Theoretical background of Hilo microscopy
HiLo microscopy was described in detail by its developers Lim and Mertz elsewhere18,22. The core equation that summarises the optical sectioning capability is
$${C}_{\delta s}^{2}({\rm{z}})={A}_{s}{\int }^{}{|BP(k)|}^{2}{|OT{F}_{det}(k,z)|}^{2}OT{F}_{ill}(k,0){d}^{2}k,$$
(1)
where \({C}_{\delta s}({\rm{z}})\) is the imaged speckle contrast, As is the average transverse area of a speckle grain, BP(k) is the Bandpass filter, OTFdet(k, z) is the detection optical transfer function and OTFill(k, 0) is the illumination optical transfer function. Equation (1) has been repeated from22 and the interested reader can find the full derivation there.
We repeat only the basic principles of the HiLo process here which are necessary to reproduce our results. HiLo microscopy performs optical sectioning of fluorescent samples by segmenting the image using contrast evaluation of the difference image of a structured illumination image and a uniform illumination image and obtaining a weighting function as a result. The uniform image iu is a simple widefield fluorescent image. To obtain the structured illumination image is, the sample is illuminated by a random laser speckle pattern. The in-focus high spatial frequencies of the image are obtained by simply applying a gaussian high-pass filter to a Fourier transformed uniform image such that:
$${{\rm{i}}}_{{\rm{HP}}}={ {\mathcal F} }^{-1}({ {\mathcal I} }_{{\rm{u}}}\times {\rm{HP}}),$$
(2)
where iHP is the high-pass filtered uniform image, \({ {\mathcal F} }^{-1}\) is the inverse Fourier Transform, \({ {\mathcal I} }_{{\rm{u}}}\) is the Fourier Transform of iu and HP is a gaussian high-pass filter with cut-off frequency kc, such that HP(kc) = 1/2.
The high spatial frequencies are inherently in focus and thus do not need to be further processed. To obtain the in-focus low spatial frequencies, first the difference image, id, must be calculated
$${{\rm{i}}}_{{\rm{d}}}={{\rm{i}}}_{{\rm{s}}}-{{\rm{i}}}_{{\rm{u}}}.$$
(3)
Subtracting iu, the uniform illumination image, from is, the speckle illumination image, removes the sample induced bias and allows the evaluation of local speckle contrast to be performed on the variations of the speckle pattern only.
The local contrast of speckle grains tends to zero with defocus and thus allows to distinguish between in-focus and out-of-focus signal. This decay to zero can be accelerated by applying a bandpass filter to id prior to contrast evaluation
$${\rm{BP}}=\exp (-\frac{{{\rm{k}}}^{2}}{4{{\rm{\sigma }}}^{2}})-\exp (-\frac{{{\rm{k}}}^{2}}{2{{\rm{\sigma }}}^{2}}),$$
(4)
where BP is the bandpass filter, generated by subtracting two Gaussian lowpass filters, k is the spatial frequency and σ is the bandpass filter standard deviation.
Correct evaluation of local speckle contrast is key to separate in-focus from out-of-focus signal. Local contrast evaluation can be performed by calculating the quotient of standard deviation and mean in a local neighbourhood with a sliding window23,24.
$${{\rm{C}}}_{\langle {\rm{\Lambda }}\rangle }=\frac{{{\rm{sd}}}_{\langle {\rm{\Lambda }}\rangle }}{{\mu }_{\langle {\rm{\Lambda }}\rangle }}.$$
(5)
where C〈Λ〉 is the contrast in the local neighbourhood, evaluated within a sliding window of side length Λ (pixels). sd〈Λ〉 and µ〈Λ〉 are the standard deviation and mean intensity in the local neighbourhood respectively.
The side length Λ of the sliding window is determined depending on the cut-off frequency kc as described in reference23.
$${\rm{\Lambda }}=\frac{1}{2{{\rm{k}}}_{{\rm{c}}}}$$
(6)
Applying the local contrast as a weighting function to iu results in a coarse image of in-focus low spatial frequencies.
$${{\rm{i}}}_{{\rm{su}}}={{\rm{i}}}_{{\rm{u}}}\times {{\rm{C}}}_{\langle {\rm{\Lambda }}\rangle },$$
(7)
where isu is the weighted uniform illumination image. By applying a gaussian low-pass filter LP complementary to HP, i.e. LP + HP = 1, the in-focus low spatial frequencies are obtained
$${{\rm{i}}}_{{\rm{LP}}}={ {\mathcal F} }^{-1}({{\mathscr{ {\mathcal I} }}}_{{\rm{su}}}\times {\rm{LP}}),$$
(8)
where iLP is the in-focus low spatial frequency image, \({ {\mathcal I} }_{{\rm{su}}}\) is the Fourier Transform of isu and LP is the complementary low-pass filter. To ensure a smooth transition between iLP and iHP , a scaling factor is calculated
$${\rm{\eta }}={ {\mathcal I} }_{{\rm{HP}}}({{\rm{k}}}_{{\rm{c}}})/{ {\mathcal I} }_{{\rm{LP}}}({{\rm{k}}}_{{\rm{c}}}),$$
(9)
where η is the scaling factor, \({ {\mathcal I} }_{{\rm{HP}}}({{\rm{k}}}_{{\rm{c}}})\) and \({ {\mathcal I} }_{{\rm{LP}}}({{\rm{k}}}_{{\rm{c}}})\) are the Fourier Transforms of iHP and iLP respectively evaluated at the cut-off frequency kc.
The final optically sectioned HiLo image is obtained by adding the in-focus high and low spatial frequency images together.
$${{\rm{i}}}_{{\rm{HiLo}}}={{\rm{i}}}_{{\rm{HP}}}+{\rm{\eta }}\ast {{\rm{i}}}_{{\rm{LP}}}$$
(10)
Where iHiLo is the final optically sectioned image. By setting kc = 0.18σ22,23, the optical sectioning strength can be controlled by changing only the σ parameter.
Experimental setup
A schematic of the setup is shown in Fig.1a. A Coherent Sapphire 488-10 CDRH laser was used as a light source that emitted light at 488 nm at 4 mW peak power at the sample. The beam was guided through a variable beam expander (Thorlabs BE02-05, 2x-5x variable zoom Galilean beam expander), increasing the beam diameter from 1 mm to a minimum of 2 mm and maximum of 5 mm. The final beam diameter resulted in coarse speckle (beam expander set 2x) with higher contrast in thick specimen at the cost of optical sectioning strength or fine speckle (beam expander set to 5x) allowing thin sectioning but contrast degradation in thick samples as described in the Results section. Subsequently the beam illuminated a 1500 grit ground glass diffuser (DG20-1500, Thorlabs). The diffuser was glued to a DC motor controlled via an Arduino Uno board connected to a PC via USB. It was imaged onto the back aperture of the 0.6 NA Mesolens condenser (Mesolens Ltd.) using an aspheric lens with 0.6 NA (ACL5040U-A, Thorlabs). With the diffuser stationary, a speckle pattern was generated in the sample. Rotating the diffuser via the DC motor (6/9 V, 12000 ± 15% rpm) resulted in uniform illumination, thus allowing acquisition of both uniform and speckle illumination images in quick succession at one minute raw acquisition time per image pair on the full 4.4 mm FOV of the camera. The samples were imaged by the Mesolens onto a camera detector. The triple band emission filter was part of the commercial Mesolens system and transmitted light at 470 ± 10 nm, 540 ± 10 nm and 645 ± 50 nm. Not shown in this diagram in Fig.1a are two mirrors that are placed before and after the beam expander to guide the laser beam. Images were acquired with a thermoelectric Peltier cooled camera (VNP-29MC, Vieworks) with a chip-shifting mechanism. The chip-shifting mechanism was essential to benefit from the large FOV and high resolution (700 nm lateral, 7 µm axial1) provided by the Mesolens. The camera port on the Mesolens system contains a focusing lens providing an additional magnification of 2x bringing the total system magnification to 8x. The technical specifications of the Mesolens system have been published elsewhere1. The camera could be operated without chip-shift at a resolution of 6576 × 4384 pixels (28.8 Megapixel), with 4x chip-shift at 13152 × 8768 pixels (115.3 Megapixel) and with 9x chip-shift at 19728 × 13152 pixels (259.5 Megapixel). For HiLo imaging with the Mesolens, the chosen mode was always 9x chip-shift. In this mode, the sampling rate was 4.46 px/µm, corresponding to a 224 nm pixel size, satisfying Nyquist sampling. The sampling rate of the image was determined by imaging a 1 mm graticule (Graticule Ltd., Tonbridge, England) and equating the known distance in µm to a distance in pixels in ImageJ. The minimum frame time of the Vieworks camera was 200 ms resulting in acquisition time for one full FOV image with 9x pixel shift of 1800 ms excluding time to transfer the image data from the camera to the PC which usually took on the order of 10 seconds. In practice, acquisition of one image took 12–15 seconds including transfer of data and beginning of new image capture.
Data processing
To process the speckle and uniform images a MATLAB (R2016b version 9.1.0.441655, 64 bit, MathWorks, Inc.) script25 was written that performed HiLo imaging in the same manner as described in the previous section. This allowed more control over individual parameters (optical sectioning factor, low frequency scaling and cut-off frequency) and opened the possibility to use the parallel processing toolbox of MATLAB to use a graphics processing unit (GPU). Because of the file size of Mesolens images, it was necessary to process z-stacks of samples on a server as commercially available desktop PCs do not have enough memory to open or process such large files. The server was a Dell PowerEdge R740 with 1TB RAM and an NVIDIA Quadro P4000 GPU with 8GB video memory.
Measuring the optical sectioning capability of Mesolens HiLo
To determine the optical sectioning strength of the HiLo method, a thin fluorescent layer was set at a tilt by wedging a microscope slide under one end of the sample microscope slide, a known height difference between the two ends of the sample slide was introduced. The resulting image then showed the fluorescent layer as a narrow strip as shown in Fig.1, coming into focus in the centre of the field of view and go out of focus towards the left and right. Since the length of the sample slide is also known, a measured FWHM in the lateral direction can be translated into an axial FWHM, thus giving an experimental measure of the optical section thickness. This method was adapted after26. To prepare the thin fluorescent layer, first a 170 μm thick microscope cover slip (22 mm × 22 mm, #1.5, Thermo Fisher Scientific) was rinsed in dry acetone (Acetone 20066.330, VWR Chemicals). It was then submerged in an APTMS-acetone (3-Aminopropyldrimethoxysaline, 281778-100 ML, Sigma Aldrich) solution for six hours (0.2 mL APTMS, 9.8 mL of dry acetone). After this period, the cover slip was rinsed three times in dry acetone and blow-dried with compressed air. The cover slip was put in a 10 μM solution of fluorescein salt (Fluorescein sodium salt, 46960-25G-F, Sigma Aldrich) in distilled water. Care was taken to only let one side of the cover slip get in contact with the fluorescein solution to avoid having two thin fluorescent layers (one on either side). The bath was carefully wrapped in aluminium foil and left overnight in a dark place. The next day, the cover slip was rinsed with distilled water twice and again blow-dried with compressed air. Finally, the cover slip was mounted on a microscope slide with the dye-coated surface in contact with the slide and was sealed with nail varnish. Imaging was done with glycerol immersion.
Mouse hippocampal neuron sample preparation
The mouse hippocampal neuron sample was prepared from C57BL/6J pups (1-2 days old) as described previously27,28 and fluorescently stained29,30. All experimental procedures were performed in accordance with UK legislation including the Animals (Scientific Procedures) Act 1986 and with approval of the University of Strathclyde Animal Welfare and Ethical Review Body (AWERB). In short, neurons were fixed in ice-cold 4% paraformaldehyde (PFA). The sample was then incubated with a primary anti-mouse antibody (anti-βIII-tubulin (1:500), Sigma-Aldrich) and fluorescently labelled using a secondary antibody (anti-rabbit AlexaFluor 488 (1:200), Thermo Fisher Scientific). The fixed and stained sample was mounted onto a glass microscope slide (VWR, UK) using Vectashield mounting medium (H-1200, Vector Laboratories) and imaging was performed with glycerol immersion on the Mesolens.
5-day-old zebrafish larva sample preparation
The zebrafish were fixed in ethanol: glacial acetic acid at a 3:1 ratio at 4 °C for 72 hours, then washed in 100% ethanol and rehydrated progressively in ethanol/saline solutions before staining in 0.01% acridine orange (A1301, Thermo Fisher Scientific) in phosphate-buffered saline with gentle agitation before dehydration in an ethanol/ saline series. The dehydrated specimens were washed three times in absolute ethanol (dried with molecular sieve) and then transferred via xylene, changed twice and left in xylene for two hours and checked for transparency. They were tumbled gently overnight in a solution Fluoromount31 (Fluoromount is no longer available commercially: we would advise Histomount (Thermo Fisher Scientific) as similar substitute) before mounting in a single-cavity slide under a standard coverslip, with the specimen left uncovered to facilitate the evaporation of the xylene solvent and more mountant being added to reduce shrinkage. Imaging was performed with glycerol immersion on the Mesolensand custom built acquisition software based on WinFluor32.
