughput of 20 µL/min with 10× dilution of whole blood and leukocyte extraction using magnetophoresis17,19 with efficiency of 97%, but throughput is 0.04 and 0.33 µL/min. Further, since the throughput and efficiency was calculated Figure 25. Hemocytometer results with normalized cell count. Cells were counted at each outlet, showing ~ 90% efficiency of separation of RBCs from WBCs. 65 using only the viable cells collected at the outlets, the throughput is indicative of cell viability. As such blood cells are quite robust as shown in previous work by Toner et al.32 and Han et al.47 Summary In this chapter, we successfully demonstrated continuous separation of erythrocytes and leukocytes from a diluted sample of blood. Due to high separation efficiency (~95%) and high throughput (1-2 mL/min of ≥0.1% hct), these spiral microfluidic devices offer a label-free alternative to a key step in CBC analysis. These devices solve the issue of trade off which has been faced by earlier microfluidic devices along with providing label-free sorting which has been difficult to achieve especially in case of active devices. The considerable reduction in size allows these devices to be integrated with on-chip blood-cell analysis or plasma analysis systems without any sample contamination or sample loss. These passive devices provide both higher sorting efficiency and higher throughput as compared to microfluidic sorting techniques like microfluidic filtration8,9 and magnetophoretic devices17,19 and hence can be used as a sorting platform for point of care blood-analysis devices. Further optimization of these devices can enable a WBC differential with minimal sample volume and on integration with an on-chip cell- counter, can potentially provide a CBC-on-chip. Other spiral have also been used for cancer stem cell sorting as shown by Sun et al.56 with ~90% efficiency. Hence, our design which has the ability to sort cells with size in the same range could possibly be used for other more size selective applications like cancer stem cell sorting. Blood dilution is one of the challenges that need to be further addressed. Although >100× dilution was used in the experiments discussed, blood dilution can be as low as 20× if very high efficiency of sorting is not a requirement. This option of variable dilution allows the use of these devices in not only commercial but laboratory platform. Due to passive mechanism of separation 66 simple planar structure and cost effectiveness, these devices are easy to use and disposable. Ultimately, these devices offer a pathway towards development of a “blood-on-a-chip” point-of- care system with high efficiency, low cost, and reduced analysis time. 67 CHAPTER 4 ISOLATION OF PCA STEM CELL SUB-POPULATIONS Introduction Prostate cancer is an adenocarcinoma of the prostate gland. There are no specific symptoms to diagnose prostate cancer except through standard-of-care digital rectal exams or tissue biopsy to confirm a diagnosis of cancer.81-83 In recent years, there has been a rise in the frequency of prostate cancer diagnosis in men. A recent study from the U.S. Military Cancer Institute reported a significant rise in prostate cancer (PCa) in military personnel, especially in 20-59 year old men84. Similarly, firefighters also show an increased risk of PCa 15 to 20 years earlier than the average male population.85,86 These reports suggest that the work environment may increase the risk of PCa in some occupations. Effective treatment for advanced PCa remains challenging. Initially, most patients with advanced disease respond to androgen deprivation therapy (ADT); however 100% of these patients will develop treatment resistant PCa and succumb to their disease.86-89 Hence, it is critical to understand the underlying mechanisms underlying metastatic spread and the emergence of therapeutic resistance. The cancer stem cell (CSC) model proposed by Cohnheim and Durante in 1875 indicated that tumors contained a reservoir of self-renewing cells that maintained the tumor characteristics and were resistant to therapy.82,90-92 Similarly, it is generally 68 thought that prostate CSCs are self-renewing cells that promote tumor development, support metastasis, and underlie the processes that give rise to treatment-resistant disease.82,88-90 Androgens are the primary growth factor for prostate cancer development, progression and metastasis. Since the bulk of tumor cells (>99%) are androgen receptor (AR) positive (+) they remain androgen-sensitive cells and are initially eliminated by ADT.85,86,93 However, a rare subpopulation of cells within a tumor are AR (-) CSCs and therefore believed to be resistant to ADT. While CSCs comprise of only ~0.1% of total tumor cells, they possess unique characteristics essential for promoting tumor growth, metastasis and the development of castration-resistant prostate cancer (CRPC). The first characteristic is self-renewal (or the ability of creating itself over and over again). The second property is the ability to differentiate into the heterogeneous prostate cancer cell types which make up the bulk of tumor cells.83,94 The biology of CSCs is still not fully understood since it has been difficult to isolate CSCs from solid human tumor biopsies obtained as surgical waste material and analyze their characteristics in vitro.82,89,95 The two CSC-like cell lines used in this work, HPET (Human prostate epithelial-TERT) and HuSLC (Human stem cell- like cells) are therefore well-suited for studying the biology of CSCs in vitro and analyzing their characteristics to identify biomarkers which could potentially be used as targets in the treatment of CRPC .81,88,95 Conventional cell sorting techniques like FACS and MACS have been previously used to sort these cells. These techniques use either sheath flow or magnetic beads labeled with antibodies of cell surface markers respectively. When Dr. Kasper’s lab attempted to sort HPET- CSCs using the commercial FACS at CCHMC (Cincinnati Children’s Hospital Medical Center), the rate of viable HPET cell recovery was 0% (Figure 26). Therefore, they searched for a new method which would yield viable CSC-like HPET cells for further analysis. As described below, 69 Figure 26. Sorting of HPET cells using FACS. (a) Bright field image of the sample of HPET cells sent to CCHMC for FACS sorting. (b) Size estimation of the set of cells. (c) Zero viability obtained at the outlet of the FACS sorter. the spiral microfluidic devices are suitable for CSC sorting and can provide a cost-effective solution with not only the prospect of high viability, but also high efficiency of separation in a single pass. Sorting protocol and methods Sorting protocol As described in the previous chapter, the working principle of spiral sorting device exploits the inherent hydrodynamic forces caused by parabolic profile of a Poiseuille’s flow and the Dean drag caused by curvature of the device.54,77,96 Counter rotating Dean vortices interact with the net hydrodynamic/inertial lift forces (FL) to focus cells/particles in a single stream close to the inner channel wall/convex wall with the largest particle focusing closest to the channel wall. 42,48,54,55 This concept of size based sorting was used to design spiral microfluidic sorting devices for this application. The design and optimization of the devices was done as described in Chapter3. However to take heterogeneous cell size distribution into consideration, additional 70 loops were added to the device. The devices were operated at flow rates such that De < DeC and only primary vortices are responsible for focusing (1-2 mL/min). Beyond DeC, vortices transition into secondary Dean vortices which can destabilize the focusing regime. To visualize the cultured cells, they were stained with a fluorescing dye, CellTracker™ Green CMFDA (5-chloromethylfluorescein diacetate) CMFDA (3mM), for 30 minutes. Cells were then trypsinized (1 mL Trypsin) and re-suspended in DMEM-F12 (+KOSR-Knock-Out Serum) medium to a concentration of 2,500 cells/mL (Figure 27). The suspended cells were passed through the spiral device at the optimized flow rate to sort them according to size. The optimized flow rate was defined as the flow rate at which maximum viability of cells was obtained which will be discussed in more detail in the next section. The sorted cells were collected from the individual outlets. To standardized volume across all outlet collections, the cells in each sample were centrifuged, and re-suspended in 1.5 mL serum free medium. Cells were counted using the Trypan Blue viability test and size estimation was done using hemocytometer and Image pro-plus. Since the range of cell sizes may vary from plate to plate, cells from one 10cm plate were used for one set of experiments to ensure that cell confluency and the range of cell size remained consistent throughout in a given set of experiments. A plate of cells was considered optimally confluent when the sheet of epithelial cells exhibited a cobblestone morphology without any spaces between them or where the culture plate bottom was visible. Cell culture protocols The sorting protocol required culturing the cells on a 10 cm PrimariaTM cell culture plate coated with matrigel. This form of 3D-cell culture allows the cells to adhere and proliferate. The Kasper lab has generated two CSC-like cell lines from human prostate cancer biopsy specimens 71 (Gleason Score 9) derived from unrelated patients, the HPET cell line and a novel HuSLC (Human Stem Cell-like Cell) line.86,88 Both HPET cells and HuSLCs exhibit an AR(-) prostate stem cell phenotype and express stem cell markers. HPET cells and HuSLCs were cultured for 5 days (or the optimum confluency) in DMEM-F12 (+KOSR-Knock-Out Serum) medium with 10 µL bFGF (basic Fibroblast Growth Factor). bFGF is a growth factor which supports the maintenance of undifferentiated human stem cell-like cells and promotes their proliferation. For the purpose of sorting doublets and triplets from single HPET cells, a confluent plate of cells was trypsinized and cells were re-plated at a concentration of 2500 cell/ml on ultra-low adhesion 6- well plates for 1 day to allow the formation of doublets and triplets and maintain the presence of single cells. LNCaP cells were purchased from the American Type Culture Collection (ATCC) and used to represent the bulk of tumor cells which do not exhibit CSC-like characteristics. LNCaP cells are human prostate adenocarcinoma cells derived from a left supraclavicular lymph node metastasis. These cells express AR, require androgens for viability and cell growth, and will undergo apoptosis in response to ADT. LNCaP cells were cultured on 10 cm glass plates in RPMI medium supplemented with 10% fetal bovine serum and 1× penicillin/streptomycin Figure 27. Schematic showing the protocol of sorting of cells in the spiral device from trypsinizing and suspending the cells to cell count after sorting. 72 DU-145 cells were purchased from ATCC and used to represent non-CSC, human prostate cancer cells which do not express AR. They were derived from a metastatic brain lesion and do not require androgens for viability. They were cultured on standard 10cm culture plates in RPMI medium supplemented with 10% fetal bovine serum and 1× penicillin/streptomycin The following sections discuss the details of each of the cell assays used to determine the effects of inertial sorting on cell viability, proliferation and function. The assay used to assess LNCaP and DU-145 cell function was the dual luciferase reporter assay which measures transcriptional activity. The assay used to assess HPET and HuSLC cell function was the sphere formation assay which measures the ability of stem-like cells to undergo self-renewal. Assessment of cell viability and proliferation after separation Viability Cell viability is a key parameter for cell sorting devices. It is defined as the number of cells that remain intact and viable after the cells have been passed through the device and sorted by size. Since the shear force can affect cells depending on their type and morphology, CSCs and bulk tumor cells were tested for viability for a range of flow rates (0.5 mL-3 mL/min.). HPET, LNCaP and DU-145 cells were removed from the culture plate by trypsinization, counted and diluted down to 5000 cells/mL. Prior to sorting, cells were counted at the inlet using the Trypan Blue viability test and hemocytometer. In this test, viable cells exclude the dye and remain clear while dead cells take up the dye and become blue. The number of viable cells was determined by the standard formula for hemocytometer count: number of viable cells = (number of viable cells in 1mmx1mm square) x dilution factor x 104.97 After sorting, the cells were collected, centrifuged and re-suspended. 73 The re-suspended cells were then stained with trypan blue and viable cells counted as above. To evaluate the efficiency with which the device could isolated viable cells, the sorted cell count was normalized to the initial cell count at the inlet for each cell type and flow rate. As shown in Figure 28, cell viability was differentially modulated over a range of flow rates according to cell type. This differential modulation was dependent on both the flow rate and the cell type. Of note was that HPET cell viability was found to be extremely sensitive to flow rate. It increased from 50% at Q = 0.5 mL/min to 95% at Q = 1-2 mL/min. At 0.5 mL/min, the resident time is increased and therefore cells are exposed to shear rate for a longer period of time as compared to cells sorted at a flow rate of 1-2 mL/min. Therefore, the low viability of HPET cells at lower flow rate (Q = 0.5 mL/min) could be attributed to the increase in the residual time that the cells spend in the iMF device and consequently to shear rate for a longer period of time. Although the shear rate is ~13,000/s, it is considerably reduced when compared to other inertial microfluidic sorters such as the vortex trapping sorter by Sollier et al.98 with a shear rate of 78,000/s and MARCS sorter by Zhou et al.50 with a shear rate of 55,000/s. Figure 28. Plot of viability of HPET, LNCaP and DU-145 cells determined over a range of flow rates. 74 Unlike HPET cells, LNCaP and DU-145 cells were not as sensitive to changes in flow rate. Taken together, these results suggest that cancer stem cell-like cell (HPET) viability could be differentially modulated by shear as compared to cancer cells which comprise the bulk of the tumor (e.g., LNCaP and DU-145 cells). Furthermore, it suggests that flow and shear rates need to be considered when designing devices for isolating and sorting different tumor cell populations. Sorting Using the optimized spiral device with a cross section of 250 µm × 100 µm, cells were suspended (2500 cells/mL) and sorted at Q = 2 mL/min since this was the flow rate at which maximum HPET cell viability was obtained. Larger HPET cells (>17 µm) focused closer to the inner channel wall as seen in the pseudo-colored image in Figure 29a. Cells smaller than 17 µm focused near the center of the channel eluting in the second and third outlet. The fourth outlet is used mainly for the purpose of balancing the flow in the channel and it did not contain any cells. Cells eluting from the individual outlets were then collected and counted (Figure 29b). HPET cells were sorted with an efficiency of >85% and throughput of 5000 cells/min. In addition, approximately 98% of sorted cells remained viable, demonstrating that the spiral device configuration supported the efficient isolation of viable cells. . As described in the sections below, the sorted cells were further analyzed to evaluate the effects of passing cells through a spiral device on cell proliferation and function. Proliferation Although the cell viability test measures whether cells remain ‘alive’ after sorting, it is also necessary to determine whether the cells have retained other normal characteristics. For example, CSCs are defined by their ability to proliferate, self-renew and differentiate. Therefore 75 Figure 29. Sorting of PCa cells using spiral sorter. (a) The pseudo-colored image of the focused larger cells near the inner wall of the spiral microchannel and eluting in outlet 1. (b) Normalized cell count of the sorted cells. it is essential that these characteristics be maintained following sorting, especially if CSCs are subjected to further analyses. For bulk tumor cells, it is essential to determine whether the hydrodynamic forces, which cells are subjected to during the sorting process , affect their ability proliferate and respond to hormone and/or drug treatment. Therefore our next approach was to evaluate any changes in the rate of cell proliferation before and after sorting (Figure 30). The proliferation assay monitors the rate at which cell number increases over a defined period of time. LNCaP and HPET cells were cultured for 0 to 120 h following sorting. The 5,000 cells were plated in triplicates in 24 well plates for each case (sorted and un-sorted). Cells were trypsinized and counted using the Trypan Blue viability test as described above. The number of viable cells over time was then plotted to determine the proliferation rates for both the cell lines. Figure 30 shows the proliferation rates for both un-sorted and sorted cells for each of the cell lines. The results obtained clearly indicate that the proliferation rate of both hormonally sensitive, bulk tumor cells (LNCaP cells) and CSC-like cells (HPET cells) is retained after sorting. Since the curve defined by the proliferation of sorted cells is virtually identical to that of 76 Figure 30. Effect of sorting on rate of proliferation. (a) Proliferation assay data over a period of 120 h for LNCaP cells. (b) Proliferation assay data over a period of 72 h for HPET cells unsorted cells, it suggests that the shear rate and Dean drag have minimal effect on cell cycle progression. Following the proliferation study, we performed luciferase assay to determine the effects of inertial sorting on androgen-mediated transcription in LNCaP and HPET cells. Assessment of cell functionality after separation Androgen-mediated transcription Androgens not only promote PCa cell growth, they also regulate AR-mediated transcription in prostate cancer bulk tumor cells, including LNCaP and DU-145 cells. To determine the effects of sorting on AR-mediated transcription, we performed the Promega Dual- Luciferase® Reporter Assay on both sorted and unsorted LNCaP cells. Briefly, LNCaP cells were transfected with an expression vector containing the firefly luciferase reporter gene linked to the androgen-regulated probasin promoter ARR2PB as well as with the Renilla luciferase control vector.99 Cells were treated with/without the androgen dihydrotestosterone (DHT, 10-8 M) and/or the antiandrogen Casodex (CSX, 10-5 M) and firefly luciferase/Renilla luciferase control activity was determined according the manufacturer’s instructions as described below. The same 77 assay was performed using HPET cells with one key modification. Since HPET cells do not express AR, they were transfected with an expression vector, pSVoAR, which enabled them to express the human AR. HPET cells were then treated with DHT and the anti-androgens CSX or hydroxyflutamide (OHF). In the luciferase assay, DHT was used as a positive control to promote androgen regulated reporter gene transcription. The antiandrogens Casodex and/or hydroxyflutamide (OHF) were used to inhibit androgen-mediated transcription. Of note is that LNCaP cells express a T877A mutation in AR.83,86,95 While Casodex inhibits mutant AR activity, OHF activates mutant AR-regulated transcription. Therefore, LNCaP cells were only treated with Casodex and not OHF (Figure 31a). In contrast, HPET cells express pSVoAR which is the “wild-type” (or normal) androgen receptor. Therefore HPET cells were treated with CSX and OHF as indicated in Figure 31b. Each experiment was performed on cells derived from the same transfection plate to ensure that the levels of transgene expression were consistent within a given experiment. This was achieved by transfecting the plasmids into cells cultured on one large 10 cm plate. After transfection, cells were cultured for 24 to permit transgene expression. Cells were trypsinized, counted, and 5x104 cells/well were plated in 24 well culture plates and allowed to attach overnight. Cells were treated for 24 h with/without DHT and with/without Casodex or OHF as indicated in Figure 31 and then harvested to measure luciferase activity. The dual luciferase assay is based on the measurement of bioluminescence of the firefly luciferase which is then normalized to Renilla luciferase activity. Since Renilla luciferase activity is not hormonally regulated, it serves as an internal control. The harvested cells were lysed using Passive Lysis Buffer (Promega). After centrifugation, the cleared lysate was 78 Figure 31. Effect of sorting on androgen-mediated transcription. (a) Luciferase expression in LNCaP cells treated with DHT and Casodex (CSX). (b) Luciferase expression in HPET cells treated with pSVoAR, DHT and OHF. transferred to a fresh tube and protein concentration was determined prior to reporter enzyme analyses.86,87 Each set of experiments was repeated at least twice and the data in each column represents the mean +/- standard error of three replicate wells (Figure 31). The results of the luciferase assay demonstrated that in both sorted and unsorted LNCaP cells, treatment with DHT upregulated luciferase activity; and addition of Casodex inhibited DHT-induced luciferase activity (Figure 31a). Similarly, luciferase activity increased in response to DHT treatment in both sorted and unsorted HPET cells expressing the wild-type AR; and addition of either Casodex or OHF effectively inhibited DHT-induced luciferase activity (Figure 31b). Since the responsiveness to androgen and antiandrogen treatment was similar in both sorted and un-sorted cells, these observations indicate that AR-regulated transcription in prostate CSCs is not altered by the process of sorting viable cells using spiral devices. Sphere formation assay CSCs exhibit the properties of self-renewal and the ability to differentiate into the 79 heterogeneous prostate cancer cell populations that comprise a tumor. 85,90,95 The sphere formation assay is performed to identify cells which exhibit the stem cell property of self- renewal. This assay was first used by Reynolds and Weiss who discovered that neuroepithelial stem cells could proliferate while in suspension and form free-floating spheres, called neurospheres.91,92,100 Since then, the sphere formation assay has been widely used for determining whether a cell type has stem cell-like characteristics. Only cells with stem cell characteristics grow in suspension and form floating spheres (Figure 32). Since non-stem-like epithelial cells require attachment to a basement membrane, they will die when cultured in suspension. The Kasper lab has generated two CSC-like cell lines from prostate cancer biopsy specimens (Gleason Score 9) derived from two unrelated patients. Both the HPET (Human Prostate Epithelial-TERT) cell line and a novel HuSLC (Human Stem Cell-like Cell) line exhibit an AR (-) prostate stem cell phenotype and express stem cell markers. Therefore, these cell lines were used to represent AR (-) treatment-resistant PCa CSCs. In the sphere formation assay, HPET cells and HuSLCs survive and form spheres in suspension, indicating that they are stem cell-like. Since inertial microfluidic devices are well suited to sort cells according to size, we Figure 32. Schematic representing the concept behind sphere formation assay. In suspension, cells with stem cell like properties form spheres, whereas cells without any stem cell like properties die. 80 postulated that our spiral device will be able to sort out different tumor cell sub-populations, including cancer stem cell subpopulations, by size. To test this hypothesis, HPET cells were plated on 10 cm Primaria plates coated with Matrigel (Figure 33a). The cells were trypsinized, and diluted to 2,500 cells/mL. HPET cells were then sorted in the spiral device into sub-population 1 (>17 µm) and sub-population 2 (10-17 µm). The sorted cells were then centrifuged, re-suspended to the concentration of 2500 cells/mL and plated in ultra-low adhesion 6-well plates (n = 3) to allow sphere formation. After 14 days, the spheres were analyzed and counted for each case: control-unsorted, outlet1-sub-population1 and outlet 2&3-sub-population2. Spheres that formed from un-sorted cells showed three distinct Figure 33. Evaluation of stem-ness of sorted HPET cells. (a) Confocal image of the HPET cells plated on 10cm Primaria plate coated with matrigel, before sorting. Sphere formation assay is performed on the un-sorted (control) and sorted (sub-population 1(outlet1) and sub-population1 (outlet 2&3)) HPET cells. After 14 days, the cells in each case were imaged and counted. Control/un-sorted cells had large (b), medium (c) and small (d) spheres. (e) Spheres from cells eluting in Outlet 1 had medium sized spheres, (f) Spheres from cells eluting in Outlet 2&3 were smaller. (g) The spheres in each case were counted using Image J and plotted. Error bars were calculated using n = 3. 81 sizes, including large, medium and small spheres (Figure 33b-d). The spheres that formed from larger sub-population 1 HPET cells (Outlet 1) were medium size and morphologically different from those of control (Figure 33e). Spheres which formed from smaller sub-population 2 (Outlets 2&3) were medium and small size and morphologically very similar to those of Control (Figure 33f). Figure 33g shows the plot of the number of spheres in each case. Smaller HPET cells (sub-population 2) formed more spheres as compared to larger HPET cells (sub-population 1), although both the sub-populations retained their characteristics of self-renewal. A similar trend was observed when the sphere formation assay was performed using sorted and un-sorted HuSLCs (Figure 34). It was observed that they formed morphologically Figure 34. Evaluation of stem-ness of sorted HuSLCs. (a) Confocal image of the HuSLCs plated on 10cm Primaria plate coated with matrigel, before sorting. Sphere formation assay is performed on the un-sorted (control) and sorted (sub-population 1 (outlet1) and sub-population2 (outlet 2&3)) of HuSLCs. After 14 days, the cells in each case were imaged and counted. (b) Spheres that formed in Control were very dense and large in size as seen in the zoomed-in image (c). (d) Spheres from cells eluting in Outlet 1 had medium sized spheres, (e) Spheres from cells eluting in Outlet 2&3 were smaller. (f) The spheres in each case were counted using Image J and plotted. Error bars were calculated using n = 3. 82 denser and higher number of spheres in the same conditions as HPET cells. More spheres were formed by the smaller sub-population of HuSLCs (Outlet 2&3) as compared to the larger HuSLC sub-population (Outlet1). Additionally, as observed in the case of HPET cells, the morphology of spheres from outlet 1 cells was observed to be distinctly different from spheres from outlet 2&3. This indicated that although both the sub-populations retain their characteristics of self- renewal, the degree of their self-renewal and differentiation could possible differ. Doublet and single-cell sorting of HPET cells While HPET cells provide insight into the role of CSCs in prostate cancer tumorigenesis, cancer progression and metastasis, one of the critical challenges of working with these cells is separating cells undergoing cell division (cell doublets and triplets). These doublets could potentially provide insight into understanding the process of reconstitution of a tumor from undifferentiated CSCs. Since these doublets and triplets are considerably larger than the single cells, size-based sorting is an effective way to isolate single cells from doublets and triplets. As described earlier in the chapter, the equilibrium position of the focused cells depends on their size and flow rate. In this particular section, a 6-loop Archimedean spiral with 500 µm × 100 µm rectangular channel was used to demonstrate the separation. An 8-outlet system was used for this purpose. Previous work by other groups and our observations has suggested that cells with aspect ratio >1, i.e., non-spherical cells focus along their major axis. In fact, doublets and discoid shaped cells like RBCs, tend to tumble and rotate along their major axis in the focusing stream (Figure 35a). The inset in Figure 35a shows the aggregated doublet 20 µm diameter particles rotating along their major axis as they travel downstream one of the loops of the spiral. 83 Figure 35. Sorting of doublet and single HPET cells. (a) Image of the spiral microfluidic device used to sort the doublets from single HPET cells. (b) Fluorescent image of the focused streams of HPET cells near the inner channel wall (as indicated by the arrows)-FITC filter. (c) Bright-field image of the unsorted cells at the inlet of the device (20× objective). (d) Bright-image of the larger cells-doublets and triplets that eluted in Outlet 1. (e) Bright-field image of the smaller single cells eluting in outlets 2 and 3(f). (g) Normalized cell count of the sorted doublet and single cells in each of the outlets. As described above, HPET cells were stained with 3 mM CMFDA, a fluorescent living dye used to image the focused streams of cells within the device, and separated at a flow rate of 2 mL/min (~5,000 cells/min). Due to the heterogeneous distribution of cell sizes, the cells focused in bands near the inner channel wall (Figure 35b) and were collected at the outlets depending on their focusing positions with respect to the inner channel wall. A bright field image of the cells at the inlet are shown in Figure 35c, indicating the range of sizes that the doublets and single cells encompass. The larger doublets (>20 µm) eluted in the first outlet closest to the channel inner wall (Figure 35d). Smaller single cells (~15 µm) eluted in the second and third outlets closest to the channel inner wall (Figure 35e, f). Although the HPET 84 cells are quite fragile, the low flow rate in the 6-loop Archimedean spiral prevented the cells from lysing, leading to 100% viability at the outlets. This work demonstrates that spiral inertial microfluidic devices can be used to sort CSCs in different phases of division, whether it is proliferation or differentiation. The doublets are indicative of the cancer cells with stem cell-like properties and them being in the process of division into either a progenitor cell or a differentiated cell. The sorted cells can be further used to study the cancer stem cell niche and the factor that affect tumorigenesis. Summary This chapter reported on the use of spiral iMF devices for size-based sorting of bulk tumor cells and CSCs. The bulk tumor cells retained their viability, proliferation rates, and hormonal responsiveness after sorting. The CSCs also remained viable after sorting, and retained their ability to proliferate. Furthermore, sorting of CSCs showed the presence of different sub- types within CSC-like cell populations. These findings suggest that cell size can be used as a marker for “stem-ness”; and that stem-like subpopulations are heterogeneous. In conclusion, this work successfully demonstrates that spiral iMF devices are an effective laboratory tool for sorting epithelial cells, which are larger and more fragile to fluid shear forces. Importantly, the different cells remained viable, proliferated, and retained their physiological functions, including hormonal responsiveness and/or “stem-ness”. Therefore we suggest that iMF devices are well suited to analyze all of the different tumor cell sub-populations found in a given tumor. This analysis will bring us one step closer to determining the interactions of bulk tumor cells and CSCs in cancer development, progression and metastatic spread. 85 CHAPTER 5 CONCLUSIONS Summary The concept of lab-on-chip systems constitutes integration of a variety of processes from sample preparation to sample analysis on a single chip. One of the most significant steps in sample analysis is cell sorting, especially, where cell biology research and therapeutics is concerned. The conventional cell sorting techniques like flow cytometry (FACS and MACS) require specialized equipment and training.16,50,75,101,102 The sheath flow in flow cytometry affects the viability of fragile cells like CSCs. In fact, when flow cytometry was used to sort human prostate epithelial-TERT (HPET) cells, the sorted cells were no longer viable. Inertial microfluidic sorting devices addresses the need for cost-effective, label-free, integrable sorting system which provides high viability at the output. In this work, fluid flow dynamic and cell sorting regime was investigated in spiral inertial microfluidic channels. The spiral devices were then designed and optimized for cell sorting. The sorting approach takes advantage of the principles of inertial microfluidics and Dean drag forces in spiral microchannels to sort cells based on size with high efficiency and throughput. Previous work6,53 on curved inertial microfluidic devices along with spiral microchannels have been based on two counter-rotating Dean vortices. This work provides the first experimental proof that it is possible for the flow to evolve into multiple vortices across the 86 channel beyond a critical De. Simulations (Star-CCM+) and confocal microscopy were used to determine the parametric conditions in which multiple vortices arise. In a 3-inlet, one-outlet system spiral devices with channel dimensions 250 µm × 150 µm (w x h), it was observed that there is a sequential development of the number of vortices as the dye moves downstream towards the center of the spiral. These secondary vortices affect the cell focusing regime by means of entrapment of cells in the secondary Dean vortices near the outer channel wall. This process of trapping has been quantified by measuring the intensity of the trapped particle in one of the additional vortices using confocal microscopy. This phenomenon explained the uncharacteristic focusing behavior of the cells/ particles in the channel. The effect of multiple vortices on cell focusing was first observed when a 500× diluted blood sample was introduced in the channel and the RBCs focused closer to the outer channel wall instead of the inner channel wall. This work improves the understanding of the concept of particle focusing in the spiral/curved devices and assist in manipulation of the interaction of multiple vortices and other inherent fluid forces to achieve higher efficiency and selectivity in cell sorting. This work also paves way into understanding the reason behind the need for a certain range of De for cell focusing in sorting techniques. After determining the range of operational and design parameters for spiral sorting devices, a numerical and empirical model was created based on primary counter-rotating Dean vortices and the variation of De as the radius of curvature changes within the spiral device. The resulting device design had 1/5th size compared to the previous spiral sorting devices and had higher sorting efficiency. This work reports on successful isolation of plasma and separation of blood cells with high throughput (1-2 mL/min) and high separation efficiency (>90%). The developed approach caters to the need for blood cell sorting devices that can be integrated with 87 an on-chip analysis system and rapidly provide separated sample with high purity. Devices for this work were fabricated using by roll-to-roll embossing of PMMA and standard soft lithography. Roll-to-roll processing provides a step towards mass production of microfluidic devices. These devices can be disposable and low cost. A tumor consists of heterogeneous cell populations. These cell populations include the bulk of tumor cells which are responsive to therapy and a small subpopulation of cells called CSCs which have the capacity to regenerate and replace tumor cells which have bene eliminated by therapy. 83,91,92 The net outcome of this replacement is the emergence of therapeutic disease.88,89 Effective cell sorting is central to sample preparation and to identifying and characterizing cell populations that exhibit different tumor cell populations. In this work, spiral inertial microfluidic sorting devices were designed for sorting different types of tumor cells including the bulk of hormonally-responsive tumor cells and prostate cancer stem cells. HPET cells and HuSLCs derived from tissue biopsy of high grade, Gleason 9 score prostate cancer biopsy, were used to represent the unique characteristics of CSCs found in the clinical human prostate cancer tissue biopsies. The single focusing position in spiral inertial device enabled the size-based sorting and isolation of sub-populations of CSCs with high efficiency and throughput. Cell viability and cell proliferation of bulk of prostate cancer cells (LNCaP and DU-145 cells) were retained following sorting, demonstrating that sorting of cells using the spiral inertial device does not alter the cellular response to androgen and antiandrogen treatment in both LNCaP and HPET cells. In addition, the sphere formation assay was performed on sorted sub- populations of HPET cells and HuSLCs to measure the ability of the sorted cells to undergo self- renewal. These results indicate that prostate CSCs contain sub-populations of cancer stem-like cells and that the spiral inertial microfluidic device can be used to selectively sort and isolate 88 these sub-populations. In conclusion, the use of spiral devices would allow us to sort and further analyze the interactions of bulk tumor cells and CSCs in cancer development, progression and metastatic spread. This knowledge could be used to develop more effective treatments that target specific tumor cell populations. Future directions Spiral devices investigated in this work offer a powerful and efficient approach to cell sorting. One of the major challenges in studying the fluid behavior in spiral microchannels was imaging of the cross-sectional development of the flow. The development of secondary Dean vortices was observed at very high flow rate. The only way to image the formation of these vortices in the cross-section was confocal microscopy and contrast by using a 1/3rd confinement of the fluorescein dye. Since spiral microchannels are efficient mixers, combined with the fact that confocal image capture is not fast enough to capture steady state images for a very high flow rate, the images were not very clear near the inner loops. It was especially difficult to image the focusing position of particles within the transitionary phase and the trapping of the particles in the secondary Dean vortices. Although, this work has provided considerable insight into the mechanics of fluid behavior and cell focusing in spiral microchannels, there is scope for further analysis of mechanics of sorting in spiral devices. Furthermore, the development trend of secondary Dean vortices in high aspect ratio spiral channels will probably be very different from the low aspect ratio channels discussed in this work. Overall, this work not only explains the behavior of particles when the flow in spiral microchannels evolves from primary vortices to secondary vortices, but also paves way into a deeper understanding of focusing states in curved 89 microchannels. This development directly affects the concept of particle focusing in the spiral devices along with development of highly efficient spiral sorters. 90 REFERENCES 1. Kumar A, Galaev IY, Mattiasson B. Cell Separation: Fundamentals, Analytical and Preparative Methods. Springer Berlin Heidelberg, 2010 2. Orfao A, Ruiz-Argüelles A. General concepts about cell sorting techniques. Clin Biochem. 1996;29:5-9. 3. Schor S. Cell Separation:Methods and Selected Applications/Vol 2. Br J Cancer. 1984; 50:571-571. 4. Shapiro HM. Practical Flow Cytometry. Wiley; 1994. 5. Tomlinson MJ, Tomlinson S, Yang XB, Kirkham J. Cell separation: Terminology and practical considerations. J Tissue Eng. 2013;4:1-14. 6. Bhagat A, Bow H, Hou H, Tan S, Han J, Lim C. Microfluidics for cell separation. Med Biol Eng Comput. 2010;48:999-1014. 7. Shields, CW. Reyes, CD. López, GP. Microfluidic Cell Sorting: A Review of the Advances in the Separation of Cells from Debulking to Rare Cell Isolation. Lab Chip. 2015;15:1230. 8. Whitesides GM. The origins and the future of microfluidics. Nature. 2006;442:368-373. 9. Gao Y, Li W, Pappas D. Recent advances in microfluidic cell separations. Analyst. 2013;138:4714-4721. 10. Yu ZTF, Aw Yong KM, Fu J. Microfluidic Blood Cell Sorting: Now and Beyond. Small. 2014;10:1687-1703. 11. Pamme N. Continuous flow separations in microfluidic devices. Lab Chip. - The Royal 2007;7(12):1644-59. 12. Toner M, Irimia D. Blood-on-a-Chip. Annu Rev Biomed Eng. 2005;7:77-103. 13. Hu X, Bessette PH, Qian J, Meinhart CD, Daugherty PS, Soh HT. Marker-specific sorting of rare cells using dielectrophoresis. Proc Natl Acad Sci U S A. 2005;102:15757-15761. 14. Voldman J. Electrical Forces for Microscale Cell Manipulation.Annu Rev Biomed Eng. 2006;8:425-54. 91 15. Wang L, Flanagan LA, Jeon NL, Monuki E, Lee AP. Dielectrophoresis switching with vertical sidewall electrodes for microfluidic flow cytometry. Lab Chip. 2007;7:1114-1120. 16. Chen A, Byvank T, Chang W. On-chip magnetic separation and encapsulation of cells in droplets. Lab Chip. 2013;13:1172-1181. 17. Huang Y, Joo S, Duhon M, Heller M, Wallace B, Xu X. Dielectrophoretic Cell Separation and Gene Expression Profiling on Microelectronic Chip Arrays. Anal Chem.2002; 74(14):3362- 3371. 18. Huang LR, Cox EC, Austin RH, Sturm JC. Continuous Particle Separation Through deterministic Lateral Displacement. Science. 2004;304:987-990. 19. Cheng F, Chang H, Hou D, Chang H. An integrated dielectrophoretic chip for continuous bioparticle filtering, focusing, sorting, trapping, and detecting. Biomicrofluidic. 2007; 1:021503. 20. Melville D, Paul F, Roath S. Direct magnetic separation of red cells from whole blood. Nature. 1975;255: 706-706. 21. Zborowski M, Chalmers JJ. Rare cell separation and analysis by magnetic sorting. Anal Chem. 2011;83: 8050-8056. 22. Yamada M, Kano K, Tsuda Y, et al. Microfluidic devices for size-dependent separation of liver cells. Biomed Microdevices. 2007;9:637-645. 23. Jakobsson O, Grenvall C, Nordin M, Evander M, Laurell T. Acoustic actuated fluorescence activated sorting of microparticles. Lab Chip. 2014;14:1943-1950. 24. Lenshof A, Laurell T. Continuous separation of cells and particles in microfluidic systems. Chem Soc Rev.2010; 39:1203-1217. 25. Patel MV, Nanayakkara IA, Simon MG, Lee AP. Cavity-induced microstreaming for simultaneous on-chip pumping and size-based separation of cells and particles. Lab Chip. 2014;14: 3860-3872. 26. Joná A, Zemánek P. Light at work: The use of optical forces for particle manipulation, sorting, and analysis. Electrophoresis. 2008;29:4813-4851. 27. Lau AY, Lee LP, Chan JW. An integrated optofluidic platform for Raman-activated cell sorting. Lab Chip. 2008;8:1116-1120. 28. Neuman KC, Nagy A. Single-molecule force spectroscopy: optical tweezers, magnetic tweezers and atomic force microscopy. Nat Meth. 2008;5:491-505. 29. Huang LR, Cox EC, Austin RH, Sturm JC. Continuous Particle Separation Through Deterministic Lateral Displacement. Science. 2004;304:987-990. 92 30. Sajeesh P, Sen AK. Particle separation and sorting in microfluidic devices: a review. Microfluidics and Nanofluidics. 2014;17:1-52. 31. Seo J, Lean MH, Kole A. Membrane-free microfiltration by asymmetric inertial migration. Appl Phys Lett. 2007;91:033901. 32. Yamada M, Nakashima M, Seki M. Pinched Flow Fractionation: Continuous Size Separation of Particles Utilizing a Laminar Flow Profile in a Pinched Microchannel. Anal Chem.2004; 76(18):5465-5471 33. Yamada M, Seki M. Nanoliter-sized liquid dispenser array for multiple biochemical analysis in microfluidic devices. Anal Chem. 2004; 76(4):895-9. 34. Yamada M, Seki M. Microfluidic Particle Sorter Employing Flow Splitting and Recombining. Anal Chem. 2006; 78:1357-1362. 35. Wei Hou H, Gan HY, Bhagat AAS, Li LD, Lim CT, Han J. A microfluidics approach towards high-throughput pathogen removal from blood using margination. Biomicrofluidics. 2012;6: 024115-024115-13. 36. Nagrath S, Sequist LV, Maheswaran S, et al. Isolation of rare circulating tumour cells in cancer patients by microchip technology. Nature. 2007; 450:1235-1239. 37. Cheng I-, Huang W-, Chen T-, Liu C-, Lin Y-, Su W-. Antibody-free isolation of rare cancer cells from blood based on 3D lateral dielectrophoresis. Lab Chip. 2015;15:2950-2959. 38. Patil P, Madhuprasad, Kumeria T, Losic D, Kurkuri M. Isolation of circulating tumour cells by physical means in a microfluidic device: a review. RSC Adv. 2015;5:89745-89762. 39. Hyun K, Lee TY, Jung H. Negative Enrichment of Circulating Tumor Cells Using a Geometrically Activated Surface Interaction Chip. Anal Chem. 2013;85:4439-4445. 40. Martel JM, Toner M. Inertial Focusing in Microfluidics. Annu Rev Biomed Eng. 2014; 16:371-396. 41. Karabacak NM, Spuhler PS, Fachin F, et al. Microfluidic, marker-free isolation of circulating tumor cells from blood samples. Nat Protocols. 2014; 9:694-710. 42. Kuntaegowdanahalli SS, Bhagat AAS, Kumar G, Papautsky I. Inertial microfluidics for continuous particle separation in spiral microchannels. Lab Chip. 2009; 9(20): 2973-80. 43. Wu Z, Willing B, Bjerketorp J, Jansson JK, Hjort K. Soft inertial microfluidics for high throughput separation of bacteria from human blood cells. Lab Chip. 2009; 9(9):1193-9. 44. Amini H, Lee W, Di Carlo D. Inertial microfluidic physics. Lab Chip. 2014; 14:2739-2761. 45. Asmolov ES. The inertial lift on a spherical particle in a plane Poiseuille flow at large channel Reynolds number. J Fluid Mech. 1999; 381:63-87. 93 46. Eichhorn R, Small S. Experiments on the lift and drag of spheres suspended in a Poiseuille flow. J Fluid Mech. 1964; 20:513-527. 47. Di Carlo D, Irimia D, Tompkins RG, Toner M. Continuous inertial focusing, ordering, and separation of particles in microchannels. Proc Natl Acad Sci USA. 2007;104:18892-18897. 48. Di Carlo D. Inertial microfluidics. Lab Chip. 2009;9:3038-3046. 49. Aman Russom and Amit K Gupta and Sunitha Nagrath and Dino Di Carlo and Jon F Edd and,Mehmet Toner. Differential inertial focusing of particles in curved low-aspect-ratio microchannels. New Journal of Physics. 2009;11:075025. 50. Zhou J, Papautsky I. Fundamentals of inertial focusing in microchannels. Lab Chip. 2013;13:1121-1132. 51. Saffman PG. The lift on a small sphere in a slow shear flow. J Fluid Mech. 1965;22:385-400. 52. Russom A, Gupta A, Nagrath S, Carlo D, Edd J, Toner M. Differential inertial focusing of particles in curved low-aspect-ratio microchannels. New J. Phys.2009; 11:075025. 53. Bhagat AAS, Kuntaegowdanahalli SS, Papautsky I. Continuous particle separation in spiral microchannels using dean flows and differential migration. Lab Chip. 2008;8:1906-1914. 54. Nivedita N, Papautsky I. Continuous separation of blood cells in spiral microfluidic devices. Biomicrofluidics. 2013; 7:054101. 55. Guan G, Wu L, Bhagat AA, et al. Spiral microchannel with rectangular and trapezoidal cross-sections for size based particle separation. Scientific Reports. 2013;3:1475. 56. Sun J, Liu C, Li M, et al. Size-based hydrodynamic rare tumor cell separation in curved microfluidic channels. Biomicrofluidics. 2013;7:11802. 57. Tse JM, Cheng G, Tyrrell JA, et al. Mechanical compression drives cancer cells toward invasive phenotype. Proc Natl Acad Sci U S A. 2011;109:911-916. 58. Jung W, Han J, Kai J, Lim J, Sul D, Ahn CH. An innovative sample-to-answer polymer lab- on-a-chip with on-chip reservoirs for the POCT of thyroid stimulating hormone (TSH). Lab Chip. 2013; 13:4653-4662. 59. Manlapaz RL, Churchill SW. Fully developed laminar convection from a helical coil. Chem Eng Commun. 1981;9:185-200. 60. Dean WR. XVI. Note on the motion of fluid in a curved pipe. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science. 1927;4:208-223. 61. Dean WR. LXXII. The stream-line motion of fluid in a curved pipe (Second paper). The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science. 1928;5:673- 695. 94 62. Cheng KC, Lin R, Ou J. Fully Developed Laminar Flow in Curved Rectangular Channels. Journal of Fluids Engineering. 1976;98:41-48. 63. Jarrahi M, Castelain C, Peerhossaini H. Secondary flow patterns and mixing in laminar pulsating flow through a curved pipe. Exp Fluids. 2011;50:1539-1558. 64. Ligrani PM, Longest JE, Kendall MR, Fields WA. Splitting, merging and spanwise wavenumber selection of Dean vortex pairs. Exp Fluids. 1994;18:41-58. 65. Ligrani PM, Niver RD. Flow visualization of Dean vortices in a curved channel with 40 to 1 aspect ratio. Phys Fluids. 1988;31:3605-3617. 66. Eustice J. Experiments on Stream-Line Motion in Curved Pipes. Proceedings of the Royal Society of London Series A. 1911;85:119-131. 67. Taylor AMKP, Whitelaw JH, Yianneskis M. Curved Ducts With Strong Secondary Motion: Velocity Measurements of Developing Laminar and Turbulent Flow. Journal of Fluids Engineering. 1982;104:350-359. 68. Sugiyama S, Hayashi T, Yamazaki K. Flow characteristics in the curved rectangular channels (visualization of secondary flow).Bull JSME. 1983;26:964-969. 69. Howell J,Peter B., Mott DR, Golden JP, Ligler FS. Design and evaluation of a Dean vortex- based micromixer. Lab Chip.2004;4:663-669. 70. Sudarsan AP, Ugaz VM. Fluid mixing in planar spiral microchannels. Lab Chip.2006; 6: 74- 82. 71. Picardo JR, Pushpavanam S. Laterally stratified flow in a curved microchannel. Int J Multiph Flow. 2015;75:39-53. 72. Martel J, Toner M. Inertial focusing dynamics in spiral microchannels. Phys. Fluids.2012; 24:032001. 73. Zborowski M, Ostera GR, Moore LR, Milliron S, Chalmers JJ, Schechter AN. Red Blood Cell Magnetophoresis. Biophys J. 2003;84:2638-2645. 74. Givan AL. The sorting of cells. In: Flow Cytometry: First Principles. John Wiley & Sons, Inc.; 2001; 2001:159-174. 75. Bonner WA, Hulett HR, Sweet RG, Herzenberg LA. Fluorescence activated cell sorting. Rev Sci Instrum. 1972;43:404-409. 76. Oakey J, Applegate RW, Arellano E, Carlo DD, Graves SW, Toner M. Particle Focusing in Staged Inertial Microfluidic Devices for Flow Cytometry. Anal Chem. 2010;82:3862-3867. 77. Kuntaegowdanahalli SS, Bhagat AAS, Kumar G, Papautsky I. Inertial microfluidics for continuous particle separation in spiral microchannels. Lab Chip. 2009;9:2973-2980. 95 78. Liedert R, Amundsen LK, Hokkanen A. Disposable roll-to-roll hot embossed electrophoresis chip for detection of antibiotic resistance genemecA in bacteria. Lab Chip. 2012; 12:333-339. 79. Parichehreh V, Estrada R, Kumar SS, et al. Exploiting osmosis for blood cell sorting. Biomed Microdevices. 2011; 13:453-462. 80. Lim EJ, Ober TJ, McKinley GH, Toner M.Visualization of microscale particle focusing in diluted and whole blood using particle trajectory analysis. Lab Chip.2012; 12(12):2199-210. 81. Santoni M, Piva F, Scarpelli M, et al. The origin of prostate metastases: emerging insights. Cancer Metastasis Rev. 2015; 34(4):765-73. 82. Kasper S. Stem cells: The root of prostate cancer? J Cell Physiol. 2008; 216:332-336. 83. Collins AT, Maitland NJ. Prostate cancer stem cells. Eur J Cancer. 2006;42:1213-1218. 84. Zhu K, Devesa SS, Wu H, et al. Cancer Incidence in the U.S. Military Population: Comparison with Rates from the SEER Program. Cancer epidemiology, biomarkers & prevention. 2009;18:1740-1745. 85. Kasper S, Cookson MS. Mechanisms Leading to the Development of Hormone-Resistant Prostate Cancer. Urol Clin North Am. 2006;33:201-210. 86. Kopras E, Potluri V, Bermudez M, Williams K, Belcher S, Kasper S. Actions of endocrine- disrupting chemicals on stem/progenitor cells during development and disease. Endocrine- Related Cancer. 2014;21:T1-T12. 87. Kerr CL, Hussain A. Regulators of prostate cancer stem cells. Curr Opin Oncol. 2014;26:328-333. 88. Kasper S. Identification, characterization, and biological relevance of prostate cancer stem cells from clinical specimens. Urologic Oncology: Seminars and Original Investigations. 2009;27:301-303. 89. Kasper S. Exploring the origins of the normal prostate and prostate cancer stem cell. Stem Cell Rev. 2008;4:329. 90. Gu G, Yuan J, Wills M, Kasper S. Prostate Cancer Cells with Stem Cell Characteristics Reconstitute the Original Human Tumor In vivo. Cancer Research. 2007;67:4807-4815. 91. Blau HM, Brazelton TR, Weimann JM. The evolving concept of a stem cell: Entity or function? Cell. 2001;105:829-841. 92. Al-Hajj M, Becker MW, Wicha M, Weissman I, Clarke MF. Therapeutic implications of cancer stem cells. Curr Opin Genet Dev. 2004;14:43-47. 93. Pfeiffer MJ, Schalken JA. Stem Cell Characteristics in Prostate Cancer Cell Lines. Eur Urol. 2010;57:246-255. 96 94. Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and cancer stem cells. Nature. 2001;414:105-111. 95. Chen X, Rycaj K, Xin Liu, Tang DG. New insights into prostate cancer stem cells. Cell Cycle. 2013;12:579-586. 96. Nivedita N, Giridhar PV, Kasper S, Papautsky I. Sorting human prostate epithelial (HPET) cells in an inertial microfluidic device. 15th International Conference on Miniaturized Systems for Chemistry and Life Sciences 2011, MicroTAS 2011. 2011;2:1230-1232 Accessed 15 August 2014. 97. Strober W. Monitoring cell growth. In: Current Protocols in Immunology. John Wiley & Sons, Inc.; 2001. 98. Sollier E, Go DE, Che J, et al. Size-selective collection of circulating tumor cells using Vortex technology. Lab Chip. 2014;14:63-77. 99. Tillman JE, Yuan J, Gu G, et al. DJ-1 binds androgen receptor directly and mediates its activity in hormonally treated prostate cancer cells. Cancer Res. 2007;67:4630-4637. 100. Pastrana E, Silva-Vargas V, Doetsch F. Eyes Wide Open: A Critical Review of Sphere- Formation as an Assay for Stem Cells. Cell stem cell. 2011;8:486-498. 101. Hulett HR, Bonner WA, Barrett J, Herzenberg LA. Cell sorting: Automated separation of mammalian cells as a function of intracellular fluorescence. Science. 1969;166:747-749. 102. Wu Z, Willing B, Bjerketorp J, Jansson JK, Hjort K. Soft inertial microfluidics for high throughput separation of bacteria from human blood cells. Lab Chip. 2009;9:1193-1199. 97