Rapid full-color serial sectioning tomography with speckle illumination and ultraviolet excitation

Rapid full-color serial sectioning tomography with speckle illumination and ultraviolet excitation

Introduction

Two-dimensional (2D) slide imaging offers detailed information about planar morphology and anatomical structures of tissues and cells, yet it fails to visualize the comprehensive internal three-dimensional (3D) structural information or spatial correspondence, which can be crucial for cancer diagnosis1,2 or embryonic development study3. Conventional 3D histology can be exceedingly labor-intensive and time-consuming, as it involves manually sectioning and staining thousands of thin tissue slices to avoid light scattering or tissue absorption4. Moreover, a sophisticated image registration algorithm is necessary to restore the lost relative spatial information among different sections5,6. Regarding this, various automated 3D optical imaging techniques have emerged and can be broadly categorized into two groups. The first category employs tissue clearing to significantly reduce light propagation issues in biological tissue7,8,9. As a representative example, light-sheet fluorescence microscopy (LSFM)10 is typically combined with tissue clearing for rapid and non-destructive whole-organ imaging. Despite significant achievements, both tissue clearing and LSFM are still under development. Tissue clearing must balance the clearing effect and time expenditure to prevent tissue degradation, especially for large samples, while LSFM needs to balance imaging resolution and volume11.

The second group relies on block-face serial sectioning tomography (BSST) to expand the imaging volume, where scanning is only required for the tissue block’s surface layer in each round12,13,14,15. Although this sectioning procedure is destructive, BSST generates inherently aligned images, and its imaging volume can be relatively easily scaled up through simple modifications to the sectioning system. Block-face imaging in BSST systems depends on the optical sectioning capabilities of the specific light microscopy applied (e.g., spinning disk-based confocal microscope16). Consequently, while straightforward in principle, the overall complexity of BSST systems can still be high, particularly when multicolor imaging is needed. Moreover, many BSST systems require whole-mount staining and embedding of imaged samples in hard supporting materials (e.g., resin or paraffin), which adds time and labor costs and induces noticeable tissue shrinkage due to dehydration17,18.

To this end, we previously introduced a whole-organ/embryo imaging method called translational rapid ultraviolet-excited sectioning tomography (TRUST)11. TRUST achieves an exceptional balance in time efficiency, imaging resolution, and cost-effectiveness by enjoying the benefits of ultraviolet (UV) surface excitation19 and real-time staining with fluorogenic dyes20. TRUST accomplishes optical sectioning by taking advantage of the shallow penetration of deep UV light in tissue. Therefore, imaging axial resolution and contrast can considerably deteriorate when the light penetration depth is significantly larger than the objective lens’s depth of field (DOF). This can occur when imaging tissue with relatively low scattering properties (e.g., human breast, ~100 µm UV penetration depth21) or when using a high-magnification objective lens22. In addition, the coarse sectioning thickness (50 µm) in TRUST worsens axial resolution and leads to tissue information loss, as certain tissue layers might not be exposed to the imaging system.

To address these issues, we present HiLoTRUST, which integrates the high-and-low-frequency (HiLo) microscopy23 into TRUST for better optical sectioning capability and imaging contrast by rejecting the out-of-focus fluorescence background. Unlike other structured illumination microscopy (SIM)-based widefield optical sectioning methods, HiLo only requires two shots per field of view (FOV): one captured under speckle illumination and another under uniform illumination. Furthermore, compared to patterned illumination techniques, this approach is more robust against aberrations or scattering in tissues due to the statistically invariant nature of fully developed speckles24. Also, HiLoTRUST utilizes a finer sectioning thickness (10–15 µm) to enhance tissue continuity in the axial direction, resulting in more accurate and detailed 3D image reconstruction (Supplementary Figure 1).

To retain the full-color imaging capability of TRUST, we have also developed full-color HiLo microscopy, which offers improved color contrast and more information compared to conventional monochrome HiLo microscopy. Although we have previously proposed dual-channel HiLo microscopy for 2D histological imaging, it relies on sequential channel imaging at the expense of acquisition time25. Full-color HiLo microscopy employs deep UV light-emitting diodes (LEDs) for uniform illumination, allowing the excitation of labeled stains and endogenous tissue components (Supplementary Figure 2). The resulting fluorescence signals can be simultaneously captured by the integrated color camera. For speckle illumination, a green 532 nm laser diode (LD) is used, and a notch filter (NF) at 532 nm serves as the excitation bandstop filter while permitting most of the visible spectra to pass. Unlike traditional illumination mode switching methods (i.e., uniform or speckle illumination) that depend on mechanical high-speed scanning with a galvanometer24 or rotating the diffuser25, HiLoTRUST uses a microcontroller unit for illumination light sources and mode switching. This low-cost electronic approach eliminates potential system vibration and lengthy start-stop times associated with mechanical methods.

To evaluate the effectiveness and reliability of full-color HiLo microscopy, we initially conducted 2D imaging on various mouse organs (e.g., liver, lung, kidney, and brain). We have also devised adapted HiLo processing pipelines for challenging situations where low-frequency non-nuclear structures may not be excited under speckle illumination. Then, we implemented 3D imaging with HiLoTRUST for mouse organs (e.g., kidney, lung, and liver), and compared the obtained 3D datasets with imaging results from TRUST, highlighting a superior optical sectioning capability. Finally, we captured images of two human lung cancer samples, showcasing HiLoTRUST’s potential for clinical applications. In summary, HiLoTRUST offers a cost-effective, full-color, and high-resolution 3D imaging approach, holding great promise for 3D histology applications.

Results

Setup of the full-color HiLo microscopy

The schematic of the full-color HiLo microscopy for 2D block-face imaging is shown in Fig. 1a. To acquire the speckle illumination image (({I}_{s})), the 532 nm green LD is turned on, emitting the light beam with a diameter of 2 mm. After passing a 220-grit ground glass diffuser (DG10-220, Thorlabs Inc.), the generated speckle pattern is then relayed onto the back focal plane of the objective lens25,26,27 (UPlanFL N 10× /0.3 numerical aperture (NA), RMS10X-PF, Olympus Corp.) by a pair of expansion lenses (L1 and L2, with focal lengths of 25 mm and 75 mm, respectively), forming a speckle pattern with an average grain size of ~4 µm. The side window length (Lambda) is set as 10 µm when calculating the local contrast ({C}_{s}). Subsequently, the fluorescence signals from the excited cell nuclei labeled with PI will be collected by the same objective lens. After passing the beam splitter (BS) (30 R:70 T, BSS10R, Thorlabs Inc.) and the 532 nm NF, the fluorescence signals will be focused on the color camera (DS-Fi3, Nikon Inc.) by an infinity-corrected tube lens (TTL180-A, Thorlabs Inc.). The 532 nm NF positioned just before the tube lens (TL) functions as an excitation bandstop filter, allowing the majority of visible signals to pass through.

Fig. 1: Illustrations of full-color HiLo microscopy.
Rapid full-color serial sectioning tomography with speckle illumination and ultraviolet excitation

a Schematic of full-color HiLo microscopy. Two shots are required for each FOV. Under speckle illumination (left), the beam from the 532 nm laser diode is projected to the surface of the diffuser. The resulting speckle pattern is relayed to the back focal plane of the objective lens with a pair of expansion lenses, and finally projected onto the surface of the imaged sample. The excited fluorescence signals will then be collected by the objective lens, pass through the beam splitter, and finally be focused on the color camera. Under uniform illumination (right), the LD is powered off, and two UV-LEDs are activated, projecting UV beams obliquely from two slides. The fluorescence signals will follow the same collection path as in the speckle image acquisition step. bg An intuitive example illustrating the principles of full-color HiLo microscopy, where speckle illumination image (({I}_{s})), uniform illumination image (({I}_{u})), speckle contrast (({C}_{s})), in-focus low-frequency components (({I}_{{Lo}})), and the final output HiLo image (({I}_{{HiLo}})) of the same FOV of a mouse kidney is presented. LD, laser diode; L1, lens 1; L2, lens 2; OL, objective lens; BS, beam splitter; NF, notch filter; TL, tube lens; M, mirror. Note: images in bg have been normalized with min-max normalization for better visualization.

Full size image

To transit to uniform illumination, the LD is turned off while the two mounted deep UV-LEDs ( ~ 285 nm, M285L4, Thorlabs Inc.) are activated simultaneously. Dual UV-LEDs are applied for high illumination energy and uniformity. The UV light beams are roughly focused on the surface of the specimen by two pairs of lenses (#67-270, Edmund Optics Inc.; LA4306-UV, Thorlabs Inc.) with an oblique orientation. The excited fluorescence signals will be collected by the objective lens and propagate back following the same optical path as it is in speckle illumination. The narrow bandwidth of the 532 nm NF ( ~ 10 nm) allows the transmission of most fluorescence signals in the visible spectrum, enabling full-color imaging (Supplementary Figure 3). The results in Fig. 1b–g present an intuitive example to better illustrate the principles of full-color HiLo microscopy. Figure 1b,c are two shots (i.e., ({I}_{s}) and ({I}_{u})) acquired under speckle illumination with LD and uniform illumination with UV-LEDs, respectively. Under speckle illumination, only PI-labeled cell nuclei can be excited. Because the emission spectrum presents a peak around 615 nm, the red channel of the acquired color image is extracted as the speckle illumination image ({I}_{s}). The speckle contrast image ({C}_{s}) (Fig. 1d) is calculated following Eq. (4) to reject the out-of-focus low-frequency components in Fig. 1e, outputting only the in-focus low-frequency image ({I}_{{Lo}}) (Fig. 1f). Finally, the calculated in-focus high-frequency components (({I}_{{Hi}})) and in-focus low-frequency components (({I}_{{Lo}})) can be merged as the final output ({I}_{{HiLo}}) (Fig. 1g) based on Eq. (6).

HiLoTRUST system setup

The HiLoTRUST system is developed by integrating the 3-axis motorized stage (Motor-xyz, L-509.20SD00, PI miCos GmbH) and the vibratome (VF-500-0Z, Precisionary Instruments Inc.) into the full-color HiLo microscopy, as depicted in Fig. 2a. Prior to imaging, the specimen is translated near the objective lens for focusing. At the same time, the specimen is also near the wall of the water tank such that the gap between the objective lens and the tissue surface will be mostly filled with air instead of stains to minimize the fluorescence background arising from staining solutions. Part of the water tank near the objective lens is made of quartz, allowing transmission of both the UV illumination and the excited fluorescence signals in the visible wavelength range.

Fig. 2: Overview of 3D imaging by HiLoTRUST.
figure 2

a Schematic of the HiLoTRUST system. b Workflow of the 3D imaging with HiLoTRUST, which contains six steps: (1) fluorescent staining with DAPI and PI for ~150 s, (2) widefield imaging with raster-scanning under speckle illumination by LD, (3) widefield imaging with raster-scanning under uniform illumination by two UV-LEDs, (4) shaving off the imaged layer with the vibratome to expose a layer underneath, (5) HiLo image processing, (6) 3D reconstruction by direct image stacking without image spatial registration. During imaging, the previous four steps will be repeated until the entire sample has been imaged and the whole procedure is fully automated with lab-built hardware and control programs. Note: TRUST and HiLoTRUST images have been normalized with min-max normalization for better visualization.

Full size image

During imaging, two motorized linear stages (Motor-x and Motor-y) translate the specimen along the xy plane for raster scanning when the FOV is insufficient to cover the entire section. An additional Motor-z is employed for focal scanning along the axial direction. After image scanning, Motor-xyz positions the specimen near the vibratome’s blade for subsequent sectioning. The tissue sectioning thickness depends on the relative spatial position between the blade and the specimen. Hence, precise manual spatial localization of the translation stage is necessary during system alignment to guarantee sectioning accuracy. Once the system is calibrated, further experiments can proceed without manual recalibration, as the spatial positions are tracked by three custom limit switches attached to the motors (i.e., Motor-x, Motor-y, and Motor-z).

The entire 3D imaging part consists of repeatedly performing the first four steps shown in Fig. 2b: (1) fluorescent staining of the tissue’s top layer for approximately 150 s, (2) widefield imaging using raster-scanning and focal scanning under speckle illumination by the LD, (3) widefield imaging using raster-scanning and focal scanning under uniform illumination by two deep UV-LEDs, and (4) removing the imaged layer with the vibratome to reveal the layer beneath. After HiLo image processing in Step (5), 3D reconstruction in Step (6) can be realized by direct image stacking because the produced image sections are inherently aligned.

The selected vibratome is capable of sectioning samples with diameters up to 20 mm, and if necessary, an alternative model (VF-800-0Z, Precision Instruments Inc.) can be employed to increase the sectioning diameter to 100 mm. We also modified the VF-500-0Z vibratome to enhance its compatibility with the HiLoTRUST system. For example, we replaced the original water tank, which contained chemical solutions (DAPI and PI) for real-time staining, with a larger 3D-printed version to accommodate more space for 3D scanning and the placement of mechanical components such as the sample stage shown in Fig. 2a. Furthermore, we removed the original vibratome’s sectioning thickness control unit and integrated another motorized stage (Motor-s: L-509.10SD00; PI miCos GmbH; not depicted in Fig. 2a) to regulate the sectioning thickness automatically by adjusting the sample stage’s position, considering the blade’s position remains fixed. To guarantee that the objective lens’s focal plane aligns parallel to the sample surface defined by the vibratome’s blade angle, we placed a custom-built 2-axis angle adjustable platform (adjusting pitch and roll, as seen in TRUST11) beneath the vibratome and used a commercial high-precision rotation mount (Rotator in Fig. 2a) to modify the yaw angle.

Validation of full-color HiLo microscopy

We initially evaluated the high-content and optical sectioning ability of the full-color HiLo microscopy by capturing 2D HiLo images (i.e., HiLoTRUST images) of various mouse organs (such as lung, kidney, liver, and brain) and comparing the results with widefield images (i.e., TRUST images), as illustrated in Fig. 3. The HiLo images (right half in Fig. 3a1–a4,b1–b4) display a significant improvement in image contrast compared to the widefield images (left half in Fig. 3a1–a4,b1–b4) due to the reduction of out-of-focus background, which helps in identifying finer structures. Root mean square (RMS) contrast28 was also computed for images in Fig. 3b1–b4, which shows a quantitative comparison of the image contrast. This feature is particularly important when dealing with highly transparent tissue samples, where the UV light’s penetration depth may exceed the objective lens’s DOF, resulting in reduced image contrast. For example, in the zoomed-in image of lung tissue (Fig. 3b1), elastic fibers or alveoli exhibit increased clarity and distinguishability in the HiLo image (right) compared to the widefield image (left). Similarly, the nucleus in kidney tissue (Fig. 3b2), hepatocyte in liver tissue (Fig. 3b3), and nerve fibers in brain tissue (Fig. 3b4) all appear more distinct in the HiLo images (right) compared to the widefield images (left). As shown in Fig. 3c1–c4, we also conducted a quantitative analysis of the imaging contrast improvement by plotting the normalized intensity distributions along the solid white lines in Fig. 3b1–b4 and calculating the standard deviation (({std})) of the line profiles.

Fig. 3: Experimental validation of full-color HiLo microscopy.
figure 3

a1a4 Pairs of widefield images (left) and HiLo images (right) of mouse lung, kidney, liver, and brain tissue, respectively. b1b4 Zoomed-in pairs of TRUST images (left) and HiLoTRUST images (right) of the regions indicated in a1a4. c1c4 Line profiles of solid white lines regions marked in b1b4 with min-max normalization, respectively. Note: widefield and HiLo images have been normalized with min-max normalization for better visualization (see Supplementary Figure 4 for HiLo images without normalization).

Full size image

Under uniform illumination with UV-LEDs, autofluorescence signals from unlabeled components such as lipids, collagen, and fibers can also be excited and captured by the color camera11, providing HiLoTRUST with high-content imaging capability. However, under speckle illumination with 532 nm LD, only the cell nucleus is excited. The speckle illumination image ({I}_{s}) (e.g., Fig. 1b) in HiLo microscopy is mainly used to generate the contrast image ({C}_{s}) (e.g., Fig. 1g) working as the weighting map to reject the out-of-focus background. Considering the dense distribution of cell nuclei and the current axial resolution achieved ( ~ 5.8 µm), the weighting map ({C}_{s}) based on cell nuclear information can be employed to evaluate the focusing status throughout the entire FOV in typical situations (e.g., Fig. 1b–g). Nonetheless, in regions where cell nuclear density is low and large non-nuclear components exist, ({C}_{s}) may not accurately represent the focus status. For instance, the hyaline cartilage shown in the widefield image (Fig. 4a, top) is not adequately represented in the corresponding HiLo image (Fig. 4a, bottom). To alleviate this challenging situation, we have modified the standard HiLo processing pipeline with two feasible solutions. The first method aims to mask out regions that HiLo processing is not suitable. The mask (Fig. 4c, top) is segmented based on its low-frequency property using the background removal plugin in ImageJ29. The resulting image (Fig. 4c, bottom) retains the hyaline cartilage region while maintaining optical sectioning capability for other regions.

Fig. 4: Adapted HiLo processing pipelines for challenging cases.
figure 4

a, b Two examples of challenging cases where directly applying standard HiLo processing is not appropriate. c, d The first adapted HiLo processing pipeline by masking out regions where standard HiLo processing is unsuitable. e, f The second adapted HiLo processing assisted with HFE. HFE, high-frequency emphasis. Note: widefield and HiLo images have been normalized with min-max normalization for better visualization.

Full size image

The second method involves selectively applying HiLo processing only in the color channel with the majority of fluorescence signals excited by speckle illumination (i.e., red channel, Fig. 4e, top). For other color channels, instead of HiLo processing, a high-frequency emphasis (HFE) filter30 is applied, which preserves the low-frequency information content of the input image (({I}_{{input}})) while enhancing its high-frequency components. Note that compared to HiLo, HFE is also an image contrast enhancement technique but cannot effectively reject out-of-focus information as HiLo (Supplementary Figure 5). The resulting output image (({I}_{{HFE}})) can be described as:

$${I}_{{HFE}}left(vec{r}right)={{mathcal{F}}}^{-1}left{{rm{a}}+{rm{b}}cdot {mathcal{F}}left[{I}_{{input}}left(vec{r}right)right]times {HP}left(vec{k}right)right}$$
(1)

where (age 0) and (b ,>, 1). The outcome of combining HFE with HiLo processing is displayed in the bottom half of Fig. 4e by setting (a=0.5) and (b=2). To verify the versatility, the two adapted HiLo processing pipelines have also been applied to mouse brain tissue (Fig. 4b, d, f).

3D imaging with HiLoTRUST

It is important to note that, HiLoTRUST achieved a finer sectioning interval ( ~ 10 µm) by upgrading the vibratome model from VF-700-0z (used in TRUST) to VF-500-0z (used in HiLoTRUST) compared to the original TRUST ( ~ 50 µm). This improvement and additional focal scanning already resulted in better continuity along the axial direction for the reconstructed 3D image (Supplementary Figure 1). However, to make the comparison fair and emphasize the optical sectioning performance improvement of HiLoTRUST, the sectioning thickness and focal scanning parameters were set to be the same when acquiring TRUST or HiLoTRUST images in Fig. 5 and Fig. 6.

Fig. 5: 2D/3D HiLoTRUST images of mouse organs.
figure 5

a 2D imaging results of the mouse kidney with TRUST (top left) and HiLoTRUST (bottom right). b Zoomed-in image comparison between TRUST (top) and HiLoTRUST (bottom) of the region in a marked with a solid green box. c Reconstructed 3D imaging results of the kidney tissue with TRUST (left) and HiLoTRUST (right). d 2D imaging results of the mouse lung with TRUST (top left) and HiLoTRUST (bottom right). e, f Close-up image comparisons between TRUST (top) and HiLoTRUST (bottom) of the region in d marked with a dashed orange box, and a solid red box, respectively. g Reconstructed 3D imaging results of the lung tissue block with TRUST (left) and HiLoTRUST (right). h 2D imaging results of the mouse liver comparing TRUST (top) and HiLoTRUST (bottom). i Reconstructed 3D imaging results of the liver tissue block with TRUST (top) and HiLoTRUST (bottom). j The extracted vessel network from the same liver tissue block based on negative contrast. (k) The histogram of the vessel radius in j. Note: TRUST and HiLoTRUST images have been normalized with min-max normalization for better visualization.

Full size image
Fig. 6: 2D/3D HiLoTRUST images of human lung cancer specimens.
figure 6

a Cross-sectional imaging of the first specimen using HiLoTRUST (top left) and TRUST (bottom right). The inset at the bottom left shows a photograph of the specimen. b Zoomed-in TRUST image (left) and HiLoTRUST image (right) of the region in (a) marked with a solid orange box. c Another cross-section of the first specimen imaged with HiLoTRUST (top left) and TRUST (bottom right). d Zoomed-in TRUST image and HiLoTRUST image of the region in c marked with a solid green box. e 3D TRUST image of the first specimen. f Zoomed-in 3D TRUST image of the region marked by a solid orange box in e. g 3D HiLoTRUST image of the first specimen. h Zoomed-in 3D HiLoTRUST image of the region marked by a solid green box in g. i Zoomed-in 3D HiLoTRUST image of the region marked by a solid red box in h. j Cross-sectional imaging with HiLoTRUST (top left) and TRUST (bottom right) of the second specimen. The inset at the bottom left shows a photograph of the specimen. k Zoomed-in TRUST image (left) and HiLoTRUST image (right) of the region in j marked with a solid orange box. l 3D distribution of the desmoplastic stroma extracted from the reconstructed 3D model in m utilizing color difference. n A magnified image of the solid white box region in (l) overlaid with nuclei within cells (predominantly cancer cells). This visualization provides an approximation of the 3D distribution relationship between collagen fibers and cancer cells. Note: TRUST and HiLoTRUST images have been normalized with min-max normalization for better visualization.

Full size image

To assess the 3D imaging capability of HiLoTRUST, mouse kidney (Fig. 5a–c), lung (Fig. 5d–g, Supplementary Figure 6), and liver (Fig. 5h–j) tissue blocks were imaged. 2D/3D results of the same tissue block acquired with TRUST and HiLoTRUST are compared to demonstrate the enhanced imaging contrast achieved through HiLo processing. For example, Fig. 5b,c are 2D/3D zoomed-in images showing the vasa recta in the medulla region marked with a solid green box in Fig. 5a, which serve as the countercurrent exchange system for solutes and water. Also, Fig. 5e,f are two close-up 2D image pairs of a mouse lung tissue, acquired with TRUST (top) and HiLoTRUST (bottom). Figure 5(g) is the rendered 3D model of a mouse lung tissue block imaged with TRUST (left) and HiLoTRUST (right), highlighting the significance of improved optical sectioning ability for tissue-level volumetric imaging. The 3D animation of the whole mouse lung block, including serial cross-sections, has also been rendered in Supplementary Movie 1. Finally, as shown in Fig. 5j, the vessel network in the liver tissue block (Fig. 5i) can be extracted based on the signal difference because their autofluorescence signal intensity is lower than that of the surrounding tissues31,32. With the trainable Weka segmentation plugin in Fiji33, the vessel network can be segmented and the histogram of the vessel radius can be analyzed as in Fig. 5k with a mean value of 55 μm. It should be noted that only relatively large vessels ( > 20 μm in radius) have been successfully extracted, possibly due to the relatively low image contrast of vessels.

To demonstrate the potential of HiLoTRUST in clinical applications, imaging results of two human lung cancer specimens are presented in Fig. 6. Figure 6a shows the 2D imaging results of the lung specimen’s cross-section layer from the first patient, with the top left half featuring the HiLoTRUST image and the bottom right half showcasing the TRUST image. Figure 6b offers a magnified comparison of the solid orange box region marked in Fig. 6a. Similarly, Fig. 6c depicts another cross-section layer of the same specimen. The magnified comparison in Fig. 6d demonstrates that the rejection of an out-of-focus background in HiLoTRUST significantly enhances the distinguishability of components such as elastin fibers, making them easier to identify. Figure 6e–i presents the comparison between TRUST and HiLoTRUST on 3D imaging. The zoomed-in result (Fig. 6i) of the region in Fig. 6h marked with the solid red box shows the clear 3D distribution of cells and fibers, providing evidence of the improved image contrast achieved by HiLoTRUST.

Similarly, Fig. 6j shows the 2D image of the lung specimen’s surface layer from the second patient, acquired with HiLoTRUST (top left) and TRUST (bottom right). Figure 6k provides a magnified 2D comparison between TRUST and HiLoTRUST of the solid orange box marked in Fig. 6j. Due to the full-color imaging capability of HiLoTRUST, it is feasible to extract the 3D distribution of the desmoplastic stroma by utilizing color difference (i.e., green channel, Fig. 6l) based on the reconstructed 3D model in Fig. 6m. Figure 6n displays a zoomed-in view of the solid white box area in Fig. 6l, with cell nuclei superimposed, primarily from cancer cells. This visualization demonstrates the 3D distribution relationship between collagen fibers and cancer cells to some degree, showcasing HiLoTRUST’s potential in cancer development research.

Discussion

In summary, TRUST has proven its potential in 3D histology or developmental study by whole-organ/embryo imaging. Here, regarding its limited axial resolution, we developed HiLoTRUST based on the full-color HiLo microscopy and have experimentally demonstrated that HiLoTRUST enables high-content full-color 3D imaging with subcellular resolution in a time- and cost-effective manner by the two-shot widefield illumination optical sectioning technique and the inexpensive light sources (i.e., green LD and a pair of UV-LEDs). To the best of our knowledge, this work demonstrates the first application of using HiLo microscopy assisted with deep UV surface excitation for full-color serial sectioning tomography. The improved performance has been validated by imaging diverse types of mouse organs (e.g., lung, liver, kidney, and brain). Moreover, human lung specimens with adenocarcinoma have also been imaged with the 3D distribution of desmoplasia stroma and cell nuclei being extracted, demonstrating its promising potential as a 3D histological imaging platform for clinical applications.

HiLoTRUST is still in its early development stage, and future research development can be conducted to further improve the performance of the system. The axial resolution of HiLo microscopy defined by FWHM has been reported to be less than 2 µm when imaging an optically thin sample23. To maintain acceptable in-focus speckle contrast, relatively large speckle patterns can be applied but at the cost of compromised axial resolution25. Even though, the current axial resolution of 5.8 µm is sufficient to produce an optical section comparable to a physical section in conventional slide-based histology.

One significant advantage of HiLoTRUST, when compared with other advanced 3D imaging systems (Supplementary Table 1), is related to its high-content multicolor imaging capability19 with an outstanding balance in terms of time efficiency (sample preparation time and imaging speed), imaging resolution, and cost-effectiveness by enjoying the benefits of the UV surface excitation and real-time labeling. In order to further exploit the spectral information of the color images, computational spectral super-resolution methods34,35,36 can be a feasible working direction, which can generate hyperspectral images with RGB images as the input. To broaden the applications of HiLoTRUST in the life science field, it is vital to explore more types of fluorogenic stains (including aggregation-induced emission-active fluorescent probes37), to enrich the imaging capabilities of HiLoTRUST, such as the protein (8-anilino-1-naphthalene sulfonic acid, Sigma-Aldrich) stain and lipids stain (Nile red or Dil, Sigma-Aldrich). However, HiLoTRUST’s speckle illumination relies on a single green LD, which poses challenges. Although, as shown in Fig. 4, two adapted HiLo processing pipelines have been proposed to mitigate this issue, the imaging performance is still compromised. To address this, multiple LDs at different wavelengths combined with stains targeting various tissue components can be applied (Supplementary Figure 7). Additionally, a similar setup to the CHAMP system38, originally designed for super-resolution slide-free imaging, can be adapted for full-color HiLo imaging, where speckle illumination generated by a 266-nm laser is obliquely projected onto the tissue surface (Supplementary Figure 8).

As a proof of concept, although formalin-fixed tumor specimens were imaged, this study is limited by the small number of specimens. To show 3D HiLoTRUST’s superiority over 2D histology in clinics, future work covering large amounts of human samples is necessary to quantify the diagnostic metrics (e.g., cancer diagnostic accuracy improvement). Finally, the practicality of HiLoTRUST can be further improved for clinical translations. For example, deep learning-assisted virtual staining could be incorporated with HiLoTRUST to further improve the efficiency of the current pathological workflow39.

One key superiority of TRUST lies in its rapid whole-organ imaging speed (e.g., ~2 days for an E15 mouse embryo). In comparison to TRUST, the enhanced axial resolution of HiLoTRUST necessitates finer mechanical sectioning, precise focal scanning, and double shots per FOV, consequently multiplying the time required for imaging. To accelerate the acquisition speed, firstly, incorporating more light sources, such as multiple UV-LEDs (e.g., four), and upgrading to a higher-powered UV-LED (M280L, 78 mW, Thorlabs Inc.) from the current model (M285L4, 45 mW, Thorlabs Inc.) could reduce the exposure time. Additionally, using a low-magnification objective lens (e.g., 4×) and a camera with a substantially larger chip size (e.g., PCO Edge 5.5; chip size: 16.64 × 14 mm) can significantly decrease the total number of FOVs. The solution also enhances tolerance to surface roughness due to the increased DOF of the low-magnification objective lens. As for the degraded imaging resolution, it can be recovered with super-resolution neural networks40 (Supplementary Figure 9). With the aid of deep learning (e.g., pix2pix41), it is also possible to achieve single-shot imaging for each FOV in HiLo microscopy (Supplementary Figure 10). Furthermore, unlike the original TRUST system, where the imaging is achieved by translating the whole optical system raster scanning over the tissue surface, the sample is translated instead in the HiLoTRUST system. The much lighter weight makes high-speed mechanical scanning feasible. By updating the current motor (20 mm/s; L-509.x0SD00, PI miCos GmbH) to an advanced model (90 mm/s; L-511, PI miCos GmbH), the motor scanning speed can be enhanced by a factor of five approximately. Finally, regarding the low-speed mechanical axial scanning, extended DOF via dynamic remote focusing42 holds great promise to speed up the whole imaging process by eliminating the need for axial scanning.

Methods

Sample preparation and processing

Once two mice (wild-type C57BL/6, two months old) were euthanized through five minutes of carbon dioxide inhalation following guidelines provided by the American Veterinary Medical Association, internal organs (e.g., kidney, lung, and liver) were harvested, rinsed with phosphate-buffered saline solution for a minute, and finally fixed by submerging in 10% neutral-buffered formalin (NBF) at room temperature for 24 h. Two human cancer specimens, considered as leftover tissue (i.e., no longer needed for assessment of diagnostic, prognostic, and other parameters in the diagnosis and treatment of the patient), were immediately fixed in formalin for at least 24 h after surgical excision.

Embedding tissue samples in agarose/gelatin contributes to improved sectioning quality. The suggested concentration of agarose or vibratome working parameters should be adjusted based on the tissue type43. To section hard specimens (e.g., cancer tissue), 10% (w/v) gelatin embedding with post-fixation by NBF overnight is preferred, as it can tighten the connection between the sample surface and surrounding gelatin after crosslinking44,45.

Whole-mount staining is not applied in HiLoTRUST. To reduce the labor and time cost for staining, imaged samples will be immersed in staining solutions throughout the experiment for real-time staining11. The specimen will be labeled along with image scanning and serial sectioning. To minimize fluorescence background noise from the staining solution itself, we used two of the most common and commercially available fluorogenic stains20, 4’,6-diamidino-2-phenylindole (DAPI, ({lambda }_{{Ex}}{lambda }_{{Em}}:,364/454{rm{nm}})) and propidium iodide (PI, ({lambda }_{{Ex}}{/lambda }_{{Em}}:,535/615{rm{nm}})) at a concentration of 5 µg/ml, which exhibit increased fluorescence upon binding to cell nuclei.

The human sample protocol was approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (HKU/HA HKW) (reference number: UW20-335) and conforms with the human research ethics protocol approved by the Health, Safety, and Environment Office (HSEO) of The Hong Kong University of Science and Technology (HKUST) (license number: HREP-2021-0270). Informed consent was obtained from all lung cancer tissue donors. All animal experiments were conducted in conformity with the animal protocol approved by the HSEO of the HKUST (license number: AEP-2021-0082).

Basics of HiLo microscopy and resolution measurement

The principles of HiLo microscopy have been extensively discussed, including our recent research on its application for rapid slide-free histological imaging25. In brief, two shots are acquired for each FOV: the speckle illumination image (({I}_{s})) and the uniform illumination image (({I}_{u})). The optical sectioning capability of HiLo microscopy is based on the distinct characteristics of information at different frequencies. In-focus high-frequency components (({I}_{{Hi}})) and in-focus low-frequency components (({I}_{{Lo}})) are calculated separately and then combined as the final output image (({I}_{{HiLo}})). ({I}_{{Hi}}) is naturally in-focus and can be directly extracted by applying a high-pass Gaussian filter on ({I}_{u}) as below:

$${I}_{{Hi}}left(vec{r}right)={{mathscr{F}}}^{-1}left{{mathscr{F}}left[{I}_{u}left(vec{r}right)right]times {HP}left(vec{k}right)right}$$
(2)

where ({mathscr{F}}{mathscr{(}}{mathscr{bullet }}{mathscr{)}}) and ({{mathscr{F}}}^{-1}(bullet )) denote the Fourier transform and inverse Fourier transform, respectively. (vec{r}) and (vec{k}) represent the spatial and frequency coordinate, respectively. ({HP}) is a high-pass Gaussian filter with a cut-off frequency at ({k}_{c}). ({I}_{{Lo}}) cannot be extracted simply by applying a complementary low-pass Gaussian filter on ({I}_{u}) (i.e., ({{mathscr{F}}}^{-1}left{{mathscr{F}}left[{I}_{u}left(vec{r}right)right]times {LP}left(vec{k}right)right})), because the processed image results (e.g., Fig. 1e) would still contain out-of-focus information. To address the issue, ({I}_{s}) is needed for evaluating the local focusing status and can effectively eliminate unwanted out-of-focus information. ({I}_{{Lo}}left(vec{r}right)) can be determined by

$${I}_{{Lo}}left(vec{r}right)={{mathcal{F}}}^{-1}left{{mathcal{F}}left[{{C}_{s}left(vec{r}right)cdot I}_{u}left(vec{r}right)right]times {LP}left(vec{k}right)right}$$
(3)

where the low-pass filter ({LP}) is complementary to the high-pass filter ({HP}) by ({LP}left(vec{k}right)=1-{HP}left(vec{k}right)). Speckle contrast (({C}_{s})) functions as an object-independent measure of the relative proportion of the uniform illumination image ({I}_{u}) that is in-focus, providing the optical sectioning ability for the low-frequency components of the object and rejecting out-of-focus backgrounds23. ({C}_{s}) can be calculated based on the local contrast of the speckle image ({I}_{s}) with the equation below:

$${C}_{s}left(vec{r}right)=frac{{{sd}}_{Lambda }left[{I}_{d}left(vec{r}right)right]}{{mu }_{Lambda }left[{I}_{s}left(vec{r}right)right]}$$
(4)

where ({{sd}}_{Lambda }left(bullet right)) and ({mu }_{Lambda }left(bullet right)) represent the local standard deviation and local mean value over a square window with a side length of (Lambda), respectively. (Lambda) can be determined by (Lambda =1/2{k}_{c}) to obtain the full spatial spectrum of the object27. ({I}_{d}) is the image difference which can be calculated as ({I}_{d}left(vec{r}right)={I}_{s}left(vec{r}right)-{I}_{u}left(vec{r}right)).

To further accelerate the decay efficiency of ({C}_{s}left(vec{r}right)) for better axial resolution, band-pass filter ({BP}) can be applied to ({I}_{d}) before the calculation of ({C}_{s}). ({BP}) can be generated by subtracting two Gaussian low-pass filters as below:

$${BP}left(vec{k}right)=exp left(-frac{{left|kright|}^{2}}{2{sigma }^{2}}right)-exp left(-frac{{left|kright|}^{2}}{{sigma }^{2}}right)$$
(5)

where (sigma) is the optical sectioning parameter determining the axial resolution of HiLo microscopy23,25,46. Smaller (sigma) although makes a better axial resolution, but at the cost of decreased contrast of ({C}_{s}) (Supplementary Figure 11). Finally, the calculated in-focus high-frequency components (({I}_{{Hi}})) and in-focus low-frequency components (({I}_{{Lo}})) can be merged as the final output ({I}_{{HiLo}}) by the equation below:

$${I}_{{HiLo}}left(vec{r}right)={I}_{{Hi}}left(vec{r}right)+eta cdot {I}_{{Lo}}left(vec{r}right)$$
(6)

where (eta) is a scaling factor to ensure the seamless transition from the low-pass and high-pass information across the cut-off frequency ({k}_{c}) (approximately equal to (0.18sigma)24). (eta) can be determined experimentally by imaging results25 or directly calculated based on the equation below at ({k}_{c})23.

$$eta =frac{left|{mathscr{F}}left[{I}_{{Hi}}left(vec{r}right)right]right|}{left|{mathscr{F}}left[{I}_{u}left(vec{r}right)right]times {LP}left(vec{k}right)right|}$$
(7)

To quantify the axial resolution ({R}_{{axial}}) of the HiLo microscopy, a specimen by dissolving 10-µm-diameter fluorescent microspheres into 2% w/v melted agarose was prepared. Then, focal scanning over 50 µm with a step size of 1 µm was conducted, with representative results shown in Supplementary Figure 12a–d. The axial intensity distribution curves for multiple fluorescent microsphere beads can be extracted with an averaged Gaussian fitted curve shown in Supplementary Figure 12e. The full width at half maximum (FWHM) was measured as 11.6 µm. Assuming Gaussian-shaped intensity profiles of the beads as well as a Gaussian-shaped point spread function of the microscope, axial resolution ({R}_{{axial}}) was calculated as 5.8 µm based on the equation25,46 below:

$${R}_{{axial}}=sqrt{{{FWHM}}^{2}-{d}_{{bead}}^{2}}$$
(8)

where ({d}_{{bead}}) is the average diameter of fluorescent microspheres (i.e., 10 µm).

Related Articles

Optical sorting: past, present and future

Optical sorting combines optical tweezers with diverse techniques, including optical spectrum, artificial intelligence (AI) and immunoassay, to endow unprecedented capabilities in particle sorting. In comparison to other methods such as microfluidics, acoustics and electrophoresis, optical sorting offers appreciable advantages in nanoscale precision, high resolution, non-invasiveness, and is becoming increasingly indispensable in fields of biophysics, chemistry, and materials science. This review aims to offer a comprehensive overview of the history, development, and perspectives of various optical sorting techniques, categorised as passive and active sorting methods. To begin, we elucidate the fundamental physics and attributes of both conventional and exotic optical forces. We then explore sorting capabilities of active optical sorting, which fuses optical tweezers with a diversity of techniques, including Raman spectroscopy and machine learning. Afterwards, we reveal the essential roles played by deterministic light fields, configured with lens systems or metasurfaces, in the passive sorting of particles based on their varying sizes and shapes, sorting resolutions and speeds. We conclude with our vision of the most promising and futuristic directions, including AI-facilitated ultrafast and bio-morphology-selective sorting. It can be envisioned that optical sorting will inevitably become a revolutionary tool in scientific research and practical biomedical applications.

Photovoltaic bioelectronics merging biology with new generation semiconductors and light in biophotovoltaics photobiomodulation and biosensing

This review covers advancements in biosensing, biophotovoltaics, and photobiomodulation, focusing on the synergistic use of light, biomaterials, cells or tissues, interfaced with photosensitive dye-sensitized, perovskite, and conjugated polymer organic semiconductors or nanoparticles. Integration of semiconductor and biological systems, using non-invasive light-probes or -stimuli for both sensing and controlling biological behavior, has led to groundbreaking applications like artificial retinas. From fusion of photovoltaics and biology, a new research field emerges: photovoltaic bioelectronics.

Label-free live cell recognition and tracking for biological discoveries and translational applications

Label-free, live cell recognition (i.e. instance segmentation) and tracking using computer vision-aided recognition can be a powerful tool that rapidly generates multi-modal readouts of cell populations at single cell resolution. However, this technology remains hindered by the lack of accurate, universal algorithms. This review presents related biological and computer vision concepts to bridge these disciplines, paving the way for broad applications in cell-based diagnostics, drug discovery, and biomanufacturing.

From superior contrast to super resolution label free optical microscopy

Label-free optical microscopy utilizes the information encoded in light scattered off unlabeled particles to generate the images. This review article starts off with a discussion on how this light matter interaction gives rise to the issues of poor-contrast and diffraction-limited spatial resolution. Then, this article reviews the various far-field label-free optical microscopy techniques that have been developed, with an emphasis on the physical mechanisms behind the image formation processes in such techniques. Thus the article aims to elucidate the various state-of-the-art label-free techniques and their current applications.

Targeting of TAMs: can we be more clever than cancer cells?

With increasing incidence and geography, cancer is one of the leading causes of death, reduced quality of life and disability worldwide. Principal progress in the development of new anticancer therapies, in improving the efficiency of immunotherapeutic tools, and in the personification of conventional therapies needs to consider cancer-specific and patient-specific programming of innate immunity. Intratumoral TAMs and their precursors, resident macrophages and monocytes, are principal regulators of tumor progression and therapy resistance. Our review summarizes the accumulated evidence for the subpopulations of TAMs and their increasing number of biomarkers, indicating their predictive value for the clinical parameters of carcinogenesis and therapy resistance, with a focus on solid cancers of non-infectious etiology. We present the state-of-the-art knowledge about the tumor-supporting functions of TAMs at all stages of tumor progression and highlight biomarkers, recently identified by single-cell and spatial analytical methods, that discriminate between tumor-promoting and tumor-inhibiting TAMs, where both subtypes express a combination of prototype M1 and M2 genes. Our review focuses on novel mechanisms involved in the crosstalk among epigenetic, signaling, transcriptional and metabolic pathways in TAMs. Particular attention has been given to the recently identified link between cancer cell metabolism and the epigenetic programming of TAMs by histone lactylation, which can be responsible for the unlimited protumoral programming of TAMs. Finally, we explain how TAMs interfere with currently used anticancer therapeutics and summarize the most advanced data from clinical trials, which we divide into four categories: inhibition of TAM survival and differentiation, inhibition of monocyte/TAM recruitment into tumors, functional reprogramming of TAMs, and genetic enhancement of macrophages.

Responses

Your email address will not be published. Required fields are marked *