System design
The VFC imaging system has three major sub-systems: lightsheet illumination, multichannel microfluidic chip (specimen holder) and high-speed detection.
Lightsheet illumination
Simultaneous illumination of multiple microfluidic channels allow interrogation of a large population of cells. Accordingly, a large light sheet is generated in order to cross-section the entire array of micro-channels. A laser (532 nm Excel Laser, Quantum Lasers, UK) of wavelength 532 nm and beam-width 1.5 mm is used as the light source. The beam is expanded using a beam-expander (consists of two biconvex lens of focal-lengths 25 mm and 125 mm procured from Thorlans, USA ) by a factor 5X-times. This is essential to just over-fill the back-aperture of cylindrical lens ((f=150) mm, Edmund Optics, Singapore) to utilize its full NA. The expanded beam is 1D-focused by the cylindrical lens to form a horizontal light-sheet. An objective lens (Olympus 10X, 0.30 NA) is placed at the focus of cylindrical lens to generate diffraction-limited light sheet. The resultant light sheet served as the illumination PSF of VFC system.
Microfluidic chip based sample flow system
Cells were flown through the Y-shapped microfluidic channel array as shown in Fig. 1 and Fig. S1 (see, Supplementary 1). The microfluidic chip is fixed to a home-built chip holder and placed on a XYZ-translator for precise position with respect to light sheet. This enabled optical sectioning of the entire channel array (an array of 4 channels of size 100 µm2). The channel inlet is linked to sample reservoir containing the cells whereas, the outlet is connected to a flow-pump. The pump is operated in a suction-mode and the operations are controlled by computer based interface software. Both the specimens, fluorescent bead (FluoSpheres F8803, Invitrogen) and HeLa cells (labelled with Mitotracker Orange CMTMRos, ThermoFisher) were flown through the channel-array and imaged in-parallel.
The proposed VFC technique uses oblique illumination coupled with orthogonal detection geometry. While the configuration is similar to that of classical SPIM, the presence of a PDMS-based microfluidic chip introduces optical aberration primarily due to refractive index mismatch as the light pass through air-PDMS-sample interfaces. Specifically, the mismatch between the specimen flow device (PDMS) and the sample buffer (cell medium) introduces aberrations, thereby degrading the image resolution. To minimize the effect, we designed the microfluidic channel so as to avoid undue refraction, especially at the chip edges. Accordingly, the fabricated open channels are first realized and then sealed with a microscope-grade refractive-index matching coverslip. The sealing was carried using an oxygen plasma machine where oxygen plasma treatment of both the glass substrate and PDMS’s surfaces are carried out before contacting them with each other immediately after activation. This produces a strong and irreversible seal. The geometrical issue associated with PDMS microfluidic device can be effectively addressed by constructing the channel from a material (such as Teflon, RI=1.35–1.38) with the same refractive index as the cell flow medium (RI= 1.33–1.38). This could resolve many issues arising out of refractive index mismatch. Future improvements require that the number of material interfaces is reduced within the light-path to avoid aberration occurring due to reflection and refraction. Another critical issue is associated with the preparation and cutting of PDMS microfluidic devices that may introduce irregular interfaces (air-PDMS). An additional step to flatten the surfaces could go a long way. All these factors strongly affect image quality and, ultimately the 3D volume.
It may be noted that the angle between the microfluidic chip and light-sheet illumination is critical. The design geometry requires the specimen flow chip to be placed at (45^{circ }) to the illumination system, and the detection is carried out at an oblique angle of (45^{circ }) to collect the light sheet images. The detection scheme is similar to that used in open-top light-sheet (OTLS) microscopy61,62. Oblique light sheet images are recorded and stored initially in a rectangular data cube. Subsequently, the data is sheared to represent the geometry of the reconstructed volume.
Fast detection
The mitochondria in HeLa cell was labelled and the emission peaks at 576 nm. The fluorescence from specimen (HeLa cells) is collected by the detection objective (Meiji, 20X, (0.4,NA)). Subsequently, the light is filtered by a set of filters (notch filter (ZET532nf purchased from Chroma) to remove the illumination 532 nm light and a long-pass filter (purchased from Thorlabs) to filter-out the background to retain fluorescence. The filtered light is then focused to the camera chip (pixel size ~ 5.5 µm × 5.5 µm) by a tube lens ((f=125) mm). The detector is a superfast CCD camera (GZL-CL-22C5M-C, Point gray, USA) with a maximum frame rate of (2.3K ,{rm frames/s}) and have a quantum efficiency of 0.56. The images were collected and sent to the computer for further processing.
System point spread function
To determine the system PSF, fluorescent beads (size = 1 µm) were flown through the microfluidic channels at varying flow-rates. The beads recorded by the camera are appropriately modelled as point source and characterized.
Beads flowing through the channel are conveniently modelled as 2D Gaussian,
$$begin{aligned} f(y,z) = G_0 ,e^{-[(y-y_0)^2 / 2sigma _y^2 + (z-z_0)^2 / 2sigma _z^2 ]} end{aligned}$$
(1)
where, (G_0) is a constant. Here, (z_0) and ((sigma _x , ,sigma _y)) are respectively the mean and standard deviations.
The diameter of bead is approximately given by its full-width at half-maxima (FWHM) i.e,
$$begin{aligned} D approx FWHM approx 2sqrt{2ln 2} ,sigma end{aligned}$$
(2)
During flow, the beads undergo motion-blur. As a result, the beads appear elongated in the recorded image. Considering a flow-rate of Q through the channel of cross-section A, and a camera exposure of (t_{exp}), the elongation of a point (along the direction of flow) is given by,
$$begin{aligned} Delta z = v,t_{exp} =frac{Q}{A},t_{exp} end{aligned}$$
(3)
where (v=Q/A) is the average velocity of fluid flowing (at a flow-rate, Q) through the channel of cross-section A.
Due to flow, the detection PSF appear elongated along the flow direction (z-axis). So the changes in (sigma _z) in the recorded image can be expressed as,
$$begin{aligned} sigma _z = frac{1}{2sqrt{ 2 ln 2 }} FWHM_{z} = frac{(D_z +Delta z)}{2sqrt{ 2 ln 2 }} = frac{(D_z + v,t_{exp})}{ 2sqrt{ 2 ln 2 } } end{aligned}$$
(4)
where, (D_z) is the diameter in static condition, and (FWHM_z) is the full-width at half-maximum of PSF along z-axis.
The standard deviation for the beads along y- and z- axes are given by,
$$begin{aligned} left{ begin{array}{ll} sigma _y = sigma _{y0} = D_y/2sqrt{2ln 2} \ \ sigma _{z} = left[ sigma _{z0} +frac{v ,t_{exp}}{{2sqrt{2ln 2}} } right] = frac{1}{2sqrt{2ln 2}} ,left[ D_z +frac{Q,t_{exp}}{A} right] end{array}right. end{aligned}$$
(5)
where, (sigma _{y0}) and (sigma _{z0}) are respectively the standard deviation along y and z axis at zero velocity ((v=0)).
So, the beads flowing at a flow-rate Q can be approximated by a bivariate Gaussian PSF given by,
$$begin{aligned} G_Q = G_0 ,exp { left{ -left[ frac{(y-y_0)^2 }{ 2sigma _y^2} + frac{(z-z_0)^2 }{ 2sigma _{z}^{2}} right] right} } end{aligned}$$
(6)
where, (sigma _y) and (sigma _z) are as given by Eq. (5).
Experimentally, the images of bead samples (YZ-plane) are recorded. A bivariate-Gaussian function is fit to calculate the flow-induced shifts (along z-axis) with respect to beads at (v=0). With the calculated variances ((sigma _y) and (sigma _z)), a new 2D Gaussian is generated, which is then used as the system PSF. The process is carried out for all flow-rates. The corresponding system PSF at different flow-rates is the used for reconstruction sectional images (see, details in Supplementary 3). For our case, we have taken fluorescent bead of size 1 µm for which the emission occurs at (lambda _{em} = 575) nm, and recording is carried out for all flow-rates.
Image reconstruction using flow-variant PSF
For VFC, the recorded image ‘g’ of the object (cell) ‘o’ flowing through the channel can be modelled as,
$$begin{aligned} g(y,z) = h(y,z) otimes o(y,z) end{aligned}$$
(7)
where, h is the PSF of the dynamic flow system, and (otimes ) denotes the convolution operator.
In the Fourier domain, the above equation can be expressed as,
$$begin{aligned} G = H times O end{aligned}$$
(8)
where, G, H, O are respectively the Fourier transform of the recorded image ‘g’, point spread function ‘h’ and object function ‘o’.
We seek the object function ‘o’ from the recorded image. The function can be retrieved from the above equation by inversion. Computationally, this is accomplished by inverting the function in Fourier domain followed by inverse Fourier transform i.e,
$$begin{aligned} o(y,z) = F^{-1} { G / H } end{aligned}$$
(9)
The above expression is a well-known inverse problem, with the exception that here ‘h’ is a flow-variant PSF. The entire process is commonly known as deconvolution.
We have used HeLa cells (of size, 15–25 µm), and the mitochondria is labelled using mitotracker orange dye following the process described in sample preparation section. The cells were flown at different flow-rates and the images were recorded at a video-rate of (38 ,mathrm{Hz}). The ergonomic design of our system allowed collection of 4–6 sectional images as the cells pass through the light sheet. The 2D images were deconvolved using flow-variant PSF and images were reconstructed. Subsequently, 2D sectional images were stacked together to reconstruct the cell volume (see, Supplementary 3 and 5).
Microfluidic chip fabrication
Using Clewin 4, channel features are designed on 4-inch silicon disc. Negative mask is printed from Clewin 4 .gbr file. Master mold fabrication is done in clean room facilities at Nanoscience Facility, Indian Institute of Science, Bangalore, India. Subsequently, Y-type microfluidic chips were fabricated using standard protocol. Silicon elastomer and curing reagent are mixed thoroughly in the ratio of 10:1. A net mixture amount of 33gm is dicicated with vacuum pump for 20 min until air bubbles are removed. Decicated mixture is gently poured on the top of master mold and is cured in hot oven at (60,^{circ })C for 3 h. Cured PDMS is peeled off from master mold and useful region are extracted from it by cutting. Thus, replica of the micro-channels on the PDMS blocks are obtained. Inlet and outlets are punched with 1.0 mm diameter PDMS puncher and cleaned with isopropanol and acetone. Washed PDMS with microchannel channel and bonding glass (0.15 mm thickness) are plasma cleaned for 5 min. Soon after plasma cleaning is completed, PDMS is placed on the top of coverslips followed by baking on hotplate for 5 min at (90,^{circ })C. The description of master mold and fabricated Y-type microfluidic chip can be found in Supplementary 1. Using microfluidic Teflon tubing (inner diameter of 0.5 mm) reservoir is connected to inlet and outlet is joined to the flow pump (New Era Flow Pump, Model No: NE-1002X). Microfluidic chips were ensured leakage-free by flowing distilled water while the features of micro-channels are obtained using low concentration TRITC solution ((Ex/Em = 557/576) nm) before carrying out actual imaging with beads and HeLa cells.
Flow pump and data acquisition
The 2D sectional images of HeLa cells were recorded by the CCD/sCMOS camera. Two major systems were synchronized to record the data, (1) flow control system and (2) Lab-view Image Acquisition. The flow is controlled using a flow-pump operated in a withdrawl mode, and at flow-rates ranging from 500 to 2000 nl /min (for HeLa cells).
Needle of flow-pump (BD,1ml conventional syringe of diameter 4.80mm) is attached to outlet tube of the microfluidic device to withdraw the cell embedded solution from the reservoir through the micro-channels. Volumetric flow rate (in nanolitre cube) is the set point on pump. Dispense module is activated for withdrawal-mode of the pump. The pump was operated from 500 to about 2000 nl/min for data acquisition.
Raw image data are recorded using both CCD camera (Point gray camera, Model:GZL-CL-22C5M-C, Point gray, USA) and sCMOS (Zyla 4.2, Andor Tech., UK). For accurate and fast data collection the camera is interfaced with NI PCI 1433 card (frame-graber device) and is programmed using LabVIEW-12. Set point on the program are exposure-time and camera resolution. Optimum value of exposure time is decided by the thickness of light sheet and flow-rate (see, Eq. (3) ). The camera (pixel size = 5.5 µm) is configured to take data at full-frame of (2048 times 1088). Recorded data / images are saved in (*.bmp) format.
Sample preparation
Fluorescent beads
Beads were used for both calibration and as a test sample. (1 mu l) of Invitrogen FluoSpheres Carboxylate-Modified Microspheres, (1.0 upmu mathrm{m}) diameter (Nile Red fluorescent, (Ex: 535,mathrm{nm} / Em: 575) nm) (Invitrogen, USA)is suspended in (1 ,ml) of distilled water. Through mixing is done by pipetting and the mixture is loaded to the sample reservoir of VFC system for counting, imaging and parameter estimation.
Cell line and maintenance
HeLa cells (human cervical carcinoma cell line) obtained from our collaborator Dr. Upendra Nongthomba (Biological Sciences, Indian Institute of Science, Bangalore, India) were used for the experiment. The HeLa cells were cultured and maintained in incubator in complete Dulbecco’s modified minimal Eagle’s medium (DMEM) (Gibco, Thermo Fisher Scientific) supplemented with (10%) FBS( Gibco, Thermo Fisher Scientific) and (1%) penicillin –streptomycin solution (Gibco, Thermo Fisher Scientific) at (37,^{circ })C and (5% ,CO2) (CO2-incubator, Thermo Scientific). After 2 passage, the cells were prepared for experiment. Hemocytometer is used to count cells after every passage and approximately 100,000 cell count was maintained. The cells were passaged in every 2–3 days to maintain healthy cell-lines.
Mitochondrial labelling using mitotracker orange
Cells were seeded ((10^5) cells) for 24h in complete DMEM. To evaluate optimum mitotracker orange concentration needed with respect to the cell density, staining efficiency and quantum yield in imaging flow cytometry, 6 different working concentration of mitotracker orange (100, 125, 150, 175, 200, 225 nM) in DMEM were tested. Our study revealed, 175 nM mitotracker orange concentration gives optimal mitocondrial staining. For studies presented in this study, the HeLa cells (at (70%) confluence) were treated with 175 nM mitotraker orange in DMEM for 15 mins in incubator. Further, the cells were trypsinized and washes three times with phosphated buffered saline (PBS) to remove trypsine and cell debris by centrifugation (300xg for 5 mins). Subsequently, the cells were resuspended in PBS and prepared for experiments. The cells were loaded in the sample reservoir and flown through microfluidic chip for volume imaging cytometry.
