Integration of polymeric membrane/dielectric sphere assemblies in microfluidic chips for enhanced-contrast imaging with low-magnification systems

Abstract. Current microscopy systems for the imaging of microorganisms are expensive because of their optimized design toward resolution maximization and aberration correction. In situations where such an optimization is not needed, for instance to merely detect the presence of pathogens in liquids for on-site analyses, a potential approach is to use highly refractive spheres in combination with low-magnification objectives to increase the resolution and the sensitivity of the optical sensing system in a cost-effective fashion. Indeed, for point-of-need assays, integration of optical elements on a microfluidic device can bring several advantages, such as test parallelization/automation and low-volume consumption. We report a study on BaTiO3 spheres that are partially embedded in thin polymeric membranes of mismatched refractive index. We computed the transformation that the polymeric membrane/dielectric sphere assembly (PMDSA) mediates on the light originating from the sample toward the optical detector and shows its enhanced-detection potential for a low-magnification objective. We then propose a method to easily fabricate chips with custom designs and precise location of such dielectric spheres relative to the microfluidic structures for enhanced imaging of microorganisms. We applied this concept to the detection of living fluorescent bacteria, either flowing in aqueous medium or immobilized in hydrodynamic traps. We quantified the contrast gain provided by the PMDSA for short exposure when used with a low-magnification objective. By comparing with a high-magnification objective, we also show how longer-term imaging can be still reliably performed with a more cost-effective system. Since the present PMDSA concept combines the optical enhancement of low-magnification systems with the flexibility of microfluidic handling, it can be highly suitable for portable and cost-effective systems for on-site analysis, from flow cytometry to longer-term antibiotic testing.

extremely hard. Moreover, if the μSs are not well positioned relative to the sample, imaging and detection may be hindered. 3,16 For applications that target the detection of large objects (e.g., bacteria or cells) instead of small objects (e.g., biomolecule, nanoparticles or subcellular structures), a large depth-of-field is usually required. 18,19 Additionally, for applications that require long-term imaging of living microorganisms, sample handling and immobilization often constitute a significant challenge. Several microfluidic approaches have been proposed to facilitate such analyses, e.g., single-cell arraying 20 and single bacteria immobilization. 21 Integrating microoptical elements with microfluidic platforms can therefore substantially enhance the detection of microorganisms, by adding potential for portability, cost affordability, and eventually automation. 22 Optical waveguides and microfluidic channels have been embedded in polymer-based devices for fluorescence excitation of biological samples, such as fluorescently labeled phospholipids and stained cells. 23 Another work reported a device based on a polydimethylsiloxane (PDMS) 2D optical lens integrated with microfluidic channels to improve the performance of fluorescence spectroscopy performed using optical fibers. 24 Moreover, an optofluidic system based on a low-refractive index liquid with tunable biconcave PDMS lenses for focusing light has also been reported. 25 Existing microfabrication techniques allow integration of optic and fluidic elements in lab-on-a-chip devices. 26 Another work demonstrated a fluid-based microfabricated lens capable of integration into a multilens optical system for microfluidic flow cytometry. 27 Another work presented microfluidic observation channel capable of performing flow cytometry measurements using solid-state lasers and PIN-based photodetectors used for fluorescent microsphere and nucleic acid labeled fungi. 28 However, such approaches mainly employ expensive optics and fabrication methods, which may hinder the affordability one can aim at. In this work, in order to test the potential of empowering low-magnification systems for the analysis of microorganisms in a cost-effective fashion, we integrated BaTiO 3 spheres in microstructured polymeric membranes toward the creation of microfluidic chips. With respect to previous works, our device constitutes a low-cost solution to potentially enhance the detection capability of low-end microscopy systems. The possibility of readily coupling polymeric membrane/dielectric sphere assembly (PMDSA) with an optical microscope makes the device possibly suitable for applications to on-site detection. Also, our method reduces manual work required for sample preparation steps by integrating the optical elements in a microfluidic tool, operated by programmable pumps. Moreover, embedding of dielectric spheres in the PDMS membrane is not limited by the design of the microchannels, which can be adapted according to the specific application.

Analytical Modeling
The procedure to calculate the optical path through the sphere has been described previously. 3 Briefly, to quantitatively calculate the effect of the sphere on the light cone emerging from the sample, we used the property of cylindrical symmetry to reduce the calculation to a 1D problem and used linear wave propagation and Snell's law to compute the relation between the input and output angles relatively to the optical axis of the imaging system. By using the same notation as previously, 3 the theoretical contrast gain was calculated as 1−cos½θ S ðhÞ 1−cosðθ obj Þ , where θ obj is the maximum angle collected by the objective relatively to the optical axis, and θ S ðhÞ is the maximum collected angle in presence of a 1-mm-diameter sphere at a distance h from the sample along the optical axis.

Chip Fabrication
Microfluidic devices were prepared by soft lithography. Briefly, after surface activation (20 min in hexamethyldisilazane vapor) of 4-in. Si-wafers, we spin-coated 2 μm of positive photoresist (AZ9260). We then exposed in hard-contact mode for 10 s (100 mJ∕cm 2 ) and developed the photoresist. After back-side rinsing, dry etching was achieved with the standard BOSCH process at 25°C, with process time adjusted to obtain the desired height of the structure (etching rate ≈3.5 μm∕ min). PDMS moulding mixture of 10:1 (prepolymer:curing agent) was mixed manually and degassed for 1 h. The mixture obtained was then poured onto a Si-master or spincoated on a Si master depending on the desired structure. BaTiO 3 spheres (Cospheric) were manually positioned on the PDMS film with tweezers. Curing was achieved by keeping at 100°C for 1 h. The solid PDMS obtained was then peeled off and bonded to a PDMS block by standard plasma activation (0.2 mbar O2, 3.0 mA, 0.5 kV) for 60 s. Inlets and outlets were then manually drilled with a puncher, and finally the chip was bonded to a glass slide with the same plasma activation procedure.

Image Acquisition and Processing
The samples imaged consisted of Escherichia coli specimens genetically modified to express red-fluorescent protein (DsRed). Chip and bacteria were imaged with an Axio Imager M2 upright microscope, equipped with a SOLA light engine (SOLA SE 5-LCR-VB, Lumencor). A filter cube (Filter Set 43 HE, Zeiss) was placed in the optical path for fluorescence imaging. We used an air objective 10 × ∕0.25 NA (Zeiss). Images were recorded with a sCMOS camera (pco.edge 4.2, Visitron Systems). Images were recorded in 2 × 2 binning mode, with variable exposure depending on the condition of the specific experiment. In order to automate the acquisition, the microfluidic chip was fixed on a motorized xyz-stage (Visitron) of an upright microscope and connected to a neMESYS syringe pump (Cetoni GmbH). For the time series of fluorescence images recorded with flowing bacteria, the following processing was performed: the pixel-by-pixel median of the series was subtracted to each image to obtain background-free images; a Gaussian blur with σ G ¼ 2 was applied to each image (this step was performed only for images recorded in the presence of the sphere, where the ≈5.5× magnification increases the variability between adjacent pixels, which may bias the detection of a single bacterium by splitting it into several smaller particles); the standard deviation of the pixels of the entire series was calculated and a threshold was set to 10 times this value; each image was thresholded and the area and integrated intensity of each particle detected were calculated. For the time series of fluorescence images recorded with trapped bacteria, all images of the same sequence were aligned by a custom developed script based on template matching, 29 in order to remove artifacts due to eventual movement along x − y plane. Subsequently, two different regions of interest of the same area were drawn to define the immobilized bacterium and an adjacent region without for background evaluation. Finally, average grayscale values were measured in all the slices of the time series for both bacterium and background.

Data Analysis
Only the particles with an area higher than 20 pixels were considered, to prevent a potential bias from small particle artifacts. The values of the integrated intensity in the presence of the sphere were divided by 2πσ G to correct for the Gaussian blur 30 introduced upstream to ease the identification of the bacteria. All the values were then normalized to the highest one to ease comparison for different conditions. To compute the experimental CBR gain, we divided the values corresponding to the same exposure.

Embedding Highly Refractive Spheres in Microstructured Polymeric
Membranes for Microfluidic Chips

Model
Based on our previous resolution analysis on dielectric spheres immersed in homogenous media, 3 we deduced the optimal configuration for a highly refractive sphere (barium titanate, refractive index n S ¼ 1.95) embedded in a PDMS membrane (PDMS, refractive index n PDMS ¼ 1.43) [ Fig. 1(a)]: it consists of a sphere only partially embedded in PDMS, where the index mismatch between the sphere material, the polymer, and the surrounding media increases the overall collection angle of the optical system [θ out in Fig. 1(b)]. This effect can be used to generate resolution and contrast gains for low-magnification objectives: in Fig. 1(b), we show the presence of an enhanced-detection zone (EDZ) for a 10 × ∕0.25 NA objective, which approaches the collection angle of a 100 × ∕0.75 NA objective with the same immersion medium (air).

Fabrication of the PMDSA
To simplify the fabrication of PMDSAs, we developed a procedure (Fig. 2) to create microstructured PDMS membranes with both customizable designs and embedded spheres to increase resolution and sensitivity. Initially, we patterned a 5-in. chromium-and-glass mask following the microchannels design. To fabricate the Si mold used for casting the PDMS sphere-embedded membrane, conventional photolithography was used to pattern a 1.5-μm-thick layer of AZ9260 positive photoresist on a 4-in. 550-μm-thick Si wafer. Subsequently, we dry-etched 6 μm of the Si substrate with a BOSCH process and stripped the photoresist protective layer. We performed a silane treatment of the Si surface in order to prevent sticking of PDMS on the substrate during membrane peeling. To finally form the PMDSA, a 10:1 mix of PDMS prepolymer and curing agent was spin-coated on top of a microstructured Si wafer, to form a PDMS layer of ≈300 μm thickness; then the spheres were manually positioned at the desired locations, and finally the PDMS was cured and peeled-off the Si mold. Pillars and hydrodynamic traps, located underneath the dielectric sphere, are spread over the microfluidic channels in dense arrays, so that positioning of the sphere does not require particular accuracy skills and can be easily performed manually by visual inspection with a stereomicroscope. The full chip was assembled by bonding the PMDSA to a thick and flat PDMS block or to a glass slide. Embedding of the BaTiO 3 spheres directly in the patterned PDMS membrane highly improves the precision of the location of the spheres relative to the microfluidic structures compared to performing bonding between a patterned PDMS block and a membrane with embedded dielectric spheres: in the latter case, alignment is very sensitive because it must be done quickly after plasma activation of the PDMS surfaces and the bonding is irreversible; in the former procedure, spheres can be positioned very precisely before the curing step, which allows more time and reversible positioning.

Microfluidic chip design
We fabricated a PMDSA with a design to analyze bacteria (1 to 2 μm size) in aqueous media [ Fig. 3(a)]. The device consisted of a 1-mm-large, 6-μm-high channel with equally spaced cylindrical pillars [ Fig. 3(b)]. The large and narrow channel keeps the hydraulic resistance low while keeping the bacteria close to the imaging region of the sphere. The pillars, 100 μm in diameter, prevent the microchannel ceiling from collapsing [ Fig. 3(c-i)]. The central region of the channel includes a large array of trapping structures intended for bacteria immobilization and long-term imaging [ Fig. 3(c-ii)]. Trapping units have a sieve-like design and present a tapered opening of 1.2 μm narrowing to 0.8 μm, for tight trapping of E. coli specimens. This type of confinement method allows to immobilize a small number of bacteria (down to a single bacterium) in each trapping unit: once a trap is filled, the flow passing through the sieve is stopped, preventing additional bacteria from being trapped. 21 The packed array of traps built in the microchannel has also the functions to maximize the number of detected bacteria in the field of view of the dielectric sphere and to provide structural stability to the chip.

Contrast enhancement of a low-magnification system
Since such chips can be used in a portable fashion and have the potential to be combined with simple optical detection systems, they could be used to detect pathogens in water basins in remote sites. To test the feasibility of this application, after an initial cleaning of the device by flowing a 70% ethanol solution, we injected red-fluorescent bacteria (≈2000 bacteria∕μL) in the chip and recorded image sequences using a low-magnification objective (10 × ∕0.25 NA). The detailed fluidic protocol utilized for this application is reported in Figs. 3(d-i). To assess the potential for fast image recording, we performed the analysis for short exposure (1 to 20 ms). Moreover, since we do not require high resolution for such imaging, we used 2 × 2 binning to further decrease the acquisition time. After subtraction of the pixel-by-pixel median of the time series to each frame, we obtained background-free images [ Fig. 4(a)]. The PMDSA introduces a ≈5.5× magnification but also some aberrations which distort the expected oval shape of bacteria. By color-coding the different time frames, one can obtain the trajectories of the bacteria in the channel [ Fig. 4(a-ii)]. The contrast-to-background ratio (CBR) was quantified as the ratio of the average signal over the average of the background, and finally normalized to its maximum experimental realization. In Fig. 5(a), we report the normalized CBR measured for several exposures. As expected, for short exposure time, it increases with the exposure for both the conditions

Comparison with a high-magnification system
In order to assess the performances of the PMDSA coupled to a low-magnification objective (10 × ∕0.25 NA) compared to a high-magnification system (100 × ∕0.75 NA), we trapped fluorescent bacteria in a microfluidic channel and imaged them with both systems. For this purpose, we utilized an array of hydrodynamic traps designed to confine the bacterial cells during the image acquisition [ Fig. 3(c)]. The trapping section is positioned at the centre of each microchannel underneath the dielectric sphere. The fluidic protocol used for the experiment is shown in detail in Fig. 3(d-ii). After a cleaning step, a suspension of fluorescent bacteria was injected into the channels. Afterward, buffer medium was injected to only retain the stably trapped bacteria. Finally, we set a constant injection flow rate at 0.12 μL∕ min in order to avoid the escape of confined bacteria from immobilization sites during time-lapse imaging, performed over 80 min [ Fig. 4(b)]. The fluorescence signal was quantified by measuring the average intensity value inside the region of interest delimited by the elliptic shape of the bacterium. The background value was quantified as the average intensity value inside a region of interest with the same area as the one used for the bacterium but positioned away from the trap. Comparison between normalized CBR measured for the PMDSA coupled with a low-magnification objective (10 × ∕0.25 NA) with respect a high-magnification system (100 × ∕0.75 NA) is illustrated in Fig. 5(c). The CBR resulted to be stable over time for both the systems, which show that the structure composed by the thin membrane with embedded sphere does not introduce mechanical instabilities which could result in drift or oscillations of the focal plane. Conversely, the CBR of the high-magnification objective remains 3.5 larger than the one obtained with the polymeric membrane with embedded BaTiO 3 sphere, which is not surprising when considering the huge difference in terms of optimization and corresponding cost [ Fig. 5(d)].

Conclusion
We have described a method to enhance the contrast in low-magnification optical detection based on highly refractive BaTiO 3 spheres partially embedded in polymeric membranes, which we integrated on microfluidic chips for the imaging of living fluorescent bacteria. The present simple fabrication procedure enables to create custom designs for the chips and, at the same time, to precisely locate the spheres relative to the microfluidic structures. The ability of the PMDSA to enhance the detection of bacteria in a continuous flow already for short exposure [ Fig. 5(a), dark gray area] paves the way for applications where flow cytometry needs to be performed in remote sites, where bulky equipment should be suitably replaced by portable optical and microfluidic systems. Moreover, we demonstrated the opportunity to integrate specific functional microfluidic elements (e.g., trapping structures). Such customizations can guarantee reliable long-term imaging capabilities [ Fig. 5(c)] and thus potentially enable applications such as antibiotic testing directly on site.
By comparing the experimental results with the theoretical calculation for the PMDSA [ Fig. 5(b)], we observe that the system is suboptimal (≈30% lower than the expected theoretical gain). This loss may be due to several factors that can cause a disruption in the light path from the sample to the objective: • the inhomogeneities in the sphere size and material; • the aberrations introduced by the sphere [Fig. 4(b-ii), arrows]; • the reduced depth-of-field in presence of the sphere λ em n NA 2 ≈ 584 nm·1 0.66 2 ¼ 1.3 μm (which is comparable to the size of a bacterium and thus can exclude some light coming from it from being collected by the objective) compared to the conventional condition λ em n NA 2 ≈ 584 nm·1 0.25 2 ¼ 9.3 μm. Moreover, despite the focusing effect from the sphere (which increases the contrast gain), the excitation through the sphere may also be suboptimal because of: • the inhomogeneities in the sphere size and material; • internal reflexion at its bottom side for a part of the excitation light.
Given the nature of the PMDSA proposed here, we expect spherical aberration to be present because of the presence of the BaTiO 3 sphere, and we cannot exclude the presence of other aberrations. Astigmatism or coma for instance can be introduced by imperfections in the shape of the dielectric sphere. All the aberrations introduced by the addition of the PMDSA to the optical system degrade the image of the object, which can be certainly detected with higher confidence, but at the price of lowering resolution on its details. This system is therefore more adapted for optical signal detection (because of increased contrast) rather than for complete imaging (which usually requires higher resolution). To obtain quantitative information about the aberrations introduced by the PMDSA, one may study the aberrated point-spread function in terms of its decomposition with the Zernike polynomials, which analytically describe the different types of aberrations. 31 This may be accomplished by imaging fluorescent nanoparticles, which can be safely approximated as point sources, which cannot be done for the bacteria in our study. Models are also used to analyze and correct distortion, 32,33 which can be needed especially when parts of the object are imaged far from the center of the sphere, which has a total field of view of about 0.03 mm 2 . Once the aberrations are more quantitatively described, one may use hardware and software solutions to obtain higher-resolution images. From a hardware point of view, one can design the shape of the polymeric membrane (which is flat in the imaging zone for simplicity of fabrication in our study) in order to compensate for some aberrations. This method is often used in commercial objectives, where several lenses altogether provide much less aberrations than the individual ones. From a software point of view, one can apply deconvolution algorithms to the aberrated images. Such methods can highly increase the net resolution of the PMDSA system.
The problem of fabrication costs is well known in microfluidics, a domain which still enormously relies on clean-room microfabrication techniques, which are highly expensive. In fact, everything intrinsically depends on the type of structure one wants to create in the chip. Assuming that conventional photolithography techniques are used for Si wafer processing, the cost of fabrication of a mold is of the order of 100 $ (material + expenses on clean-room equipment only). A Si master can be used multiple times for PMDSAs casting, such method of production allowing to reduce the cost of a single PMDSA unit. Fabrication expenses are mainly related to material costs of PDMS together with dielectric spheres (0.5 $/PMDSA) and of utilization of PDMS processing equipment (30 $/h). However, other methods besides clean-room microfabrication are applicable to obtain reliable chips with lower costs in terms of infrastructure and material. Injection-molding, for instance, would be ideal for large scale production of chips with the same design. Given the low cost of such dielectric spheres (0.04 $/sphere) and the current continuous decrease of the fabrication methods for microfluidic chips (such as hotembossing and injection-moulding 34 ), we believe that the proposed fabrication protocol is readily adaptable to a wide variety of microfluidic chips with integrated optical elements. After initial investment in the creation of the mold (which may rise up to tens of k$), the cost of a single chip may be lowered to 1 $ or less, depending on the fabrication material used. Moreover, with a suitable mold, one may eventually improve the positioning of the dielectric spheres on the chip and decrease the time spent for this operation as well. When considered together with the potential cost reduction associated to low-end imaging systems, the present approach has the potential to highly increase the usability in low-resource settings for the often required on site analyses.