Continuous Manipulation and Characterization of Colloidal Beads and Liposomes via Diffusiophoresis in Single- and Double-Junction Microchannels

We reveal a physical mechanism that enables the preconcentration, sorting, and characterization of charged polystyrene nanobeads and liposomes dispersed in a continuous flow within a straight micron-sized channel. Initially, a single Ψ-junction microfluidic chip is used to generate a steady-state salt concentration gradient in the direction perpendicular to the flow. As a result, fluorescent nanobeads dispersed in the electrolyte solutions accumulate into symmetric regions of the channel, appearing as two distinct symmetric stripes when the channel is observed from the top via epi-fluorescence microscopy. Depending on the electrolyte flow configuration and, thus, the direction of the salt concentration gradient field, the fluorescent stripes get closer to or apart from each other as the distance from the inlet increases. Our numerical and experimental analysis shows that although nanoparticle diffusiophoresis and hydrodynamic effects are involved in the accumulation process, diffusio-osmosis along the top and bottom channel walls plays a crucial role in the observed particles dynamics. In addition, we developed a proof-of-concept double Ψ-junction microfluidic device that exploits this accumulation mechanism for the size-based separation and size detection of nanobeads as well as for the measurement of zeta potential and charged lipid composition of liposomes under continuous flow settings. This device is also used to investigate the effect of fluid-like or gel-like states of the lipid membranes on the liposome diffusiophoretic response. The proposed strategy for solute-driven manipulation and characterization of colloids has great potential for microfluidic bioanalytical testing applications, including bioparticle preconcentration, sorting, sensing, and analysis.


Introduction
Synthetic and natural nanoparticles are ubiquitous in a wide range of chemical, 1 bioanalytical, 2,3 biomedical 4-6 and environmental 7 applications.][10] Nanoparticle characterization in terms of size and surface properties -such as chemical composition and charge -is also key to several nanoparticle technologies.2][13][14] Size, surface charge and surface composition also dictate the nanoparticles' ability to penetrate through natural barriers, such as extracellular matrices 15,16 and the blood brain barrier. 17Consequently, control over particle size and surface properties is essential for precision drug delivery applications, in which these properties are tuned to increase circulation time, 18,19 achieve high therapeutic efficacy 20,21 and reduce toxicity. 22As a further example, extracellular vesicles -i.e., lipid vescicle naturally released and internalized by cells -has attracted much attention because of their potential as powerful therapeutic and diagnostic tools. 23Their size, surface charge and biochemical composition reflect their biogenesis and determine the cellular uptake pathways used for intercellular communication. 24Thus, the characterization of the size and surface properties of these lipid vesicles is necessary to elucidate the many physiological and pathological processes they are involved with, 25 and to exploit their potential as drug carrier and disease biomarkers.
In the last two decades, there has been a growing interest in microfluidic strategies for both particle manipulation 26,27 and characterization. 28,292][43][44][45] Furthermore, since the diffusiophoresis mobility of colloids can depend on particle size 35,41,46 and surface charge, [47][48][49] this transport mechanism has been exploited also for the detection and characterization these particle properties.A low-cost zeta potentiometry microchip was developed by relying on the fluid and particle motion induced within dead-end microchannels via transient salt concentration gradients. 49This microfluidic device has a simple geometry, and it is cheap and easy to fabricate.However, it allows only for batch (i.e.non-continuous) measurements, due to the transient nature of the salt concentration gradient and the need for regular flushing of the dead-end pores to replace the sample and prevent clogging.
Imposing a salt concentration gradient in a microchip also results in a diffusioosmotic slip velocity at the charged walls of the fluidic channels. 50,51Such a slip velocity is typically weak and has usually negligible effects in pressure-driven flows within open micron-sized channels.Conversely, diffusioosmotic effects can become significant in dead-end channels 35,52 or in highly confined flows within nanotubes 53 and nanochannels. 54Recently, diffusiophoresis and diffusioosmosis have been jointly exploited in a cleverly designed microfluidic platform for the separation and characterization of liposomes and extracellular vesicles. 46In this device, a shallow tapered open-ended nanochannel is exposed to a steady salt concentration gradient and this leads to the size-and charge-dependent accumulation of lipid vesicles nearby the region where the diffusioosmotic flow velocity and the particle diffusiophoretic velocity balance each other.This microfluidic system enables the pre-concentration of lipid vesicles as well as the accurate measurements of their diameter and zeta potential.Although this is the first significant exploitation of diffusioosmotic flows for particle characterization, the sample analysis is again performed in batches, with a small fraction of the sample injected into the device being analyzed at any given time.Moreover, the use of nanochannels makes the device fabrication process more complex and expensive, because it requires costly cleanroom fabrication procedures such as reactive ion etching techniques. 46,51,54Nanochannel devices are also more vulnerable to obstruction and clogging, compared to microchannel chips.
To address these limitations, we propose a novel microfluidic strategy for the manipulation and characterization of colloidal particles via solute concentration gradients in continuous flow, low-cost and easy to fabricate microfluidic devices.First, we report and investigate a newly observed particle focusing mechanism occurring in a single Ψ-junction microchannel device.The intertwined effects of particle diffusiophoresis and flow diffusioosmosis, induced by steady salt concentration gradients, enable the accumulation of colloidal beads and liposomes at multiple focusing regions nearby the charged walls of the microchannels.Notably, unlike previous studies, here the diffusiophoresis and diffusioosmosis effects do not compete one against the other by pushing particles along parallel and opposite directions, but, instead, they act synergistically by moving particles along two perpendicular directions.This allows for the first time the continuous and high-throughput manipulation of colloids in an open-ended micron-sized channel via diffusioosmotic flows without recurring to dead-end or nano-sized channel geometries, therefore reducing the risk of device clogging and avoiding the need of expensive device fabrication procedures.To exploit this novel mechanism, we introduce a new double-junction device that can be used for the continuous separation and characterization of nanoparticles based on their size or zeta potential.Furthermore, we investigate how the chemical composition and lipid phase of the membrane affect the liposome diffusiophoretic response and establish a relationship between the latter and charged lipid content of the membrane.

Particle focusing in single junction devices
A Ψ-junction microfluidic device was used to generate a steady-state gradient of salt concentration (referred to as c in the following) by pumping a low concentration (c L = 0.1 mM) LiCl aqueous solution in the inner channel and a high concentration (c H = 10 mM) LiCl aqueous solution in the outer channels (Fig. 1a).Outer and inner streams had a constant flow rate of 3.65 µL/min.A similar flow configuration was adopted in previous studies [37][38][39]55,56 to investigate the dynamics of charged colloids under steady-state salt gradients and continuous flow conditions.
In a first set of experiments, carboxylate polystyrene nanoparticles (NPs), 216.5 ± 0.9 nm in diameter (Invitrogen, USA) and ζ = −54.9±0.7 mV, were dispersed in the inner flow only, thus leading to the formation of a fluorescent stream of colloidal solution -hereby referred to as colloidal band -within the central region of the channel.The boundaries of the colloidal band are represented by the black dotted lines in the 3D schematic of the device in Fig. 1a.A x, y, z-axis reference system is introduced as shown in the figure.The origin of z-axis is located at the junction, whereas the origins of the x and y axis are located at the mid-points along the channel width and depth, respectively.The parallel streams of electrolyte solutions at different salt concentrations generated a salinity gradient in the direction transverse to the flow (red arrows parallel to x-direction in Fig. 1a), which caused the broadening of the colloidal band due to the diffusiophoresis migration of charged nanoparticles towards higher salt concentration regions.In agreement with previous studies, 37,56 the particle transverse profiles showed an enhanced diffusive dynamics in the transverse direction and the colloidal band width, ∆w, increased linearly with the square root of the longitudinal distance z from the junction (inset in Fig. 1c).As discussed by Abecassis and co-workers, the enhanced colloid diffusivity in the transverse direction is caused by the coupling of the diffusiophoretic migration of particles with the underlying salt diffusion process. 37e chemical gradient field also prompted the appearance of two distinct peaks of focused nanoparticles (when looking at the channel from above), as showed by the epi-fluorescence images in the blue rectangle of Fig. 1a.The peaks were separated from each other by a distance ∆ DO , initially equal to the inner channel width w i .The value of ∆ DO rapidly decreased with z over a typical distance d 23 w i and eventually plateaued at a constant saturation value (Fig. 1c).The decay distance d was estimated by intersecting two straight lines fitting the ∆ DO versus z plot for z/w i → 0 and z/w i → 45, respectively.Particle peaks did not form when no salt concentration gradient was imposed (Fig. S1b,c).The particle focusing effect is quantified by plotting the fluorescence intensity profile against the transverse x-direction (perpendicular to the direction of the flow) at different distances, z/w i , downstream the junction (Fig. 1b).The profiles are normalized with respect to the intensity of the colloidal solution with particle concentration, n 0 , injected in the inner channel of the device.Note that the epi-fluorescence micrographs are generated from the convolution of the particle fluorescence intensity with the microscope point-spread function. 57Thus, the intensity profiles of Fig. 1b are the result of an integration of the particle fluorescence intensity over an optical window, whose characteristic size in the vertical y-direction is given by the depth of field of the optical system (ca.10 µm).For the micrographs in Fig. 1a, the mid-point of this optical window, which coincides with the focal plane of the microscope objective, was located nearby the bottom glass wall of the device.Note that epi-fluorescence images were also acquired with the focal plane located nearby the top PDMS wall and at the mid-point along the depth of the microchannel, but no significant change in the micrographs and corresponding intensity profiles could be detect.From the intensity profiles in Fig. 1b, we can observe a clear effect of focusing of the nanoparticles (solid lines) when compared to the case without salinity gradient (dashed lines).A confocal scan of the (x, y)-plane confirms that under salinity gradient conditions, charged particles were accumulating at the top and bottom walls of the device in four symmetrical positions (red rectangle in Fig. 1a).In the absence of a salinity gradient, the particle distribution profile in the same plane was uniform (Fig. S1c), therefore the observed particle dynamics was driven by the salt contrast generated in the microchannel.
The formation of peaks at the channel walls are in agreement with the observations reported in our previous study, where we investigated the diffusiophoresis manipulation of charged nanoparticles in a Ψ-junction channel fitted with a microgrooved substrate. 38Our analysis revealed that the particle accumulation regions at the flat wall of the device were induced by the vertical component of the salt concentration gradient which is originated by the Poiseuille-like velocity profile in the rectangular microchannel. 38Conversely, the convergence of the peaks in the (z, x)-plane towards lower salinity regions is unprecedented.This particle behavior is also entirely unexpected as it seems to contradict the well-established interpretation of particle diffusiophoresis, 50 according to which colloids migrate towards higher salinity regions with a diffusiophoresis velocity u DP = Γ DP ∇(ln c) when the diffusiophoresis mobility Γ DP is positive -as it is the case for negatively charged particles in LiCl solutions. 37This implies that additional phenomena, such as hydrodynamics and diffusioosmosis induced flows, must be involved in the mechanism responsible for the transverse deviation of the peaks.
To verify our hypothesis, two additional experiments with different flow configurations were conducted (Fig. 2a,b).First, a salinity gradient was imposed as done in the experiment depicted above (high salt concentration c H in the outer channels and low concentration c L in the inner channel).However, contrary to what was done in the initial experiment, a homogeneous solution of nanoparticles was injected in both inner and outer channels (Fig. 2a).In a second configuration, the chemical gradient was swapped (high salt concentration in the inner channel, Fig. 2b).As showed by the fluorescence intensity profiles along the transverse direction (Fig. 2c,d), the focused peaks induced by diffusiophoresis effects were observed to converge (∆ DO decreases with the distance) in the first configuration (Fig. 2a) and diverge (∆ DO increases with the distance) in the second configuration (Fig. 2b).As observed in the previous experiment, the variation of ∆ DO with z (Fig. 2d,e) was rapid at shorter distances from the junction (z/w i 25) and more slowly at larger distances (z/w i 25).By swapping the salinity gradient, we effectively interchanged the chemically generated electrical field that gave birth to the fluid flow near the charged walls induced by the diffusioosmosis slip velocity, u DO = −Γ DO ∇(ln c), where the diffusioosmosis coefficient, Γ DO , has same sign and order of magnitude of Γ DP .No physical parameters or properties in our system were varied otherwise.Consequently, the systematic counter-intuitive transversal behavior of focused peaks in both configurations (Figs.2d,e) suggests that they were deeply affected by diffusioosmosis flows -which also justifies our choice of notation, ∆ DO , for the peak distance.
In other words, the peak displacement was always against the diffusiophoresis-driven motion predicted for Γ DP > 0 (red arrows in Fig. 2a,b), and, instead, it had always the same direction of the diffusioosmosis slip velocity for Γ DO > 0 (red arrows in Fig. 2a,b).Note that in these two complementary experiments the presence of particles in both inner and outer streams resulted in the formation of two additional, but narrower and more intense peaks, separated from each other by a distance ∆ DP (Fig. 2c,d) which evolved in an opposite way compared to ∆ DO (insets in Fig. 2e,f).These peaks had been previously reported 56 and are intrinsically different from the ones discussed so far.Indeed, they form as the transverse x-component of the chemical gradient induces a particle migration towards higher salt regions via diffusiophoresis only -thus the notation ∆ DP .This causes nanoparticles to accumulate along the entire depth of the channel and not just nearby the channel walls. 56In addition, the transversal migration of these peaks is extremely slow as their separation distance ∆ DP is proportional to z 1/2 , which is notably the same scaling as ∆w.Conversely, the separation distance ∆ DO changes much faster with the distance from the junction (Fig. 2e,f).
A numerical analysis was performed in COMSOL Multiphysics to confirm our interpretation of the experimental observations.The numerical hydrodynamic velocity field u, salt concentration c, and particle concentration n, were calculated in a 3D domain consisting of a straight rectangular channel.A slip velocity u DO = −Γ DO ∇(ln c) was imposed at the channel walls and the particle velocity u p was calculated as the sum of the hydrodynamic velocity and the diffusiophoresis velocity, u p = u + u DP .The value of Γ DO for the channel walls could not be measured, so it was used as an adjusting parameter in the model to achieve a good match between experimental and numerical results (see Numerical Methods section for details).The simulated transverse profiles of the normalized particle concentration n/n 0 , at increasing distances z downstream the junction, are shown in Fig. 1d, and they are in good agreement with the fluorescence intensity profiles measured in the experiments (Fig. 1b).
It is worth noting that a close quantitative match is not expected since the experimental profiles correspond to the convolution of the particle fluorescence intensity with the microscope point-spread function, 57 whereas the numerical profiles are directly obtained from the simulated particle concentration field by averaging the concentration over the channel depth (y direction).A good quantitative agreement between experiments and simulations can be seen for the peak separation ∆ DO at increasing distances from the junction (Fig. 1c), which is consistent with the fact that the effect of the microscope point-spread function on ∆ DO measurements should be negligible.Fig. 1e shows the simulated particle concentration field on the plane perpendicular to the flow direction at z/w i = 25, and this is in a good agreement with the confocal image of the channel cross-section at the same distance from the junction (Fig. 1a).As expected, the salt concentration isolines in the inner region of the channel (|x|/w i < 0.5), showed in Fig. 1e, are bent towards the outer flow because of the Poiseuille-like hydrodynamic velocity profile.Consequently, the onset of a vertical component of the salt concentration gradient and, thus, of the diffusiophoresis velocity, causes the accumulation of particles at the top and bottom walls.The diffusioosmosis slip velocity at the walls induces the formation of four symmetric recirculation regions in the in-plane total particle velocity field u p , whose streamlines are showed by white arrows in Fig. 1e.As a result, the accumulated particles are advected along the top and bottom walls toward the central region of the channel, namely from higher to lower salt concentration regions.To summarize, the observed particle behavior is governed by a combination of particle diffusiophoresis along the vertical axis, which induces particle accumulation, and diffusioosmosis flow along the horizontal axis, which pushes the accumulation peak towards the center of the channel.

Particle size detection and size-based separation in double-junction devices
The Ψ-junction microchip, depicted in Fig. 1a, could be potentially adopted for the online pre-concentration, via solute-driven transport, of charged synthetic or biological colloids, including macromolecules, 58 liposomes, 35 exosomes, 46 viruses 59 and bacterial cells. 60This can be particularly useful for applications such as microfluidic point-of-care diagnostics and point-of-need bioanalytical testing, provided that the target analytes are charged and, thus, susceptible to diffusiophoresis migration.Alternatively, the same device could be used for the solute-driven accumulation of charged nanoparticles that are conjugated to recognition moieties, such as antibodies or aptamers, for the capture and detection of the target molecules. 61Furthermore, since the diffusiophoresis mobility depends on both particle size and zeta potential, 35,47 the diffusiophoresis-driven accumulation of nanoparticles at the device walls could be exploited also for particle characterization, fractioning (commonly known as field flow fractioning) and sorting.However, the device configuration showed in Fig. 1, does not lend itself to such applications, because the accumulation peaks are advected towards the central region of the channel, thus overlapping with the bulk colloidal stream.
Consequently, particle fractioning and sorting are not possible.Moreover, the fluorescence intensity of the accumulation peaks is partially screened by the background fluorescence signal generated from the colloids in the bulk (see Fig. 1a,b), thus limiting the accuracy of the peak intensity detection and hampering the ability to characterize particles by charge or size.These limitations could be overcome if the accumulation peaks migrated away from the bulk colloidal stream.To achieve this, first we tested a different flow configuration, whereby a salinity gradient was imposed as done in the experiment of Fig. 1 -namely higher salt concentration c H in the outer channels and lower salt concentration c L in the inner channel -but the colloidal particles were present in the outer stream only.However, this test came to no avail (see Fig. S2 in Support Information Material), since the accumulation peaks did not form.This is because the salt concentration isolines, shown in Fig. 1e, are bent outwards only within the inner stream region of the channel (|x|/w i ≤ 0.5), but no colloids are now present in that region.As a result, the vertical component of the salt gradient no longer leads to the migration of colloids towards the top and bottom walls and the consequent formation of the particle accumulation peaks.Therefore, a new chip design was required to exploit the observed phenomenon of particle focusing for particle fractioning, separation and characterization.Fig. 3a,b depict the blueprint of the double Ψ-junction microchip designed for this purpose.The channel geometry is similar to the one of the previous device, but a second Ψ−junction was added upstream of the first one.The downstream junction is used to regulate the salt concentration gradient in the device by swapping the outer flow stream between a low salt concentration (c L ) solution, leading to no salt gradient (Fig. 3a), and a high salt concentration (c H ) solution, generating a steady-state salt gradient (Fig. 3b).The upstream junction allows to control the position of the colloidal stream within the main channel.All streams injected from the inlet channels of the upstream junction have a low salt concentration (c L ), but only the outer streams are laden with colloidal particles.Consequently, the inner region of the channel remains particle-free so that, upon imposition of the salt gradient, the focused particle peaks can converge into this region without overlapping with the bulk colloidal stream.To quantify this focusing effect, we introduce the focusing parameter, Ī, defined as the average value of the normalized fluorescent intensity profile within the range of interest, x/w m ∈ [−0.2, 0.2], which is equivalent to a transverse section, ca.50 µm wide, located at the center of the channel (Fig. 4a).Note that the width of the region of interest is chosen to exclude the fluorescence intensity generated by the particles in the bulk colloidal stream, namely the gray shaded regions in Fig. 4a.Fig. 4b shows the focusing parameter at increasing distances z/w m downstream the junction in the presence (empty circles) and absence (solid circles) of a salt concentration gradient.Under the examined conditions, at z/w m = 36, the focusing parameters were ĪH = 0.66 and ĪL = 0.03 with and without a salt concentration gradient, respectively.
Since the formation of the accumulation peaks is driven by the diffusiophoresis migration of charged nanoparticles from the channel bulk toward the top and bottom walls of the microchannels, it is reasonable to expect that the peak intensity, and thus the focusing parameter, can be correlated to the diffusiophoresis coefficient of the nanoparticles.For a charged nanoparticle in an electrolyte solution, the diffusiophoresis coefficient Γ DP is an increasing function of the particle size, 35,47 if the thickness of the Debye layer κ −1 formed around the particle is a few percent greater than the particles radius a, namely (κa) −1 1%.
On the other hand, for (κa) −1 → 0 the coefficient, Γ DP , levels off to a constant value that is independent of the particle size.In a 0.1 mM LiCl solution, the Debye layer thickness is κ −1 = 32 nm, therefore it is expected that for sub-micron particles (2a 1 µm and (κa) −1 > 6% ), the coefficient Γ DP and the accumulation peak intensity increase with the particle size.To verify this hypothesis, we measured the focusing parameter for carboxylate polystyrene particles with a diameter ranging from tens nanometers to one micrometer.Note that the zeta potential of the particles were very similar for all diameters (Tab.1).From the logarithmic plot in Fig. 5a, it is apparent that the focusing parameter Ī increases with the particle diameter, d DLS , determined via dynamic light scattering (DLS).Interestingly, log Ī and log d DLS are linearly correlated (R 2 = 0.99).
log Ī = 0.299 log d DLS − 0.941 (1)   where the value of d DLS is expressed in nanometers.Consequently, Eq.( 1) can be effectively used as a calibration line for the microfluidic characterization of the size of nanoparticles of similar zeta potential, whereby the focusing parameter Ī is measured experimentally to The combined effects of peak formation and drift towards the center of the channel can be exploited also for particle size-based separation.To demonstrate this additional application of the double-junction chip, a binary colloidal mixture of 36 nm and 1.098 µm diameter particles was injected into the device.The carboxylate polystyrene particles were stained with fluorophores with different emission peaks, namely 515 nm (yellow-green) for the smaller particles and 605 nm (red) for the larger particles.In absence of a salt concentration gradient, both populations of particles were advected by the flow and remained confined in the same regions of the channel where they were initially injected (Fig. 5b).Upon imposition of a salt contrast (c L = 0.1 mM, c H = 10 mM), the larger (red) particles were strongly focused at the center of the channel where they formed two intense peaks.Conversely, the smaller (yellowgreen) particles did not form any accumulation peak and only a small fraction of them drifted toward the center.Furthermore, the two colloidal bands made of larger particles, expanded along the transverse (x) direction towards the higher salt concentration (outer) regions of the channel due to diffusiophoresis, 37 with two additional focusing peaks forming at these locations.On the other hand, the colloidal bands made of smaller particles did not drift towards the outer regions because of their smaller diffusiophoresis coefficient.As a result, diffusiophoresis and diffusioosmosis could be effectively exploited for the microfluidic separation and sorting of the two populations of particles according to their size.

Detection of liposome zeta potential and membrane composition in double-junction devices
The range of applications of the double-junction device can be expanded further by leveraging the dependence of the diffusiophoresis coefficient on zeta potential, which is a key property of colloidal system.To this end, we used nano-sized unilamellar liposomes whose surface charge and, thus zeta potential, can be easily tuned by adjusting the lipid membrane composition.
Such nanoparticles, which are often used as drug carriers 62,63 and cell membrane models, 64 are hence suitable for investigating the effect of zeta potential on the focusing phenomenon in the double-junction device.
In the first set of experiments, we fabricated negatively charged liposomes by adding 10 % mol fraction of the anionic phospholipid 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (DOPS) to the zwitterionic DPPC phospholipid (d = 166 ± 4 nm, ζ ≈ -57 mV, polydispersity index ≈ 0.1).The phophoserine headgroup (PS) was specifically chosen as its presence in the outer leaflet of the membranes of biological particles can be associated with pathological and physiological processes, such as in tumor-derived exosomes 65 and during cell apoptosis. 66The vesicles were dispersed in a low salt concentration (c L = 0.01 mM) LiCl solution buffered with HEPES salt and EDTA chelating agent (pH = 8.15).The aqueous stream of charged liposomes was pumped in the outer channels of the upstream junction, whereas the same low salt concentration solution without liposomes was injected in the inner channel of the device.Finally, a high salt concentration (c H = 6.65 mM) LiCl solution, also buffered with HEPES salt and EDTA, was pumped in the outer channels of the downstream junction.
The fluorescence intensity profiles along the transverse direction are showed in Fig. 6a.It can be observed that charged liposomes display a similar behavior to the one observed for polystyrene nanoparticles.Upon imposition of a salt gradient (red curves), peaks of accumulated liposomes formed and converged to the center of the device (|x|/w m ≤ 0.2) and away from the colloidal bulk streams (|x|/w m > 0.2).Conversely, when the salt concentration was the same throughout the device (blue curves), liposome peaks did no form and no particles migrated toward the inner region of the channel.Note that for liposomes, a higher salt contrast (c H /c L = 333) was adopted in comparison to the one applied in the experiments with carboxylated polystyrene nanoparticles (c H /c L = 100).Indeed, in agreement with previous studies, 46,67 the migration speed of liposomes under salt gradients is typically smaller that the one of polystyrene nanoparticles with comparable size and zeta potential, therefore higher salt contrasts are required.This is likely due to the soft and water-permeable nature of the liposomes that may trigger additional migration mechanisms, such as osmophoresis, 68 or affect the diffusiophoresis response through membrane permeability, 69,70 membrane viscosity and vesicle shape deformations.To confirm that the imposed salt contrast did not affect the liposome stability and size in our experiments, a liposome solution in low salt concentration (c L ) buffer was analyzed via dynamic light scattering before and after dilution in high salt concentration (c H ) buffer, resulting in no detectable changes in the particle size distribution.Interestingly, the inset in Fig. 6a shows that when using zwitterionic DPPC liposomes (ζ = +10 mV) instead of negatively charged DPPC-DOPS liposomes, no focusing effect is observed under a salinity gradient.This observation is consistent with our physical interpretation of the particle focusing phenomenon.Indeed, DPPC liposomes in the buffer solution carry a very weak positive charge and, therefore, do not migrate by diffusiophoresis towards the top and bottom non-zero charged walls of the channel.In absence of a particle accumulation nearby the walls, the diffusioosmosis flows at the walls do not affect the colloid distribution in the device and no particles are directed toward the center of the channel (|x|/w m ≤ 0.2).Remarkably, this finding could be exploited for the microfluidic separation of liposomes based on their surface charge.Indeed, for a mixture of negatively charged liposomes and zwitterionic lipid vesicles, only the charged liposomes will migrate towards the central region of the microchannel, whereas the trajectories of the zwitterionic lipid vesicles won't be affected by the salt concentration gradient and they will keep clear of the central region of the channel.
In another set of experiments, we investigated the effect of the lipid composition of the liposome membrane on particle focusing.Specifically, we adjusted the zeta potential of the liposomes by varying the amount of charged lipid content in the membrane composition, and we controlled the viscosity/fluidity of the membrane by using lipid mixtures with a fluid/gel phase transition temperature either above or below the room temperature.Since altering the zeta potential of the liposomes affect significantly their diffusiophoresis mobility, it should be possible to correlate the charged lipid content with the intensity of the focusing effect in the device.On the other hand, it is known that the diffusiophoresis speed depends also on the Newtonian and Maxwell stress balance at the particle surface, 71 which in turn depends on the viscosity of the membrane.However, to date the role of membrane viscosity on the diffusiophoresis of lipid vesicles or living cells has yet to be explored.To carry out this analysis, we considered six different liposome populations at 1,2 and 10% DOPS (charged lipid) mole fraction added to zwitterionic lipids, either 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) for a disordered fluid phase membrane or 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) for an ordered gel phase membrane.Three independent samples were produced for each liposome population, and for each sample both the zeta potential and the corresponding focusing parameter were determined.All liposome populations had similar size (d = 166 ± 4 nm) and polydispersity index.Fig. 6b and Fig. 6c show how the focusing parameter I is positively correlated with the particle zeta potential and the DOPS (charged lipid) molar fraction.
Remarkably, for a given DOPS concentration, there is no statistically significant difference By combining Eq. ( 5) and Eq. ( 4), it follows x DOPS = 7.91 Ī − 1.74 (6)   which allows one to estimate the DOPS lipid molar fraction from the experimental measurements of the focusing parameter Ī.To conclude, Eq. ( 4) and Eq. ( 6) show how the double-junction device can be used effectively for the quantitative estimate of both the zeta potential and the DOPS lipid content of the outer leaflet of liposome membranes.

CONCLUSIONS
We Those peaks are of an intrinsically different nature of those observed in previous works, 37 the dynamics of which was exclusively driven by diffusiophoresis.
We also showcased the exploitation of this novel mechanism for the continuous separation and characterization of colloidal particles.A proof-of-concept double-junction device was developed and used for the accurate measurements of the diameter of colloidal beads with same ζ-potential, based on the measurement of an ad-hoc, purposely defined, focusing parameter Ī.We also demonstrated how the dependence of the transverse drift dynamics of the accumulation peaks on the colloid size can be further leveraged to separate large particles (1.098 µm) from smaller ones (36 nm) in a bi-modal mixture.Moreover, the setup is proven to allow for a reasonable assessment of the ζ-potential of same size particles through the measurement of the focusing parameter Ī.Based on this latter principle, we eventually showed how, in the case of liposomes, the measurement of the ζ-potential can be used to assess the chemical composition of the membrane (here the molar fraction of the charged DOPS lipid), at least in the low concentration limit.Note that, whilst previous studies allowed only for batch measurements of particles located in dead-end pores or open nanochannels, our double-junction device enables the online, continuous and high-throughput characterization of particles directly within the colloid stream.This also facilitates the recollection of the analyzed sample for further off-chip downstream analysis.Finally, we showed how, under the examined experimental conditions, the fluid-like or gel-like states of the membrane, and thus the membrane viscosity, does not affect the diffusiophoretic response of the liposomes.
We envisage that this study will open up new exciting routes for the use of diffusiophoresis and diffusioosmosis for the microfluidic manipulation and characterization of both synthetic and natural particles.Our microfluidic devices can unlock a wide scope of possibilities related to bio-oriented applications, such as the pre-concentration, sorting, sensing and analysis of biological entities, including liposomes, extracellular vesicles and bacteria.By relying on the chemical energies of the electrolyte solutions rather than on external energy sources and by adopting cheap and easy-to-fabricate microfluidic chips, the proposed particle manipulation and characterization strategies naturally lend themselves to the development of portable, cost-effective, point-of-need microdevices for chemical and biochemical analysis, diagnostics, drug screening and drug delivery applications.

Particle characterization
All size and zeta potential measurements were performed using a ZetaSizer nano ZS (Malvern Panalytical) at 25°C.All samples were analyzed a minimum of 3 times using the monodomal measurement mode for a monodisperse single population of particles.The instrument provided the zeta potential values calculated according to the Smoluchowski's theory for which the Debye length is much smaller than the particle size.The zeta potential values were corrected to account for the finite size of the Debye length according to Henry's model. 76

Numerical methods
Numerical simulations were performed in Comsol Multiphysics according to the procedures detailed in our previous work. 38Briefly, the 3D computational domain consisted in a rectangular channel of width 2w i = 400 µm, depth h = 52 µm and length 25w i + 5h.The

Salinity conditions of NP solution for peak formation
We find that peak formation due to the reported focusing effect does not occur when the carboxylate polystyrene nanoparticles are dispersed in the high salt (c H = 10 mM) solution in the outer channels and not in the low salt (c L = 0.1 mM) solution in the inner channel (Fig. S2a).Consequently, this led to the development of the two Ψ-junction microchip for nanoparticle fractioning to target advanced applications.Moreover, we observe a previously

Figure 1 :
Figure 1: (a) Schematic and micrographs of the single Ψ-junction device.Outer streams: LiCl solution at high concentration (c H ). Inner stream: LiCl solution at low concentration (c L ) seeded with fluorescent (d = 216.5 ± 0.9 nm) carboxylate polystyrene nanoparticles (NPs), shown as green dots.The decay of the distance ∆ DO between the accumulation peaks with the distance from the junction (z) is visible in the top view epi-fluorescence micrographs (blue rectangle).The micrographs were acquired with a focal plane located nearby the bottom glass wall of the device.The four accumulation regions in the transverse (x, y)-plane are visible in the confocal image of the channel cross-section at a distance z/w i = 25 from the junction (front view, red rectangle).(b) Experimental intensity profiles along the transverse x-direction with a salt gradient (solid lines) and without a salt gradient (dashed lines), at increasing distances z/w i .(c) Experimental and numerical peak separation distance ∆ DO as a function of the distance z/w i .Inset shows the profile of the colloidal band width, ∆w, as a function of the distance z/w i .(d) y-averaged NP concentration profiles along the transverse x-direction at increasing distances z/w i , predicted by numerical simulations.(e) Numerical map of the channel cross-section at a distance z/w i = 25 showing the NP concentration field, the salt concentration isolines and the total particle velocity field, u p = u + u DP , streamlines (white arrows) with diffusioosmotic slip velocity, u DO , at the channel walls.

Figure 2 :
Figure 2: Relationship between the direction of the salt gradient and the dynamics of the peak displacement mechanism due to diffusioosmosis.(a,b) Schematic diagrams of a single Ψ-junction microchip, where carboxylated NPs are uniformly injected (in all channels) with two different configurations of salt gradient.In (a), high concentration salt is in the outer channels (c H /c L = 100), whilst in (b) high salt concentration is in the inner channel.Red-towhite shade qualitatively indicates the field of salt concentration.Red (blue) arrows show the direction of diffusiophoresis transport (diffusioosmosis flow).(c,d) Normalized fluorescent intensity profile plots along the transverse direction at various distances z/w i downstream the junction, for the configurations depicted in schematics (a) and (b), respectively.(g,h) Normalized longitudinal distance ∆ DO plotted with respect to the normalized distance z/w i for the configurations (a) and (b), respectively.

Figure 3 :
Figure 3: Particle focusing and fractioning in a double Ψ-junction device.(a,b) Schematic diagrams of the double-junction microchip and the corresponding epi-fluorescence micrographs showing the fluorescent carboxylate polystyrene nanoparticles (d = 549.8±6.8 nm, ζ= -50.4 ±0.2 mV) at various distances from the junction when there is no salinity gradient (∇c = 0) in (a) and when there is a salinity gradient (c H /c L =100) in (b).The micrographs were acquired by positioning the focal plane nearby the bottom wall of the device.Scale bar is 75 µm.(c) Normalized fluorescent intensity profiles when there is no salinity gradient (blue curve) and when there is a salinity gradient (red curve) at various distance downstream the junction.

Figure 4 :
Figure 4: Focusing parameter in double-junction device for carboxylate polystyrene nanoparticles (d = 549.8±6.8 nm, ζ= -50.4 ±0.2 mV) (a) Normalized fluorescent intensity profile at z/w m = 36 without salinity gradient (blue curve) and with salinity gradient (red curve).The region of interest corresponds to a selected central region of the channel, x/w m ∈ [−0.2, 0.2].The focusing parameter, Ī, is calculated as the average fluorescence intensity over the region of interest.(b) Focusing parameter at different distances downstream the junction (z/w m ) under no salinity gradients (solid circles) and with a salinity gradient (empty circles).The blue and red dashed lines are the focusing parameter values corresponding to their respective intensity plots in (a) at z/w m = 36.

Figure 5 :
Figure 5: Particle size characterization and particle separation.(a) Average focusing parameter for different particle sizes at approximately the same zeta potential (see Tab. 1).(b-c) Epi-fluorescence images taken at z/w m = 36 showing particle dynamics of mixed yellow green fluorescent (505/515) 36 nm and red fluorescent (580/605) 1098 nm carboxylate polystyrene colloids under no salinity gradient (b) and with salinity gradient (c).Scale bar is 50 µm.White dashed line corresponds to the microfluidic channel boundaries.

Figure 6 :
Figure 6: (a) Normalized fluorescent intensity profiles along the transverse direction x and at varying distances z from the downstream junction without (blue curves) and with (red curves) a salinity gradient for negatively charged (ζ = −57 mV) 10:90 DOPS:DPPC liposomes.Inset shows the fluorescent intensity profile for zwitterionic DPPC liposomes (ζ = +10 mV).(b) Experimental focusing parameter against zeta potential for DOPC (blue) and DPPC (red) based liposomes with varying anionic DOPS concentrations (navy box = 1 PS, orange = 2 PS, purple = 10 PS .(c) Experimental focusing parameter against the anionic DOPS lipid content.(d) Relation between vesicle zeta potential and anionic DOPS lipid content for DOPC-DOPS vesicles.The dashed line corresponds to the linear regression of the first four data points.Fabricated liposomes had an average size of d = 166 ± 4 nm.
between the focusing parameters observed for DOPC-DOPS fluid-like membrane vesicles (blue bars/symbols) and for DPPC-DOPS gel-like membrane vesicles (red bars/symbols), thereby showing the irrelevance of the liposome membrane viscosity for the diffusiophoresis transport under the examined conditions.The linear regression (dashed line) between liposome zeta potential and focusing parameter is given byĪ = −0.0116ζ(3)with ζ expressed in millivolt.Eq.(3) can be used as a calibration curve for the microfluidic detection of the zeta potential of similarly-sized particles, whereby the focusing parameter Ī is measured experimentally to quantify the zeta potential according to the following relationζ = −86.2Ī (4)Furthermore, we conducted an electrophoretic light scattering analysis to correlate the zeta potential of DOPC-DOPS vesicles with the DOPS (anionic lipid) content in the membrane (Fig.6d).As the DOPS molar fraction increases, the magnitude of the zeta potential of liposomes increases, and eventually it plateaus at ca. 8-10% DOPS content.Beyond this point, adding more DOPS to the membrane does not affect the zeta potential, due to the formation of a charged condensed layer of Na + counterions around the outer leaflet of the liposomes.72On the other hand, the zeta potential is highly sensitive to the DOPS content at low anionic lipid concentrations (≤ %3).This suggests a strategy for the application of the double-junction device for the quantification of small amounts of DOPS lipids in the outer leaflet membranes of liposomes.Indeed, for low anionic lipid concentrations (≤ %3), the relation between zeta potential ζ (expressed in millivolt) and the DOPS molar fraction x DOPS (expressed in percentage values) can be well approximated by the following linear relationship, plotted as a dashed line in Fig. 6d, ζ = −10.9x DOPS − 19.0 described an unreported mechanism where diffusiophoresis and diffusioosmosis are closely intertwined, leading to a strong transverse focusing of nanoparticles in a single Ψ-junction microchannel under continuous and steady axial flow conditions.Parallel electrolyte streams are merged at the junction of the device to generate a chemical gradient in both the transverse and vertical directions.As a result, the particles first migrate vertically by diffusiophoresis from the bulk towards the top and bottom walls, and subsequently undergo a transverse horizontal migration along these walls driven by diffusioosmosis.The remarkable coupling of diffusiophoresis and diffusioosmosis along two perpendicular directions allows us to take advantage of the relatively weak diffusioosmotic slip velocities in a pressure-driven microfluidic flow without resorting to dead-end or nano-sized channels.Indeed, here it is the vertical diffusiophoretic migration that confines the particles nearby the charged walls where the effects of diffusioosmotic flows are the most intense.Consequently, one can avoid the most common drawbacks associated with dead-end pores or nano-confined channels, such a risk of device obstruction and clogging, costly device fabrication procedures and difficult recovery of the colloidal sample.By means of epi-fluorescence microscopy, two accumulation peaks are formed and their separation distance decreases with the distance from the channel inlet, eventually reaching a plateau value that depends on the size and ζ-potential of the colloids.

Invitrogen™
FluoSpheres™ , carboxylate-modified nanoparticles, red fluorescent (580/605), 2 % solids, were purchased from ThermoFisher scientific at various sizes(20,100,200,500,1000 nm).In addition, Invitrogen™ FluoSpheres™ , yellow-green fluorescent (505/515), 2% solid, 20 nm carboxylate-modified nanoparticles was also purchased for separation experiments from the same supplier.Lithium chloride salt (LiCl, 99%) used for diffusiophoresis experiments was purchased from Acros Organics.Aqueous liposome solutions were prepared with buffer salt, HEPES and chelating agent EDTA purchased from Sigma Aldrich.Lipophilic dye, 3,3'-dioctadecyloxacarbocyanine perchlorate (DiO c 18 ) -used for staining liposomesand chloroform (99%) -used for preparing lipid films -were purchased from Sigma Aldrich.RTV 615 polydimethyl siloxane used for the fabrication of microfluidic devices was purchased from Techsil, UK.The phospholipids 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), 1,2dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) and 1,2-dioleoyl-sn-glycero-3-phospho-Lserine (DOPS) were purchased from Avanti Polar Lipids.Deionized water (18.2MΩ cm) produced from an ultrapure milli-Q grade purification system (Millipore, USA) was used to prepare all aqueous solutions.Fabrication and operation of the microfluidic devicesStandard photo-and soft-lithography techniques were employed in manufacturing the microfluidic devices.Briefly, the CAD drawings of both one and two-junction devices were printed on a photomask film (Micro Lithography Services, UK) and subsequently used to produce an SU-8 master mold of ca.50 µm thickness on a silicon wafer (Inseto, UK).Imprinted polydimethylsiloxane (PDMS) channels are then made via replica molding by heating a PDMS:curing agent (9:1) mixture on top of the SU-8 master mold.The channels are then irreversibly bonded to a microscope slide by using a plasma cleaner (Harrick Plasma, UK) at 18W power for 1 minute.All devices have a nominal depth of 50 µm and a total channel width of w = 400 µm.For the single junction chip, the width of the main channel and the inner inlet channel are w = 400 µm and w i = 200 µm, respectively, whereas the width of the outer inlet channels are (w − w i )/2 = 100 µm.For the double-junction device, the width of the main channel is w = 400 µm.For the downstream junction, the width of the inner channel and outer inlet channels are w m = 250 µm and (w − w m )/2 = 75 µm, respectively.For the upstream junction, the width of the inner and outer inlet channels are w i = 100 µm and (w m − w i )/2 = 75 µm, respectively.Aqueous solutions are injected into the device inlets at a constant flow rate of 3.65 µL/min by means of syringe pumps (Harvard, USA).To point or the top (PDMS) wall of the device.The location of the focal plane was established by focusing on colloidal particles permanently stuck on the bottom and top walls of the microchannel.A PicoQuant MicroTime 200 time-resolved confocal microscopy platform, built around an Olympus IX 73 microscope, was used to acquire confocal images in the x-y plane, as shown in Figure 1.A plan N 40x water immersion objective lens (0.65 NA) was used to take 1024 x 1024 px, 32 bit TIFF images.A monodirectional scanning pattern was used with a learning time of 5 seconds and a dwell time of 2 seconds.Fluorescent intensity profiles were normalized by first subtracting the background noise and then dividing by the average intensity of the bulk obtained from a no salinity gradient configuration.All fluorescent image shown in figures were processed using ImageJ (contrast enhancement, LUT color change).
hydrodynamic velocity, u, pressure p, salt concentration c and particle concentration n were calculated by solving the steady-state Navier-Stokes equation and the advection-diffusion equations for c and n.At the channel inlet, the boundary condition for the velocity field was u = u inlet , with u inlet the fully developed velocity field at a cross section of the rectangular channel perpendicular to the flow direction and with average velocity U 0 .The boundary conditions at the channel inlet for the salt and particle concentration fields are c = c H and n = 0 for the outer flow region (i.e., |x| > w i /2) and c = c L and n = n 0 for the inner flow region (i.e., |x| ≤ w i /2).At the channel outlet, the zero normal gradient boundary condition for the pressure, salt and particle concentrations were imposed.At the remaining walls, the slip boundary condition u = −Γ DO ∇(ln c) was applied together with the zero flux condition for the salt and particle concentration fields.The channel outlet was located at 5 times the channel depth h from the cross section z/w i = 25 to ensure that the boundary conditions at the channel outlet do not affect the fields near that section.The particle diffusiophoresis coefficient Γ DP = 291 µm 2 /s was calculated according to the procedure detailed in our previous work,38 where the formula provided by Prieve and co-workers47 was used to account for the particle size effect on Γ DP .A diffusioosmosis coefficient of Γ DO = 1165 µm 2 /s was chosen for the channel walls to obtain a good quantitative match between experimental results and numerical predictions.and ability to maintain a membrane potential.Biochimica et Biophysica Acta (BBA)-Biomembranes 1985, 812, 55-65.75.Inoue, S.; Oldenbourg, R. Microscopes Handbook of Optics.1995.76.Hunter, R. J. Foundations of colloid science; Oxford University Press, 2001.since both inertia and gravity effects are also negligible, the colloids behave as passive tracers.

Figure S2 :
Figure S2: Absence of the hereby reported focusing mechanisms when colloids are in high salinity.(a) Schematic diagram of a single Ψ-junction microchip, where low salt is injected in the inner channel and high salt with colloids being injected in the outer channels.(b) Epi-fluorescence images taken at different distances downstream the junction for the flow configuration in (a), scale bar = 75 µm.(c) The dynamics of particle migration against the square root of the longitudinal distance downstream the junction.

Table 1 :
Comparison between dynamic light scattering (DLS) and microfluidic (MF) characterization of sub-micron particle size.Ī measured focusing parameter, d DLS particle size determined via DLS, ζ particle zeta potential, d M F particle size determined microfluidically via Eq.(2), ε r absolute relative difference between d DLS and d M F .The error for d DLS and ζ represent the standard deviation obtained from the instrument.The error for Ī is the standard error from experimental mean values.