Miniaturized modular-array fluorescence microscopy

: Fluorescence live-cell imaging allows for continuous interrogation of cellular behaviors, and the recent development of portable live-cell imaging platforms has rapidly transformed conventional schemes with high adaptability, cost-effective functionalities and easy accessibility to cell-based assays. However, broader applications remain restrictive due to compatibility with conventional cell culture workflow and biochemical sensors, accessibility to up-right physiological imaging, or parallelization of data acquisition. Here, we introduce miniaturized modular-array fluorescence microscopy (MAM) for compact live-cell imaging in flexible formats. We advance the current miniscopy technology to devise an up-right modular architecture, each combining a gradient-index (GRIN) objective and individually-addressed illumination and acquisition components. Parallelization of an array of such modular devices allows for multi-site data acquisition in situ using conventional off-the-shelf cell chambers. Compared with existing methods, the device offers a high fluorescence sensitivity and efficiency, exquisite spatiotemporal resolution ( ∼ 3 µm and up to 60 Hz), a configuration compatible with conventional cell culture assays and physiological imaging, and an effective parallelization of data acquisition. The system has been demonstrated using various calibration and biological samples and experimental conditions, representing a promising solution to time-lapse in situ single-cell imaging and analysis.


Introduction
Visualizing diverse anatomical and functional behaviors of single cells provides critical insights into the fundamentals of living organisms. Conventional assays rely on standard cell fixation to observe and characterize cells at discrete time points, insufficient to reveal many dynamic, rare and heterogeneous cellular events. In contrast, the rapid development of time-lapse imaging techniques permits uninterrupted and noninvasive interrogation of cells with exquisite spatiotemporal contextual details [1][2][3]. These techniques have greatly enabled recent studies on various dynamic cellular mechanisms and behaviors such as cell proliferation [4], morphology evolution [5], cell-cell interaction [6], and cell interaction with microenvironment [7].
To maintain optimum culture conditions for continuous and long-term observation, commonlyused live-cell imaging instruments equip conventional microscopy with environmental control and incubation systems [8,9]. However, these modalities regulate culture conditions less accurately compared to standard cell incubators, causing potential temperature inhomogeneity and the leakage of gas and humidity compositions [1,8]. As a result, the methods can induce cell stress due to environmental alterations from native physiological conditions, detrimental to many sensitive cell types and impeding long-term observation [10]. Furthermore, many of these high-end instruments lack the flexibility to adapt to diverse project-specific requirements and become impractical for those chip-based biological systems incompatible with conventional microscopy [11][12][13].
Addressing these challenges, the emergence of portable cell imaging strategies has thus far transformed many conventional schemes with high adaptability, cost-effective functionalities and easy accessibility to cell-based assays [14]. Integrating advanced fabrication, computation and sensor technologies, these platforms have been reported in various implementations such as lens-free techniques, smartphone imaging, and miniaturized microscopy [14]. Amongst these advancements, for example, lens-free and smartphone imaging methods bypass the restrictions of traditional optical hardware, thereby effectively overcoming either the fundamental trade-offs or the instrumentation complexity of lens-based imaging systems. These breakthroughs have exploited such versatility of adaptation and integration for cell diagnosis and analysis [15][16][17][18][19][20].
On the contrary to the lens-free or smartphone techniques, miniaturized microscopy retains commonly-adopted imaging features and sample assays of conventional microscopy, while particularly optimized for time-lapse cell observation in situ [21,22]. Specifically, various fluorescence implementations have been exploited for miniaturized microscopy to investigate cells with high sensitivity and molecular specificity [23][24][25][26][27][28][29]. The current architectures require specialized cell chambers, inverted microscope configurations, and mechanical scanning across a larger field of view. However, these features restrict experimental demands such as compatibility with conventional cell culture workflow, accessibility to up-right physiological imaging, integration with biochemical sensors under the cell platform, and parallelization of data acquisition [30,31]. Furthermore, many of these fluorescence components are additionally implemented to the existing bright-field miniaturized systems, adopting a broad-beam illumination or compound objective lenses, thus less optimum in image quality, fluorescence efficiency and phototoxicity for time-lapse observation. In this context, the fluorescence imaging capability of in situ microscopy remains to be fully explored to meet ever-increasing live-cell imaging needs.
Here, we introduce miniaturized modular-array microscopy (MAM) for compact portable fluorescence live-cell imaging in flexible formats. The MAM system is formulated on the basis of the emerging miniscopy technology for functional brain imaging, which has offered remarkable advantages in the fluorescence imaging capability, high flexibility and scalability, and open accessibility to mass-fabricated micro-optics and semiconductor optoelectronics [32,33]. In this work, we advanced the miniscopy architecture and designed compact up-right modular microscopes, implemented with gradient-index (GRIN) objectives and individually-addressed illumination and digital components. The system provides high fluorescence efficiency and low photo-toxicity and enables parallel data acquisition in situ using conventional off-the-shelf cell chambers. The modular device offers a high optical sensitivity and spatiotemporal resolution (∼3 µm laterally and up to 60 Hz), a configuration compatible with conventional cell culture assays and physiological imaging, and an enhanced imaging content through parallelization of data acquisition. The system has been demonstrated using various calibration and biological samples and experimental conditions, representing a promising solution to time-lapse in situ single-cell imaging and analysis.

Cell staining
HEK-293 T and COS-7 cells were permeabilized by adding 4% Paraformaldehyde fixative solution for 12 min after washing with PBS several times. The fixative solution was prepared by diluting 3 mL of 16% aqueous paraformaldehyde solutions with 6 mL of PBS and 3 mL of Milli-Q water. Fixed cells were carefully washed with PBS and stained using fluorescent stock solutions prepared according to proper protocols provided by the distributor (Invitrogen). Nucleus were stained with 0.5 mL of 1× HCS NuclearMask Deep Red (H10294, Invitrogen) for 1 hour at room temperature. Mitochondria were co-stained by adding 0.05 µL from 500 nM working solutions of MitoTracker Deep Red FM (M22426 Invitrogen) and 0.4 µL from 200nM working solutions of MitoTracker Green FM (M7514, Invitrogen) probes. The 1 mL of complete DMEM were added with Mitochondria staining probes for 45 min at room temperature. For staining cytosol, we applied 100 nM of 1000× stock solutions of Image-iT DEAD Green viability (I10291, Invitrogen) on fixed cells for 30 min at room temperature. After proper application of fluorophore staining, we washed twice with HBSS to refrain from overloading. 2 mL of PBS were added to refrain from drying cells. For live-cell staining, the HCS NuclearMask Deep Red solutions were applied for 30 min inside the incubator and washed with PBS. Next, 2 mL of FluoroBrite DMEM (A18967-01, Gibco) were filled in multiwell plates for time-lapse imaging inside the incubator. All stained cells were prepared right before visualization.

Image alignment with inverted microscopy
We compared the image quality of the MAM system with that of an inverted microscope equipped with 10×, 0.3NA objective lens by imaging fixed COS-7 cells stained with HCS NuclearMask Deep Red using both systems. Given the different magnification and pixel size, i.e. 6.5 µm for the sCMOS camera (ORCA-Flash 4.0 V3, Hamamatsu) used in the inverted microscope and 6 µm for the CMOS sensor used in the MAM system, as well as the actual misalignment between the optical axis of the two setups, we realigned the images. First, we performed a preliminary coarse alignment by rotating the image obtained with an inverted microscope. To improve the effectiveness of image registration, we removed hot pixels and fixed-pattern noise in the MAM image through a feature-based approach [34], which could hinder feature matching. Hot pixels were substituted with the median value between neighboring pixels. Fixed-pattern stripes were removed by filtering the corresponding peak in the Fourier space.

Cytotoxicity assay
We prepared 1 mM of Paclitaxel (Taxol, P3456, Invitrogen) stock according to the protocol provided by the distributor (Invitrogen). For cytotoxicity test, we administered 400 nM of Taxol stock on live COS-7 cells for 8 hours inside the incubator. Before the end of Taxol application, we added HCS NuclearMask for 1 hour and Image-iT DEAD Green viability stain for 30 minutes inside the incubator. Next, we washed the cells with HBS several times before fixation steps. The fixation steps follow the above description in cell staining. Stained cells were filled with PBS for visualization. All cells for cytotoxicity assay were prepared right before visualization.

Image post-processing
We processed the raw images of fixed cell data by applying the Subtract Background built-in function in ImageJ with Rolling ball radius of 50 pixels with no other options applied. For live-cell data, we processed the time-lapse sequence of HCS NuclearMask Deep Red stained nucleus of COS-7 cells to remove the noise introduced by the CMOS sensor. First, we evaluated the pixel-dependent response of the chip. Then, we used the ACsN algorithm for the joint correction of CMOS-related noise in the acquired frames [35]. Briefly, the algorithm uses the sensor calibration to remove fixed-pattern noise and the optical parameters to estimate the noise variance in the Fourier space. Finally, all the input images were processed together by collaborative sparse filtering, which used the data correlation in both space and time to correct the CMOS noise while preserving the image details of the signals.

Design and construction of the MAM system
The MAM system is illustrated in Fig. 1(a) and Appendix 1 (Fig. 5), consisting of an imaging device compatible with various in situ conditions and associated with a control algorithm for system operation and data acquisition. The imaging device contains three main components: an array of modular microscopes, a mounting cover, and an illumination panel ( Fig. 1(b,c)).
In particular, each miniaturized microscope was constructed using an infinitely-corrected GRIN lens (0.25 pitch, 0.5NA, GT-IFRL-200-inf-50-NC, GRINTECH), a customized dualband emission filter (5.0 × 6.0 × 2.0 mm, ET520/30 & ET705/70, Chroma), an achromatic tube lens Edmund), and a CMOS sensor (MT9V032C12STM, ON Semiconductor) [36]. Individual modular microscopes were threaded onto a 3D-printed mounting cover that environmentally encloses the cell culture chamber (e.g. a 12-well plate was utilized in this work as shown in Fig. 1(b)) and positions the microscopes in a user-defined array pattern according to the specimens of interest. The illumination panel consists of a corresponding array of multi-color LEDs (NeoPixel5050, ID1612, Adafruit) with peak spectra at 470 nm, 520 nm and 625 nm, half-ball lenses (90859, Edmund), and customized dual-band excitation filters (12.0 × 12.0 × 1.0 mm, ET-FITC/Cy5x, Chroma). The illumination panel was placed on a mounting plate (MSB15_M, Thorlabs). The overall device was designed with computer-aided software (Solidworks) and fabricated using a 3D printer (RS-F2-GPBK-04, FormLabs2) ( Fig. 5(b-d)). The control system utilized a lab-written GUI (Visual Studio C++, LabVIEW, and Arduino) to operate the illumination panel and data acquisition. Flexible coaxial cables were used to supply power, operate hardware, and transmit image data with each CMOS sensor through their individual data acquisition boards (DAQ) (Fig. 1(d)). General hardware and CMOS operating software are openly available on miniscope.org. The MAM design documents and codes are available from the authors and on GitHub (https://github.com/shujialab/MAM). As a proof-of-concept, the fully assembled imaging device consisting of 12 modules is highly compact in 15 cm × 15 cm × 3 cm, permitting hand-held portability and accessibility to various in situ conditions (Fig. 1(e)).
Notably, the design and architecture of the MAM system offer several unique features compared to existing miniaturized microscopy. First, unlike compound objective lenses, GRIN lenses allow for water immersion and thus up-right imaging of cell populations with low cost, light weight (<1 g), small diameters (<1 mm), long relay lengths (>1 cm), and relatively high numerical aperture (NA, >0.45). Next, each modular microscope can be individually addressed, and their array distribution can be customized into various patterns to facilitate parallel acquisition from user-defined areas of interest. The associated mounting cover forms an effective enclosure to the cell culture condition and is adaptable to a wide range of imaging chambers such as cell culture dishes, plates and slides, as well as microfluidic chips. Furthermore, the design of the illumination and detection optics in MAM utilized schemes consistent with the miniscopy techniques to obtain optimum fluorescence efficiency, high image quality, and low phototoxicity [36,37].

System characterization
To characterize the MAM system, we first used the blue LED and imaged a water-immersed negative USAF target (R1DS1N, Thorlabs), attached to a green fluorescent slide (92001, Chroma) ( Fig. 2(a)). A uniform fluorescence emission was observed throughout the full CMOS sensor (752 pixels × 480 pixels, physical pixel size = 6 µm), and the measurement of the calibration structures determined the magnification (∼6×) and the effective pixel size (1 µm) of the system ( Fig. 2(a-c)).
Next, we imaged 2-µm fluorescent beads (FSDG005, Bangs Laboratories) immersed in water on a glass slide and recorded the point-spread function (PSF) of the system at varying depths ( Fig. 2(d-g) and Appendix 2 (Fig. 6)). The PSF images were Gaussian-fitted and exhibited FWHM values of ∼3 µm and ∼30 µm in the lateral and axial dimensions, respectively ( Fig. 2(f,g)).
It should be mentioned that the GRIN lens is not diffraction-limited due to aberrations. Therefore, the use of 2-µm beads allows viable measurements below the resolution of the system. As seen, the axial PSF is substantially extended due to the intrinsic spherical effects of the GRIN lens [38], which, however, effectively enhance the depth of focus (DOF) to cover the thickness of most single-cell specimens. Furthermore, we recorded the PSFs of the MAM system in different spectra using 4-µm fluorescent beads (TetraSpeck T7283, Invitrogen), showing a good agreement between the two-color PSFs in all three dimensions (Fig. 2(h)). Parallel two-color readouts from the entire array of microscopes on a 12-well plate demonstrated consistent alignments between spectral channels and across each modular device ( Fig. 2(i,j) and Visualization 1).

In vitro cell imaging
To demonstrate cell imaging, we first recorded the nuclei of COS-7 cells stained with HCS NuclearMask Deep Red (H10294, Invitrogen), showing consistent MAM image quality as compared with the same region taken by conventional inverted wide-field microscopy using a 10×, 0.3NA objective lens (Fig. 3(a) and Appendix 3 (Fig. 7)). It should be mentioned that in Fig. 3(a), we implemented a modular microscope (without the illumination panel) atop an inverted conventional epifluorescence microscope. This configuration allows for a simultaneous acquisition of the same cell population in a multi-scale fashion combining both the up-right MAM module and inverted conventional imaging (Appendix 3 (Fig. 7)).
To examine two-color imaging, we next imaged mitochondria in HEK-293 T cells, co-stained with MitoTracker in deep red (M22426) and green (M7514, Invitrogen) ( Fig. 3(b)). Two-color MAM modular imaging showed a consistent distribution of cells across the field of view, as well as the quantitative measurement on their cross-sectional profiles in both channels ( Fig. 3(b,c)).
Next, we utilized an array of 12 modular devices and performed MAM imaging of the nucleus-stained COS-7 cells using a 12-well, black-walled, and clear-bottom cell plate (P12-1.5P, Cellvis) (Fig. 3(d)). To facilitate in-focus acquisition throughout the entire array of FOVs, the  focal planes of each miniaturized microscope can be fine-adjusted using the threads on the microscope body that position the camera sensor, which design also offers a high stability after each adjustment for time-lapse observation. As seen, individual cells can be clearly resolved and precisely identified in the MAM images using automatic nucleus segmentation [39] (Fig. 3(e)). We then measured all the segmented nucleus areas on the 12-well plate, showing consistent quantification of the cells across the array with an average nucleus size of 23.5 ± 2.7 µm, in a good agreement with the known size of ∼25.0 µm for COS-7 cells [40] (Fig. 3(f)).
To further assess the usability of MAM to image and measure cytotoxicity, we conducted a two-color assay to determine viability of COS-7 cells. The assay allows for adding the fluorescent reagents, incubating for a time period, and visualizing the results without sample transfer or wash steps in situ. Specifically, the cells were treated with paclitaxel (Taxol) as a chemotherapeutic drug for 8 hours to destabilize the formation of cytoskeletons and induce apoptosis by blocking cell cycle [41,42]. We then labeled the nuclei of all the cells with HCS NuclearMask Deep Red (H10294, Invitrogen) and the dead cells with Image-iT DEAD Green (I10291, Invitrogen), a permeant dye only when the plasma membrane of cells becomes disrupted by Taxol. Using the red and blue LEDs, the MAM system sequentially acquired and registered the optical signals from the two channels ( Fig. 3(g) and Appendix 4 (Fig. 8)). As seen, the two-color fluorescence assay distinguished the dead-cell population from the rest of the active cell population (marked as nucleus − ), exhibiting a high spatial correlation and sensitivity based on fluorescence intensity (Fig. 3(h)). The areas of the nuclei of the two population groups were processed and measured, implying a consistent cell size between dead and active cells after drug treatment (Fig. 3(i)). Furthermore, as expected, the two-color fluorescence scatter plot distinguished cells of the two population groups (dead and nucleus − ) of comparable nucleus fluorescence intensity while different expressions in cytosol (Fig. 3(j)). These measurements demonstrate the high resolution, sensitivity and assay-compatibility of the MAM system.

Live cell imaging inside a conventional incubator
Retaining the physiological function of cells is essential for continuous, time-lapse observation and analysis. Thereby, it often remains as the ultimate limiting step to maintain an appropriate incubation and environmental condition. Here, we performed live-cell imaging inside a conventional incubator, where the compactness and portability of the MAM system permits feasible in situ microscopy (Appendix 5 (Fig. 9)). We imaged the nucleus-stained live COS-7 cells and tracked the cellular movement and proliferation (Fig. 4). The cell population was continuously recorded at 2 Hz, and individual cells were identified and segmented to generate the corresponding trajectories [43] (Fig. 4(a-d)). To facilitate the cell analysis, we have developed a denoising algorithm to enhance the SNR of live-cell imaging [35] (Appendix 6 (Fig. 10)). As seen, the sub-pixel movements of individual cells were localized and continuously tracked over a time period from minutes to about an hour without noticeable focal drifts, which in longer term, can be fine-adjusted using the threads and computationally correlated for uninterrupted frame-by-frame tracking. Furthermore, cell proliferation events were clearly visualized and quantitatively depicted to resolve the continuous formation of daughter cells over time (Fig. 4(e-j) and Visualization 2). It is noted that the high resolution of the MAM system allows for imaging of cell morphology, motility, and proliferation at a highly confluency level of >1000 cells/mm 2 .

Discussion and conclusions
We have developed miniaturized modular-array microscopy (MAM) for fluorescence live-cell imaging in a flexible and portable format. The system takes advantage of the emerging miniscopy technology and presents a new up-right architecture, integrating compact GRIN objectives and illumination and detection components. The modular design allows for feasible parallelization of multi-site data acquisition from conventional well plates using an array of the devices. The system provides a high fluorescence sensitivity and efficiency and resolution for capture of spatiotemporal details. It should be noted that the proof-of-concept could be readily extended to various cell culture chambers by customizing the mounting pattern of the array of the devices. It should also be mentioned that several recent developments of miniscopy have demonstrated significant advances by implementing such as light-field imaging [44], optical design [45], multi-focal phase mask [46], and computational frameworks [47], substantially enhancing the capability for imaging large-scale tissues with miniature systems. These concepts and designs are readily adaptable to improve the MAM system for single-cell studies. Other ongoing efforts strive to downsize each modular device to enhance the parallelization with a higher throughput and content. In addition, the system can be applied to various large-scale specimens beyond single cells by implementing motorized axial scanning. The imaging scheme presents minimal instrumental complexity and full compatibility with various cell culture, conventional microscopy or sensing modalities, and microfluidic configurations, permitting different levels of integration for long-term in situ single-cell imaging and analysis at the chip scale.