2.1.

Materials and Instrumentation

The modified BPA aptamer,¹⁷ 5’-$[\mathrm{ThiSS}]{[\mathrm{HEG}]}_3$ CCG CCG TTG GTG TGG TGG GCC TAG GGC CGG CGG CGC ACA GCT GTT ATA GAC GTC TCC AGC-3’, was synthesized by Integrated DNA Technologies. Hetero-bifunctional PEG linker (${\mathrm{NH}}_2$-PEG-SH, 1 kDa) was purchased from NanoCS. MGITC reporter dye was purchased from Invitrogen (UK). All other reagents were obtained from Sigma Aldrich (USA/UK).

A Varian Cary 300Bio UV–visible spectrophotometer was used with a scan range of 200 to 800 nm for extinction measurements. The zeta potential and hydrodynamic diameters of the nanoparticles were measured on a Zetasizer Nano ZS90 (Malvern, UK). All microplate Raman and SERS spectra were collected using a ThermoScientific DXR Raman confocal microscope with a $900\text{-}\mathrm{g}/\mathrm{mm}$ grating. SERS mapping of the microfluidic channels was done using a WITec Alpha 300R confocal Raman instrument (WITec GmbH, Ulm, Germany) with a $600\text{-}\mathrm{g}/\mathrm{mm}$ grating, fitted with a piezodriven $XYZ$ scan stage. All samples were probed using a laser wavelength of 532 nm and coupled to a thermoelectrically cooled charge-coupled device (CCD).

2.2.

Colloid Synthesis

Silver colloid (AgNP) was synthesized using the “cold” method reported by Leopold and Lendl.²¹ Hydroxylamine hydrochloride (1 mL, 150 mM) was added to 89 mL of NaOH (3.33 mM) under vigorous stirring. Silver nitrate (${\mathrm{AgNO}}_3$) solution (10 mL, 10 mM) was added drop-wise and stirred for 15 min at room temperature. Dynamic light scattering (DLS) measurements revealed an average particle diameter of $\sim 45\mathrm{nm}$ (not shown) with a PDI index of 0.133. The stock particle concentration was determined to be 225 pM according to Beer’s law using an extinction coefficient of $2.87\times {10}^{10}{\mathrm{M}}^{-1}{\mathrm{cm}}^{-1}$ at 404 nm.²²

Silver-coated cobalt-ferrite nanoparticles ($\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{CoO}}_4$) were prepared by first synthesizing a stock solution of the core ${\mathrm{Fe}}_2{\mathrm{CoO}}_4$ nanoparticles through co-precipitation of iron(III) chloride (0.2 M ${\mathrm{FeCl}}_3$) and cobalt(II) chloride (0.1 M ${\mathrm{CoCl}}_2$) in a sodium hydroxide solution at $\mathrm{pH}\sim 12$ (3.0 M NaOH) using a method modified from Donnelly et al.¹⁶ The cobalt ferrite salt solution was added rapidly into 3.5 M NaOH solution under vigorous stirring, then headed at 80°C for 1 h. The cobalt-iron oxide precipitated colloid was washed with deionized water in triplicate using a permanent neodymium magnet. To coat with silver, $500\mu \mathrm{L}$ of these stock MNPs was mixed with 4 mL of 0.35 M glucose and 1.5 mL of $60\mu \mathrm{M}$ ${\mathrm{AgNO}}_3$. The solution was sonicated for 10 min then heated to 90°C for 90 min. Finally, the particles were centrifuged three times to wash and redispersed in 6 mL of 5 mM sodium citrate.

Silver-coated ferrite nanoparticles ($\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{O}}_3$) were prepared using a coprecipitation method for the synthesis of the particle core and glucose reduction to coat with silver as described by Kumar et al.²³ and Mandal et al.,²⁴ respectively. Briefly, a stock solution of maghemite ($\gamma \text{-}{\mathrm{Fe}}_2{\mathrm{O}}_3$) nanoparticles was prepared by adding 25 mL of an acidified iron salt solution (0.4 M ${\mathrm{Fe}}^{2+}$, 0.8 M ${\mathrm{Fe}}^{3+}$, 1 M HCl) drop-wise to 250 mL of 1.5 M NaOH at 50°C under vigorous nonmagnetic stirring. After 20 min, the particles were allowed to cool, washed twice with DI water, and washed once with 0.1 M ${\mathrm{HNO}}_3$. An additional 125 mL of ${\mathrm{HNO}}_3$ was then added to the solution, stirred an additional 40 min at 95°C, and resuspended in distilled water. To coat with silver, 1 mL of these stock MNPs was mixed with 4 mL of 0.35 M glucose and 1.5 mL of $60\mu \mathrm{M}$ ${\mathrm{AgNO}}_3$. The solution was sonicated for 10 min then heated to 90°C for 90 min. Finally, the particles were centrifuged three times to wash and redispersed in 6 mL of 5 mM sodium citrate.

2.3.

Target and Probe Nanoparticle Conjugation

2.4.

Analysis of Competitive Binding Assay in a Microwell

All microwell measurements were collected using the Thermo-Scientific DXR Raman (600 with a spectral range from 400 to $1800{\mathrm{cm}}^{-1}$) in a total volume of $30\mu \mathrm{L}$) in 0.1 M PBS (pH 7.4) buffer with a 532-nm laser power of 10 mW and an integration time of 10 s ($10\times 1\mathrm{s}$ exposures). A solution containing equal volumes of 225 pM of each of the target and probe nanoparticles was monitored with SERS for 5 h to allow complete binding between the immobilized BPA aptamers and BADGE. The assembled nanoparticle clusters were collected using a neodymium magnet held at the side of the glass vial, the supernatant containing any unbound AgNPs was removed, and the assembled nanoparticles were resuspended in 0.1 M PBS (pH 7.4). For confirmation of the formation of nanoclusters using DLS, $10\mu \mathrm{L}$ of the assembled target/probe NPs were mixed and allowed to shake for 3 h before collecting size measurements. To monitor competitive binding with SERS, $20\mu \mathrm{L}$ of the assay clusters were mixed with $10\mu \mathrm{L}$ of BPA (1 pM to $1\mu \mathrm{M}$) in 0.1 M PBS (pH 7.4), and SERS measurements were taken every 30 s for 6 min.

2.5.

Fabrication of Ni-Patterned Magnetic Microchannel

Each magnetofluidic SERS chip is composed of a 200-nm-thick deposited nickel patterned onto a glass microscope slide, a PDMS microfluidic channel, and two permanent neodymium magnets housed around the PDMS chip by a 3-D printed casing. COMSOL was used to simulate the distribution of induced magnetic field for different designs by mapping the theoretical magnetic flux density for the experiment and are shown in Fig. 2. From modeling, it was revealed that at least two pads are necessary to evenly distribute the magnetic field in the channel, which would lie perpendicular to the gap line in the red zone of 2(a). Figure 2 shows the initial two designs investigated: design A “the straight pattern” [Figs. 2(a) and 2(c)] and design B “the micromagnet array” pattern [Figs. 2(b) and 2(d)]. The magnetic field across the patterns (where the PDMS channel will be overlaid) is induced by the two attractive permanent neodymium magnets on either side of the pads, as shown in the schematics in Figs. 2(c) and 2(d). In these schematics, the polarity for the top magnet is north and for the bottom magnet is south; therefore, the polarity of the induced field across the pattern is reverse (top of the pattern is the south pole and bottom is north pole). This is due to the induced magnetic field propagating down the deposited nickel from the field, looping from the top to the bottom external magnets. From the magnetic flux density maps in Figs. 2(a) and 2(b), it is observed that the field is strongest at the centers and edges of the nickel pads for both patterns; thus, it is predicted that the majority of magnetic nanoparticles will be trapped at the “warm” locations within the patterns where the magnetic flux density is the strongest. Since each micropad has its own distribution of magnetic flux density, each micropad can be considered as an individual micromagnet, thereby creating a series of “magnetic microwells” when housed in a fluidic chip. It is noted that it is desirable that the magnetic field is not too strong as to aggregate the particles irreversibly on the surface of the nickel. A weaker field also allows space for competitive binding to cause the release of a nonmagnetic nanoparticle without it being captured by another magnetic probe downstream. According to Do et al.,^26,27 the magnetic field distributed to Ni-pads with nanaometer-thickness should distribute vertically with a height of $\sim 20\mu \mathrm{m}$.

Fig. 2

COMSOL simulation results of the magnetic flux density intensity distribution for (a) straight Ni-pads and (b) an array of Ni-pads, demonstrating that the magnetic field is stronger for a planar field of Ni but is more evenly distributed for the array pattern. It is desired that the magnetic field is uniform but not too strong as to aggregate the particles irreversibly on the surface of the nickel. Schematic diagram for (c) straight and (d) array micromagnet design patterns. $A=1\mathrm{cm}$, $G=50\mu \mathrm{m}$

To fabricate the chips, the nickel is deposited onto a glass microscope slide by photolithography and electron beam deposition using a methodology derived from Guven et al.¹⁵ Briefly, glass slides are spin-coated with lift-off resist (LOR) 3A and S-1813 LOR at 750 nm and $1.3\mu \mathrm{m}$, respectively. After exposure and development, the pattern of the Ni micromagnet array is visibly transferred onto the glass slides. Next, 100-nm-thick chromium and 50-nm-thick copper are continuously deposited on the pattern as the adhesion layer, and then 200-nm-thick nickel is deposited as the third and final “magnetically responsive layer.” After deposition, the entire glass slide is placed into the chemical stripper of LOR 3A at 80°C to remove the photoresist layer. A $100\text{-}\mu \mathrm{m}$-thick microfluidic channel is constructed using a silicon wafer mold via soft lithography. Uncured-PDMS solution ($10:1$) is poured over the silicon wafer mold and cured at 65°C for 2 h as per typical soft lithography methods.¹⁶ Finally, the PDMS microfluidic channel is bound to the glass slide patterned with a micromagnet array by plasma etching treatment. The nickel pads within the channel are not magnetically activated until the final 3-D printed neodymium magnet holder is placed around the channel.

2.6.

SERS Analysis of Prebound Assay Particles Housed in Magnetofluidic Chip

A $200\text{-}\mu \mathrm{L}$ solution containing equal volumes of 225 pM of each of the target and probe nanoparticles in a total volume of $300\mu \mathrm{L}$ 0.1 M PBS was allowed to flow through the Ni-patterned magnetic channels at a 2 mL syringe pump rate of $10\mu \mathrm{L}/\min$, and subsequentially left overnight to dry. After reintroducing particles to 0.1 M PBS, the SERS profile of the localized assembled nanoparticle clusters were mapped using WiTec Raman analysis software for chip designs A & B and using Thermo-Scientific DXR Raman Omnic mapping software for chip design C. Both the WiTec and Thermo-Scientific Raman microscopes use a $10\times$ objective (NA 0.25), $600\text{lines}/\mathrm{mm}$ grating, and a 10-mW 532 nm laser, where all SERS measurements were collected with an integration time of 2 s ($2\times 1\mathrm{s}$ total exposure) and a step size of $10\mu \mathrm{m}$. Seventeen maps were collected for a total 3-D mapping area of $3\times 300\times 170\mu \mathrm{m}$ ($XYZ$) for chip design A. Additionally, 2-D maps were scanned for areas of $75\mu \mathrm{m}\times 75\mu \mathrm{m}$ ($XY$) for design B and $70\mu \mathrm{m}\times 350\mu \mathrm{m}$ ($XY$) and $70\mu \mathrm{m}\times 70\mu \mathrm{m}$ ($XZ$) for chip design C. Raman spectral maps of the MGITC peak at $1175{\mathrm{cm}}^{-1}$ were processed using Origin Pro 2015 graphing software.

3.1.

Target and Probe Nanoparticle Conjugation

As shown from the SEM images and DLS plots in Fig. 3, the average hydrodynamic particle diameter of probe 1 ($\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{O}}_3$, nanoparticles coated in BADGE) was determined by DLS to be $\sim 70\mathrm{nm}$, with a final concentration of 425 pM as determined by a Nanosight NTA. For probe 2, the iron core was doped with cobalt to increase the magnetization of the probe particles and improve the collection time. Probe 2 $\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{CoO}}_4$ particles were $\sim 63\mathrm{nm}$ in diameter, and final stock concentration was 385 pM as determined by NTA. The extinction profile of plain AgNPs compared to the magnetic core–shell particles ($\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{CoO}}_4$ and $\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{O}}_3$) is shown in Fig. 3(c) for reference.

Fig. 3

(a) SEM images of the two ferric probe nanoparticles after coating in silver to enhance their plasmonic properties at 532 nm, (b) DLS size distribution, concentration, and zeta potential data demonstrating particle size, yield, and stability, and (c) comparison of the extinction profile of the target solid silver target nanoparticles (black dots) with the two silver coated Fe (blue dash) and FeCo (red line) probe nanoparticles.

A colloid’s zeta potential ($\zeta$) is indicative of the relationship between the particles’ surface charge and their ionic environment and thus can be used to predict long-term stability. The Ag target nanoparticles were functionalized with aptamers and MGITC as described and yielded an average $\zeta$ of $-39.7\mathrm{mV}$, where $\zeta >-20\mathrm{mV}$ indicates sufficient colloidal stability. Silver-coated probe particles of the ${\mathrm{Fe}}_2{\mathrm{O}}_3$ and ${\mathrm{Fe}}_2{\mathrm{CoO}}_4$ varietals-coated in PEGylated BADGE were found to have zeta potentials of $-32.1$ and $-22.7\mathrm{mV}$, respectively (Fig. 3), indicating that the PEG spacer and the HEG modified aptamer provide adequate particle stabilization in the 0.1 M PBS binding buffer but that the Co-doped particles were slightly less stable.

3.2.

Analysis of Competitive Binding Assay in a Microwell

The SERS response of the probe and target nanoparticles as they form assembled assay nanoclusters, along with the nanocluster assay’s subsequent negative SERS response when introduced to competing BPA, was first analyzed in a traditional microwellplate. As shown in Fig. 4(a), the SERS signal of the reporter molecule MGITC exhibits a 50-fold increase in SERS intensity as it experiences an increase in the electron density in its immediate environment due to the aptamer-induced plasmonic nanoparticle assembly, reaching a steady-state equilibrium after $\sim 2\mathrm{h}$. The DLS results in Fig. 4(b) show that upon probe and target binding, the average size of the colloid increases, which can be interpreted as formation of the assay nanoclusters. Bringing the nanoparticles closer together as the DLS indicates increases the net field intensity as hot spots form, which leads to the increase SERS intensity.

Fig. 4

Schematic of nanocluster assembly (a) SERS spectral intensity from tagged target nanoparticle, conjugated to either a BPA specific (black circle) or a nonspecific (red square) aptamer sequence, monitored for 5 h after exposure to the Ag@Fe probe nanoparticles at a $1:1$ molar ratio. Correlation coefficient of the exponential fit of the stagnant nanoparticle assembly is $r^2=0.94$ and reaches equilibrium after ~2 h. (b)Verification by DLS that SERS enhancement is facilitated by nanoparticle assemblies, shown before and after for each probe type.

When free BPA is added to the assay nanoparticle assembly solution, it competes with BADGE on the surface of the probe particle for the aptamer binding sites on the target particle (Fig. 5). This dissociation is verified by the decrease in SERS intensity of the peak at $1175{\mathrm{cm}}^{-1}$ [aromatic C–H bending vibrational mode,²⁸ Fig. 5(a)] and decreases to a steady state over the course of $\sim 5\min$ as competitive binding occurs [Fig. 5(b)]. This implies that the BPA aptamer immobilized on the target AgNPs must “loosen” (increased net interparticle distance) or release from the BADGE on the MNPs to more favorably bind to free BPA. This causes a decrease in the solution SERS signal due to MGITC being displaced further from the nanoparticle surface interface quantifiably as hypothesized and as demonstrated in Fig. 5(c).

Fig. 5

Schematic of nanocluster dissociation in the presence of the analyte BPA. (a) SERS spectra of assay nanoclusters taken at 1 min intervals for 6 min after exposure to 10 nM BPA. (b) Nanocluster assay response after exposure to 0 to 100 nM of the competing analyte BPA in free solution as a function of time, where the $r^2$ coefficient of the fits for the time response curves against 1, 10, and 100 nM BPA were 0.96, 0.95, 0.97, respectively. (c) Average SERS peak intensity at $1175{\mathrm{cm}}^{-1}$ after 5 min of reaction time with BPA as a function of concentration.

3.3.

SERS Analysis of Prebound Assay Particles Housed in Magnetofluidic

To develop a more repeatable SERS analysis platform for assays relying on magnetic nanoparticles, PDMS microchannels laid over two Ni-Patterned glass slides that can be magnetically activated by two neodymium magnetic encased around the optofluidic chip were developed (Fig. 6). Specifically, the magnetic collection rate and the SERS enhancement of silver-coated ${\mathrm{Fe}}_2{\mathrm{O}}_3$ and ${\mathrm{Fe}}_2{\mathrm{CoO}}_4$ nanoparticles bound to the Ag target particles were compared for two different Ni-pattern designs as shown in Fig. 6: straight (chip design A—center images) and spotted (chip design B—right images).

Fig. 6

Magnetofluidic chip designs: (a) Top-down and isometric images of the experimental setup for Raman mapping of the assay nanoparticle clusters within the magnetic microfluidic. (b) Schematic and brightfield images of the Ni-patterned detection regions through a $10\times$ objective for chip design A. (c) Same as center for chip design B.

Using 3-D Raman mapping, the straight channel (chip A) provided a more uniform nanoparticle organization with an apparent coefficient of variation of 23.6% across the entire channel when in focus with the Ni pads. However, only the assay with the less stable Co-doped $\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{CoO}}_4$ probe was magnetically trapped within chip design A. This was likely due to its rapid magnetic collection rate (on the order of minutes compared to hours required to pull down 2 mL in a glass scintillation vial) when compared to the “plain” but stable ferric particles. To investigate, a depth profile of the straight channel after exposure to $200\mu \mathrm{L}$ of Co-doped ferric assay nanoparticle clusters was obtained in the form of 3-D Raman spectroscopic maps (Fig. 7).

Fig. 7

Chip design A 3-D intensity profile: 17-stacked $XY$ Raman intensity maps of the straight channel design filled with Ag@FeCo assay clusters, revealing that the majority of particles are located at the bottom of the channel near the surface of the nickel pad.

The map in Fig. 7 reveals that the large majority of particles are located at the bottom of the channel, near the surface of the deposited nickel. Unfortunately, exposure to BPA competing analyte as high as $2\mathrm{mg}/\mathrm{mL}$ (an order of magnitude above the physiological range) revealed no statistically relevant change in the SERS intensity in the channel. It is hypothesized that this is due to the magnetic field delivered to the pads being too strong at its surface and not reaching the full height of the channel with uniformity. This forces the particles too close together, hindering competitive binding or even causing irreversible mechanical aggregation. This is consistent with the COMSOL models in Fig. 2, which predict that the magnetic field is much stronger for a planar field of Ni when compared to an array pattern. Additional error in the particle depth distribution could be due to the fact that only the polydisperse, unstable $\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{CoO}}_4$ probe assay was trapped within chip design A. While this design was simpler to pattern, and the ${\mathrm{Fe}}_2{\mathrm{CoO}}_4$ core being the much simpler choice in terms of synthesis, it was determined that this “quick and easy” modality is not sufficient for monitoring molecular binding events with SERS. However, it may still be valuable as a simple solution for controlling magnetic nano (or micro) particles for assay wash steps. This design could be used to either automate washing steps and buffer exchanges or automate the capture and enhancement of the spectra of specific components in a complex solution.

Chip design B, consisting of a $2\times 5$ Ni-pad array, was able to trap the more stable $\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{O}}_3$-based assay. Chip B demonstrated a 10-fold improvement in the localized SERS enhancement (Fig. 8) across the nickel pads when compared to chip A and, thus, theoretically allows for lower limits of detection. Since the COMSOL models in Fig. 2 revealed that the magnetic field would be weaker but more evenly distributed for the array pattern, we suspect that the increase in SERS enhancement is due to an increase in the number of assay particles trapped not an increase in the magnetic field strength. In terms of quantification limits, looking at the raw peak intensity at $1175{\mathrm{cm}}^{-1}$ the coefficient of variation (%CV) across all 10 pads was at nearly 55%, though intensity maps in Fig. 8 reveal this is mainly due to a larger portion of particles located on the top row of pads. This is likely due to several incomplete nickel pads in the bottom row, which prevent the magnetic field from propagating uniformly across the pads. Unfortunately, depth profiles for design B were unable to be obtained as the small pad’s field was still too concentrated near the surface, yielding-enhanced assay signal only when in focus with the nickel pad as before.

Fig. 8

Chip design B Raman $XY$ intensity profile: looking at the center column from the $2\times 5$ array of Ni pads it is noticeable that particles favor the top row of pads, which are more uniform than the bottom row and thus provide a more uniform field.

To overcome the issues with chip designs A and B, a more optimized chip C nickel pad pattern was designed. Chip design C involves a $1\times 5$ array of magnetically activated pads housed within the PDMS channel (Fig. 9, left), which provides similar benefits as Chip B while avoiding the error induced by having two full rows of pads within the channel. This design allows the field to propagate in a manner that provides individual pad fields strong enough to trap the weakly magnetic yet stable $\mathrm{Ag}@{\mathrm{Fe}}_2{\mathrm{O}}_3$-based assay while also providing uniform SERS enhancement in the $Z$-dimension. In other words, it is desired that the colloid is suspended in solution in the channel and that particles are not irreversibly aggregated by the field pulling them too close together or onto the nickel surface so that they may avoid steric hindrance when exposed to BPA.

Fig. 9

Chip design C Raman $XY$ intensity profile and response to BPA: looking at the entire length of the $1\times 5$ array of Ni pads. It is noticeable that SERS intensity of the particles is redistributed when buffer is introduced, indicating particle resuspension. Upon introduction of $2\mathrm{mg}/\mathrm{mL}$ of BPA, the net SERS intensity over each pad decreased in response to competitive binding.

Top-down Raman intensity maps of chip loaded with the ${\mathrm{Fe}}_2{\mathrm{O}}_3$-based assay design C are also shown in Fig. 9. When the channel is filled with dried assay particles, the chip demonstrated SERS enhancements up to 5 times higher than design B and 30 times higher than design A, as demonstrated by comparing their averaged full spectra over the nickel patterns in Fig. 10. However, for a more clear comparison across all three conditions, the map of the dried assay in Fig. 9 is scaled to the maximum intensity of the PBS chip, which was 1.5 times higher than chip B and 9 times higher than chip A. The PBS filled channel, where the map was collected $10\mu \mathrm{m}$ above the pad surface, provides ample space for particle suspension and verifies that after filling the dried assay in the channel with PBS the particles are redispersed. This is further confirmed by an overall drop in the SERS intensity, which remained steady over 1 h after the full $300\mu \mathrm{L}$ of particles passed through the chip. Additionally, upon introduction of $2\mathrm{mg}/\mathrm{mL}$ of BPA, a drastic drop in the SERS intensity at $1175{\mathrm{cm}}^{-1}$ was observed after 5 min, consistent with the behavior seen when testing the assay in the well plate and thus validating the chips capability for housing a molecularly mediated SERS assay.

Fig. 10

SERS spectra for the average across the entire Ni patterned area for all three chip designs. Here, it is observed that chip C provides a 30-fold enhancement compared to chip A, and a 3-fold enhancement compared to chip B.

To confirm that particles were not permanently bound to the nickel pads, depth profiles ($XZ$) of chip design C were collected with and without the syringe pump turned on at $10\mu \mathrm{L}/\min$ (Fig. 11). When the pump is on, the particles’ signal moves with the direction of flow. Although the %CV across the entire channel for design C is still $\sim 50\%$ as with design B, while investigating the signal variability across each individual pad it was discovered that the error could be reduced through individual normalization to the each of the five pads’ maximum intensity at time $t=0$ (i.e., when the channel is filled with PBS). The improved standard error of the mean of the five chips yielded a %CV of $\sim 25\%$, further improving upon the previous designs.

Fig. 11

Chip design C Raman intensity depth profile: looking at the depth profile of the second pad in the array. (a) Map shows that assay clusters are not bound to the surface and are able to move with the pump flow as a colloidal suspension. (b) Map shows assay cluster reorganize over the center of the pad when the pump is turned off.