You are on page 1of 7

pubs.acs.

org/ac Article

Shape-Enhanced Open-Channel Hydrodynamic Chromatography


Valentina Biagioni, Stefano Cerbelli, and Gert Desmet*

Cite This: Anal. Chem. 2022, 94, 15980−15986 Read Online

ACCESS Metrics & More Article Recommendations

ABSTRACT: Hydrodynamic chromatography (HDC) is a well-established


Downloaded via TUNKU ABDUL RAHMAN UNIV COLG on November 29, 2022 at 13:24:32 (UTC).
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

analytical separation method for the size separation of nano- and microparticles
and large molecular weight solutes such as synthetic polymers and proteins. We
report on a theoretical study showing that the separation resolution of open-
tubular HDC can be significantly enhanced by changing the cross-sectional shape
of the separation channel. By enforcing Brenner’s macro-transport approach, we
provide theoretical/numerical evidence showing how the shape of the cross section
influences quantitatively both the selectivity and the axial dispersion of the
suspended particles in HDC. The separation performance of square-, triangle-, and
star-shaped channel cross sections is compared to that of a cylindrical capillary
over three decades of the particle Péclet number in terms of the minimal
separation length and time to obtain the unit resolution of a two-particle mixture.
Enhancement factors up to 400% are found in the case of triangular shapes, with
the best performing case being the 70.6° angle, which can be obtained by KOH etching of bulk silicon.

1. INTRODUCTION and Guttmann theory.10,11 Their work culminated in the


Hydrodynamic chromatography (HDC) is a chromatographic conduction of HDC separations in flat rectangular micro-
separation technique suitable for the size-based separation of channels (1 μm deep and 1000 μm wide), envisioning to
colloidal matter, both macromolecular and particulate.1,2 It combine the increased size selectivity of the very shallow
differs from other modes of chromatography in the sense that channel depth with the increased detection sensitivity
it does not rely on a stationary phase but makes use of the emanating from the large channel width.7,12 More recently,
difference in migration velocity experienced by different-sized the HDC principle has also been applied in perfectly ordered
micro-pillar array columns, showing the separation of 20 and
solutes in the parabolic velocity field that is established in
40 nm particles in 90 s in a 4 cm long channel filled with 1.5
pressure-driven flows through channels or the interstitial pores
mm pillars.13,14 It has also been shown that the HDC effect can
in packed beds. This difference in particle migration velocity
play a role in the Taylor Dispersion Analysis of large MW
becomes appreciable when the size of the solutes becomes
solutes, such as proteins.15
significant with respect to the cross-sectional diameter. It is
In the present contribution, we report on a theoretical
caused by the fact that the center of mass of larger particles is
exploration of the effect of the cross-sectional shape on the
more effectively excluded from the low-velocity regions near
resolution of open-tubular HDC. It can be hypothesized that
the wall. Hence, bigger particles assume a larger average
an alteration of the cross-sectional shape will change the shape
velocity than smaller ones, whose centers of mass can come
of the parabolic velocity profile and hence the selectivity of the
closer to the enveloping walls and hence sample more of the
separation process. Selectivity improvements have, for
low velocities. This effect has, for the first time, been observed
example, been demonstrated for channels with a triangular
by Pedersen while studying the separation of proteins in
cross section.5 At the same time, any deviation from a circular
packed glass sphere beds and has subsequently been
cross section can also be expected to worsen the axial
theoretically explained by DiMarzio and Guttmann.3−5 Their
dispersion. In the literature, the question of whether there are
theoretical work was complemented by Brenner and Gaydos,
cross-sectional shapes whose gain in selectivity can overcome
who also included the effect of the hydrodynamic interactions
with the wall and the slip velocity.6,7 Experimentally, the
technique was fully developed in the 1970s by the seminal Received: June 28, 2022
work of Small and coworkers at the Dow Chemical Company Accepted: November 2, 2022
on the size separation of polymers as well as colloidal silica Published: November 12, 2022
mixtures on packed bed columns.8,9 In the 1980s and 1990s,
Tijssen and coworkers applied the HDC principle in open
tubular channels, showing good agreement with the DiMarzo

© 2022 American Chemical Society https://doi.org/10.1021/acs.analchem.2c02766


15980 Anal. Chem. 2022, 94, 15980−15986
Analytical Chemistry pubs.acs.org/ac Article

the inevitable increase in band broadening, leading to an 2 P


w=
improvement of the overall resolution (combination of z (2)
resolution and dispersion), has, to the best of our knowledge,
not been considered yet. Equation 2 is equipped with the no-slip conditions, w = 0,
Here, we consider conduits having an angular cross section, on the walls, ∂Ω, of the cross section Ω.
that is, shapes marked by the presence of distinct corners, as The simplest approach to HDC modeling hinges on a one-
these can be expected to lead to steeper velocity gradients (see way coupling, overdamped regime assumption for particle
Figure 2 for an overview of considered shapes). The present motion. Here, the eluent flow is considered unperturbed by the
study makes use of Brenner’s macro-transport formalism to presence of the particles, and the generic individual particle is
compute the effective velocity (via Brenner’s first order regarded as a point-sized tracer ideally located at the particle
moment) and the effective dispersion (via the second order center-of-mass. Thus, in this approach, size influences particle
moment), which can be quantified by the Taylor-Aris dynamics only through wall−particle interactions. On the
dispersion coefficient16−18 Previously, this theory has been assumption that particles are spherical and rigid and hydro-
applied to study effective macrotransport in a diverse range of dynamic interactions are negligible, particle motion is identical
fields.19−24 As this is a first exploration into the present topic, to that of passive tracers onto an effective domain, Ωp, that
the solutes are treated as hard, impermeable spheres, and the takes into account particle hindrance. The effective domain Ωp
hydrodynamic effects induced by the interaction of the can be constructed from Ω by excluding all the points of the
particles with the wall and the velocity profile (wall friction, cross section closer than the particle radius to the wall along
particle rotation, concomitant lift forces, and slip) are not the local normal direction. Figure 4 shows the fluid domain
taken into account. Hence, results have been limited to the and the particle transport domain for two of the shapes that are
range of dimensionless particle size λ = dp/l ≤ 0.2 (where dp is next considered.
the particle size, and l is the cross-sectional characteristic size Let c(x, y, z, t), represent the particle number density
parameter, see Figure 2 for further details). Previous studies on (henceforth referred to as PND), where c(x, y, z, t) dV yields
the shape optimization of micro- and nanochannels that have the fraction of the total number of particles at time t falling
been conducted in the area of analytical chemistry have mainly within an elementary volume dV embedding a generic point (x,
been focused on minimizing the axial dispersion of point-sized y, z). Let c0(x, y, z) = c(x, y, z, t = 0) be a uniform distribution
solutes.25−27 Here, the finite size of the particles is taken into of particles over the cross section. By the above assumptions,
account (Figure 1). the PND obeys the micro-transport equation28

c c 1 ijj 2
2
c yzz
+w = j c+ z
t z Pep jk 2
z z{ (3)

Equation 3 has been made dimensionless with respect to a


reference concentration, and with respect to the same
characteristic length, l, and velocity, U, used in eq 2. Here,
Pep = Ul/Dp is the particle Péclet number, where Dp is the bare
Figure 1. Excluded-volume model for finite-sized particles. The particle diffusivity. The numerical solution of eq 3 is
boundaries of the fluid dynamic domain (∂Ω) and of the particle numerically challenging in view of its intrinsic multiscale
transport domain (∂Ωp) are depicted in blue and red lines, nature. Specifically, as the particle distribution becomes more
respectively. and more dispersed along the channel axis, the smallest length
scale to resolve PND gradients is still constrained by the
Let us consider a pressure-driven flow through an empty characteristic cross-sectional length l, which, for typical HDC
microchannel of arbitrary cross-section, which is henceforth applications, is several orders of magnitude smaller than the
denoted by Ω. dispersion bandwidth. Therefore, the number of degrees of
Owing to the prevailing laminar flow regime characterizing freedom necessary to preserve accuracy of the particle
microchannel flow, the pressure-driven velocity field of the distribution swiftly becomes exceedingly large. Besides, in
eluent only consists of its axial component, v = (u, v, w) = (0, chromatography applications in general, and in HDC in
0, w(x, y)), where x and y ∈ Ω are the cross-sectional particular, the information about the particle distribution
(Cartesian) coordinates throughout the channel cross section is irrelevant, since the
signal detected at the device outlet is proportional to the total
2 1 P particle number exiting the channel in a prescribed time
w=
z (1) interval. Thus, for all practical purposes, the main piece of
2 2 2 2 2
where ∇⊥ = ∂ /∂x + ∂ /∂y is the cross-sectional Laplacian information necessary to predict the separation performance is
operator. Equation 1 can be made dimensionless by taking the provided by the marginal particle distribution
characteristic dimension of the cross section, say l, as a C(z , t ) = c(x , y , z , t )dx dy , w h e r e , b y d e fi n i t i o n ,
p
reference length, the average eluent velocity
C(z,t)dz yields the fraction of particles falling in the volume
U = w(x , y)dx dy / dx dy as a reference velocity, and the dV between the cross sections at z and z + dz at time t,
characteristic viscous stress as the reference pressure, Pref = regardless of their position on the cross section. Macro-
μU/l, where μ is the dynamic viscosity of the fluid. By using transport theory shows that in the long-time asymptotic limit,
the same symbols as those used in eq 1 for the corresponding this marginal distribution obeys the effective transport
dimensionless variables, the latter equation is written as equation28
15981 https://doi.org/10.1021/acs.analchem.2c02766
Anal. Chem. 2022, 94, 15980−15986
Analytical Chemistry pubs.acs.org/ac Article

Figure 2. Dimensionless velocity profile w(x,y)/U. Panels (a−c) depict circle-, square-, and star-shaped cross sections. Panels (d−g) depict
triangular shapes characterized by different values of the bottom angle α, α = 70.6, 60, 45, and 30°, respectively.

2 b-field (so termed after Brenner). The relationship between the


C C 1 C
+ Wp = eff 2 b-field and the Taylor-Aris coefficient can be expressed as
t z Pep z (4)
( b) × ( b)dx dy
where the dimensionless transport parameters, Wp and 1/Peeff p p
p =
are referred to as effective particle velocity and effective particle dx dy
Péclet number, respectively. The effective velocity is given by p (8)
the eluent velocity averaged over the effective cross section of The knowledge of the effective particle velocity and of the
the channel Taylor-Aris coefficient for any given particle size allows us to
w(x , y)dx dy predict the column length necessary to achieve the separation
p of a size-dispersed suspension. Specifically, the separation
Wp =
dx dy resolution of a mixture of particles of two different sizes, say
p (5) “1” and “2” in a channel of length L can be expressed as
where the dimensionless dispersion coefficient 1/Peeff
p is given 1 1
by28 1 W2 W1
Rs =
1 1 2 2(1 + 1Pe12)
+ 2(1 + 2Pe 2 2)
eff
= + pPep Pe1(W1)3 Pe 2(W2)3 (9)
Pep Pep (6)
By assuming that the bare particle diffusivity can be
It can be observed that the inverse effective particle Péclet estimated from the Stokes−Einstein relationship,
number coincides with the Height Equivalent of the = k bT /(3 d p ) (where kb, T, dp, and μ are the Boltzmann
Theoretical Plate (HETP), say Hp, normalized with respect
to the characteristic cross-sectional length l, that is, hp = Hp/l = constant, the absolute temperature, the particle diameter, and
1/Peeff the dynamic viscosity of the fluid, respectively), the fact that
p . The constant Γp appearing at the right hand side of eq
6 is referred to as Taylor-Aris coefficient, and its value controls both particle types are entrained in the same channel flow
the increase of the dispersion bandwidth at values of particle implies that Pe 2 = Pe1(d p2/d p1). Thus, if the effective particle
Péclet number larger than 1/ p . Therefore, geometries transport parameters are known, and eq 9 allows us to
determine the minimal length, Lmin, yielding a prescribed level
providing the lowest Taylor-Aris coefficient that are compat- of separation resolution.
ible with operating and constructive constraints are sought
after. Besides this, the macro-transport theory also provides an 2. MATERIALS AND METHODS
explicit expression for Γp in terms of the solution of the steady-
The geometries considered throughout are depicted in Figure
state (elliptic) equation28
2, which also shows the dimensionless axial velocity profile
2 w(x,y)/U obtained by solving eq 2 using a commercial finite-
b = Wp w(x , y) (7)
element software (Comsol Multiphysics 5.5) onto an
equipped with vanishing normal gradient at the boundary of unstructured mesh on the order of 105 degrees of freedom.
the effective particle transport domain, ∂b/∂n = 0 on the The dispersion data shown below were computed by solving
boundary of Ωp. The dependent variable b(x,y) is referred to as eq 7 using the same Comsol Multiphysics 5.5 software on a
15982 https://doi.org/10.1021/acs.analchem.2c02766
Anal. Chem. 2022, 94, 15980−15986
Analytical Chemistry pubs.acs.org/ac Article

Figure 3. (A) Effective velocity and (B) selectivity Wp′ (definition: see text) vs the dimensionless particle radius l for different cross-sectional
shapes. (C) Corresponding plot of dimensionless residence time τ vs l. The labeling is consistent with that of Figure 2. The dashed line in panel C
represents the case of a flat-rectangular channel (infinite width). The reference case (circular channel) is colored in red.

triangular unstructured mesh corresponding to an order of 106 dependence of Wp on λ, whereas the star and the triangle shape
degrees of freedom. Specifically, the steady-state PARDISO associated with α = 70.6° [curves (c,d), respectively] exhibit
solver algorithm was used with the relative tolerance set to the most sensitive dependence of the effective particle velocity
10−5. In the case of shapes with sharp corners (triangular and on the particle size. The remaining triangle shapes display a
star), mesh refinement was enforced in the near-corner regions behavior intermediate between these two extremes. This result
of the domain. is qualitatively consistent with the axial velocity profiles
We observe that, beyond the parabolic profile associated depicted in Figure 2, where the shapes (c) and (d) display
with the circular channel, analytical expressions are also the highest values of the eluent maximum velocity when
available for the square [panel (b)] and the equilateral triangle compared to circular and squared shapes. Clearly, geometries
[panel (d)] in terms of Fourier expansion. Nevertheless, since yielding a swift increase of Wp with Rp are sought after for
the solution of the two-dimensional Stokes equation poses no maximizing the separation efficiency. In fact, if one considers
severe numerical challenge, we used the same numerical two particle types of size λ and λ + Δλ, the difference Wp(λ +
approach for all the shapes. All of the shapes, except for the Δλ) − Wp(λ + Δλ) controls the distance, say δpeaks, between
star-shaped geometry of panel (c), can in principle be obtained the peaks of the Gaussian-shaped marginal distributions C(z,t)
by lithography-etching microfabrication. Specifically, the cross- of each particle type, which grows linearly with time as
sectional geometry in panel (d) can be directly obtained by wet peaks = (Wp( + ) Wp( + ))t and ultimately deter-
KOH etching of bulk silicon.29 The star-shaped geometry can mines the distance between the average residence times in the
in principle be obtained by 3d (nano)printing. One is to note chromatogram for a column of given length. By this
how the scaled value of the maximum axial velocity depends on observation, an intrinsic measure of the driving force for the
the cross-sectional shape. The highest values are those separation is represented by the derivative W′p = dWp/dλ,
associated with the star- and triangle-shaped geometries which can be referred to as the selectivity of the HDC
depicted in panels (c,d), respectively. separation and is depicted in panel (B) of Figure 3. Here, the
difference between the standard channel shapes (circular and
3. RESULTS AND DISCUSSION squared) and the triangular and star-shaped cross sections can
The results showed next are obtained through the finite- be appreciated quantitatively. For small particle diameters of λ
element solution of the b-field eq 7 enforced onto the same < 0.12, the best performing shape is the triangular one with α =
computational domain as that used to compute the axial 70.6°. For larger particle radii, however, the selectivity
velocity profile. associated with this shape drops rapidly, and falls below that
3.1. Effective Particle Velocity and Separation of the star cross-sectional geometry. By contrast, the selectivity
Selectivity. Panel A of Figure 3 shows the dimensionless of the star shape displays a nonmonotonic behavior and shows
effective particle velocity Wp (scaled with respect to the eluent an overall behavior that is relatively insensitive to the particle
velocity) versus the relative particle radius λ, defined with size. On a practical level, this means that the separation
respect to the characteristic length of the cross section (λ = dp/ performance would essentially be equally effective for all
(2l), where dp is the particle diameter). The different curves particle sizes in the range considered, which, in turn, allows us
refer to the shapes depicted in Figure 2 and are labeled to use HDC for a particle suspension characterized by a wide
consistently with this figure. size distribution. To verify with the existing HDC literature,
Note that values beyond λ = 0.2 were not considered, as the panel (C) of Figure 3 plots the dimensionless residence time τ
enforcement of the one-way coupling approximation becomes (=1/Wp) versus λ.1,7,11 For a cylindrical tube, it can be
questionable for such large particles and two-phase effects rigorously shown that, for the simplified case of a hard,
might become important. The most common shapes, namely nonrotating sphere considered here, τ is given by1
the cylindrical capillary and the square duct represented by 2 1
curves (a,b), respectively, are those showing the weakest = (1 + 2 ) (10)

15983 https://doi.org/10.1021/acs.analchem.2c02766
Anal. Chem. 2022, 94, 15980−15986
Analytical Chemistry pubs.acs.org/ac Article

This equation fits to within 99.9% with our computed result attained at the boundary of the effective transport domain to
for the cylindrical tube case, thus validating the presently the (negative) minimum value attained at the point of the core
proposed procedure. The fact that the (λ,τ)-curves for the channel region, where the eluent velocity reaches its maximum
other considered shapes are flatter than the one for the circular value. In general, the larger the range of values of Wp − w(x,y),
tube reconfirms the superior selectivity of the angular shapes the higher the average magnitude of ∇b. When the particle size
already clear from panels (A) and (B). For the sake of increases, the maximum value decreases in view of the fact that
comparison, we also added (dashed line) the selectivity of a the boundary of the effective transport domain for the particle
flat-rectangular channel.12 As can be noted, this has the worst moves away from the channel walls, whereas the minimum
selectivity of all possible shapes. This is due to the blunt shape value remains fixed. As a result, the range of values on the
of its velocity field parabola. right-hand side of eq 7 decreases as the particle size increases,
3.2. Taylor-Aris Coefficient and Scaled HETP. This as does the magnitude of the gradient of the b-field. More
much is established for the effective particle velocity, and the interesting is the structure of the integrand function ∇b × ∇b
resolution of HDC separation also depends on the dispersion (which could be regarded as the local source of axial
bandwidth, as is evident from eq 9. We, therefore, next focus dispersion) for a fixed particle size when the shape of the
on how the relationship between the Taylor-Aris coefficient, channel cross section is varied. Figure 5 shows the spatial
Γp, and the relative particle size depends on the cross-sectional
shape. Figure 4A shows the results for the different geometries
considered obtained from the expression in eq 8.

Figure 5. (∇⊥b) × (∇⊥b) for different channel cross sections. Same


panel labeling as in Figure 2.

contour of this function for the different geometries of a


particle of dimensionless radius Rp = 0.05. The white areas in
the contour depict the excluded volume that is inaccessible to
Figure 4. (A) Taylor-Aris coefficient Γ vs dimensionless particle the center of the particle. The highest values of the gradient
radius for different channel geometries. (B) Evolution of dimension- magnitude are reached in the wedge-shaped triangle (g) (note
less plate height hp with Pep for λ = 0.2. Curve labeling is consistent that the legend scale is different for the different shapes). By
with Figure 1. The reference case (circular channel) is colored in red. taking the circular capillary as a benchmark to understand the
localization behavior of the gradient of the b-field, where ∇b
In this case, the performance of the different shapes does not only possesses the radial component, one observes the
follow the same ordering as that observed for the selectivity. seemingly counter-intuitive phenomenon that the areas
The best-performing shapes (i.e., those associated with lowest contributing the most to axial dispersion are those where the
values of Γ) are the triangles (d) and (e), whereas the right-hand side of eq 7 is close to zero (i.e., where the axial
remaining wedge-shaped triangles and the squared cross eluent velocity is close to its average value). This apparent
section yield the highest Γp-value and therefore the largest paradox can be explained by the boundary conditions of the b-
dispersion. All the shapes share the same decreasing tendency field at the boundary of the effective transport domain, where
of the Taylor-Aris coefficient with the increasing particle size. the normal gradient of b(x,y) is forced to assume vanishing
To link the Γp coefficients to the general dispersion behavior, values. Also, theoretical arguments can be brought up to show
panel B of Figure 4 translates the Γp values into a plot of the that the gradient of the b-field must vanish at the center of the
dimensionless plate height hp versus Pep for the case of a channel, where the axial velocity reaches its maximum.
particle of dimensionless diameter λ = 0.2. The monotonically Combining this observation with the behavior of ∇b at the
decreasing trend of Γp with the particle size, which has been boundary of the effective domain, one concludes that the
observed in a variety of channel flows, can be explained based gradient of the b-field is constrained to become localized at an
on the following argument. From eq 8, the dispersion- intermediate value of the radial coordinate, where, incidentally,
enhancing effect caused by the axial flow is given by the the source term that ultimately causes the very presence of the
average value of the squared gradient of the b-field, which is gradient attains vanishing values.
obtained as the solution of eq 7. In this latter equation, the The situation is slightly more complex for the noncircular
magnitude of ∇b is related to the range of values attained by shapes, where the above argument strictly holds true only
the source term Wp − w(x,y) appearing at the right-hand side when ∇b is orthogonal to the boundary of the effective
of the equation, which ranges from the largest (positive) value transport domain. One notes that nonvanishing values of the
15984 https://doi.org/10.1021/acs.analchem.2c02766
Anal. Chem. 2022, 94, 15980−15986
Analytical Chemistry pubs.acs.org/ac Article

gradient magnitude are found in all of the shapes because the the overall efficiency response is readily explained by the
tangential component of ∇b in not zero in these cases. Both behavior of the Taylor-Aris coefficient for the two shapes (see
the values and the localization patterns of the gradient of the b- Figure 4, which shows that the value of Γ associated with shape
field are sensitive to the specific geometry. Specifically, by (e) is significantly lower than that of shape (c), thus implying
comparing the data, one notes how a small variation of the that the overall efficiency of the separation is more influenced
bottom angle causes a relatively large variation of ∇b × ∇b. by the axial dispersion than by the velocity difference for these
3.3. Resolution of a Two-Particle Mixture. The data particle sizes.
discussed above are all reported in their nondimensional form, The same conclusions apply to the minimal separation time,
and selectivity and dispersion are discussed separately. In order as depicted in Figure 6B, which decreases with increasing
to clarify their impact on the resolution in a practical HDC eluent velocities until it reaches a plateau, where the positive
separation, we next consider the resolution of a two-particle effect of the reduced residence time in the column at increasing
suspension in open-channel HDC, where the characteristic eluent velocities is compensated by the efficiency-hindering
cross-sectional length is fixed to l = 1 μ and the sizes of the two effect of the increased axial dispersion associated with the
particle types “1” and “2” are assumed to be equal to dp(1) = 50 Taylor-Aris convection-dominated regime. Quantitative read-
nm and dp(2) = 100 nm. As observed in Section 2, particles of outs of the gain factors are given in Table 1, showing that gain
different sizes entrained in one and the same channel flow are
characterized by different Péclet numbers, which are in the Table 1. Performance Gain with Respect to the Cylindrical
same ratio as the particle diameters. Therefore, the dimen- Geometry for 3 Selected Eluent Velocities for the Different
sional values of the particle effective velocity and of the Shapesa
dispersion coefficient entering the expression of the separation
μm/s square star α = 70.6° α = 60° α = 45° α = 30°
resolution must be properly rescaled. The performance of the
different channel shapes is compared both in terms of the (A)
minimal length, Lmin needed to achieve unit resolution of the 101 1.05 3.18 3.36 2.8 2.2 1.7
two-particle mixture and of the corresponding separation time, 102 0.8 2.85 3.77 2.9 1.67 0.58
103 0.61 2.5 4.37 2.97 1.27 0.3
based on the slowest (smaller) particle type tmin = (1) + t (1)
(B)
, where σ(1) and t(1) are the variance and the average residence 101 1.05 3.3 3.56 2.93 2.27 1.73
time of the chromatogram of particles “1”, respectively. The 102 0.8 2.98 3.99 3.02 1.72 0.59
data reporting these quantities for the different shapes as a 103 0.616 2.62 4.64 3.12 1.32 0.32
function of the eluent velocity are shown in Figure 6. In terms a
(A) Ratio between the channel length to obtain unit resolution with
respect to that of the cylindrical capillary. (B) Ratio between the
separation time to obtain unit resolution with respect to that of the
cylindrical capillary.

factors on the order of 3−4.5 can be obtained with respect to


the cylindrical channel. The figures also show the separation
time in the KOH-etched shape is almost one order of
magnitude smaller than that of the cylindrical capillary and
more than one order of magnitude smaller with respect to that
of the squared channel.

4. CONCLUSIONS
Performing HDC separations in channels with an angular cross
Figure 6. Plots of (A) minimum channel length required to achieve
section can significantly increase the separation resolution and
unit resolution vs eluent velocity U and (B) corresponding separation speed compared to the customarily considered cylindrical
time vs U. The characteristic length of the cross section is set to l = 1 capillary. This improvement is owed to the presence of the
μm, and the diameter of particles “1” and “2” is set to to dp(1) = 50 nm corners, which promote the formation of regions of low
and dp(2) = 100 nm, respectively. The reference case (circular velocity, where the smaller particles are arrested with respect to
channel) is colored in red. the mean velocity. At the same time, the velocity in the central
region is, for a given average fluid velocity, higher in the shapes
of the required separation length (Figure 6A), the square with sharp corners than in the circular capillaries to
channel and the wedge-shaped triangle with α = 30° perform compensate for their larger fraction of low velocity regions.
less well than the reference case, that is, the cylindrical Both effects combined lead to a larger spread in average
capillary. The best performing shapes are the triangular (α = velocity between larger and smaller particles, thus enhancing
70.6 and α = 60°) cross sections [curves (d) and (e), the separation selectivity. On the other hand, the presence of
respectively]. It is worth highlighting how, for the particle sizes the corners and its concomitant effect on the velocity gradients
considered (corresponding to dimensionless particle radii R(1) p leads to an increase in dispersion. Using Brenner’s macro-
= 0.025and R(2) p = 0.05), the average selectivity of the star- transport theory, it could be shown that the best overall
shaped channel (c) is higher than that of the equilateral compromise between selectivity and dispersion is found in the
triangle (e), as can be gathered from the data depicted in triangular shapes, with α = 60° and especially α = 70.6°, which
Figure 3; therefore, based only on this parameter, one would can be obtained by KOH etching of bulk silicon, being the best
expect the star-shaped channel to be more efficient. Besides, performing. The latter has the potential to increase the
15985 https://doi.org/10.1021/acs.analchem.2c02766
Anal. Chem. 2022, 94, 15980−15986
Analytical Chemistry pubs.acs.org/ac Article

separation speed by a factor of 4.64 compared to the (28) Edwards, D.; Brenner, H.Macrotransport Processes; Butterworth-
conventional cylindrical capillary case. Heinemann: Woburn, MA (USA), 1993.
(29) Barycka, I.; Zubel, I. Sens. Actuators, A 1995, 48, 229−238.

■ AUTHOR INFORMATION
Corresponding Author
Gert Desmet − Department of Chemical Engineering, Vrije
Universiteit Brussel, B-1050 Brussels, Belgium; orcid.org/
0000-0001-8781-7184; Email: gert.desmet@vub.be
Authors
Valentina Biagioni − Dipartimento di Ingegneria Chimica
Materiali Ambiente, Sapienza Università di Roma, 00184
Roma, Italy
Stefano Cerbelli − Dipartimento di Ingegneria Chimica
Materiali Ambiente, Sapienza Università di Roma, 00184
Roma, Italy; orcid.org/0000-0003-3906-6595
Complete contact information is available at:
https://pubs.acs.org/10.1021/acs.analchem.2c02766

Notes
The authors declare no competing financial interest.

■ REFERENCES
(1) Striegel, A. M.; Brewer, A. K. Annu. Rev. Anal. Chem. 2012, 5,
15−34.
(2) Brewer, A. K. Chromatographia 2021, 84, 807−811.
(3) Pedersen, K. O. Arch. Biochem. Biophys., Suppl. 1962, 1, 157−
168.
(4) DiMarzio, E. A.; Guttman, C. M. Polym. Lett. 1969, 7, 267−272.
(5) DiMarzio, E. A.; Guttman, C. M. Macromolecules 1970, 3, 131−
146.
(6) Brenner, H.; Gaydos, L. J. J. Colloid Interface Sci. 1977, 58, 312−
356.
(7) Blom, M. T.; Chmela, E.; Oosterbroek, R. E.; Tijssen, R.; van Recommended by ACS
den Berg, A. Anal. Chem. 2003, 75, 6761−6768.
(8) Small, H. J. Colloid Interface Sci. 1974, 48, 147−161. Online Quaternized Derivatization Mapping and Glycerides
(9) Small, H.; Langhorst, M. Anal. Chem. 1982, 54, 892A−898A. Profiling of Cancer Tissues by Laser Ablation Carbon Fiber
(10) Tijssen, R.; Bleumer, J. P. A.; Van Kreveld, M. E. J. Chromatogr. Ionization Mass Spectrometry
A 1983, 260, 297−304.
Yingjie Lu, Yinlong Guo, et al.
(11) Tijssen, R.; Bos, J.; Van Kreveld, M. E. Anal. Chem. 1986, 58,
FEBRUARY 22, 2022
3036−3044. ANALYTICAL CHEMISTRY READ
(12) Chmela, E.; Tijssen, R.; Blom, M. T.; Gardeniers, H. J. G. E.;
van den Berg, A. Anal. Chem. 2002, 74, 3470−3475.
(13) Op de Beeck, J.; De Malsche, W.; Vangelooven, J.; Gardeniers, Isotopic Peak Index, Relative Retention Time, and Tandem
H.; Desmet, G. J. Chromatogr. A 2010, 1217, 6077−6084. MS for Automated High Throughput IGF-1 Variants
(14) Op de Beeck, J.; De Malsche, W.; De Moor, P.; Desmet, G. J. Identification in a Clinical Laboratory
Sep. Sci. 2012, 35, 1877−1883. Ievgen Motorykin, Zengru Wu, et al.
(15) Chamieh, J.; Leclercq, L.; Martin, M.; Slaoui, S.; Jensen, H.; AUGUST 17, 2021
Østergaard, J.; Cottet, H. Anal. Chem. 2017, 89, 13487−13493. ANALYTICAL CHEMISTRY READ
(16) Brenner, H. Philos. Trans. R. Soc., A 1980, 297, 81−133.
(17) Brenner, H. J. Fluid Mech. 1989, 204, 97−119. A Universal Eluent System for Method Scouting and
(18) Aris, R. Philos. Trans. R. Soc., A 1956, 235, 67−77. Separation of Biotherapeutic Proteins by Ion-Exchange, Size-
(19) Dorfman, K. D.; Brenner, H. Biomed. Microdevices 2002, 4, Exclusion, and Hydrophobic Interaction Chromatography
237−244.
(20) Chu, H.; Garoff, S.; Tilton, R. D.; Khair, A. S. J. Fluid Mech. Alexander B. Schwahn, Ken Cook, et al.
NOVEMBER 16, 2022
2021, 917, A52.
ANALYTICAL CHEMISTRY READ
(21) Cerbelli, S.; Giona, M.; Garofalo, F. Microfluid. Nanofluidics
2013, 15, 431−449.
(22) Adrover, A.; Cerbelli, S. Phys. Fluids 2017, 29, 062005. Design Guidelines and Kinetic Performance Limits for
(23) Adrover, A.; Venditti, C.; Giona, M. Phys. Fluids 2019, 31, Spatial Comprehensive Three-Dimensional Chromatography
062003. for the Analysis of Intact Proteins
(24) Santamaría-Holek, I.; Hernández, S. I.; Alcántara, C. G.; Durán, Thomas Themelis, Sebastiaan Eeltink, et al.
A. L. Catalysts 2019, 9, 281. SEPTEMBER 02, 2022
(25) Golay, M. J. E. J. Chromatogr. A 1981, 216, 1−8. ANALYTICAL CHEMISTRY READ
(26) Dutta, D.; Leighton, D. T. Anal. Chem. 2001, 73, 504−513.
(27) Dutta, D.; Leighton, D. T. Anal. Chem. 2003, 75, 57−70. Get More Suggestions >

15986 https://doi.org/10.1021/acs.analchem.2c02766
Anal. Chem. 2022, 94, 15980−15986

You might also like