Three-Dimensional Microscopy by Milling with Ultraviolet Excitation
Jiaming Guo, Camille Artur, Jason L. Eriksen, David Mayerich
PPreprint for Review, 2019
Three-Dimensional Microscopy by Milling with UltravioletExcitation
Jiaming Guo , Camille Artur , Jason L. Eriksen , and David Mayerich * Abstract
Analysis of three-dimensional biological samples is critical to understanding tissue function and the mechanisms of disease. Manychronic conditions, like neurodegenerative diseases and cancers, correlate with complex tissue changes that are difficult to exploreusing two-dimensional histology. While three-dimensional techniques such as confocal and light-sheet microscopy are well-established,they are time consuming, require expensive instrumentation, and are limited to small tissue volumes. Three-dimensional microscopy istherefore impractical in clinical settings and often limited to core facilities at major research institutions. There would be a tremendousbenefit to providing clinicians and researchers with the ability to routinely image large three-dimensional tissue volumes at cellularresolution. In this paper, we propose an imaging methodology that enables fast and inexpensive three-dimensional imaging that canbe readily integrated into current histology pipelines. This method relies on block-face imaging of paraffin-embedded samples usingdeep-ultraviolet excitation. The imaged surface is then ablated to reveal the next tissue section for imaging. The final image stack is thenaligned and reconstructed to provide tissue models that exceed the depth and resolution achievable with modern three-dimensionalimaging systems.
Keywords
Three-Dimensional Microscopy — Ultraviolet Excitation — Microtome University of Houston , Department of Electrical and Computer Engineering University of Houston , NSF BRAIN Center University of Houston , Department of Pharmacology*
Corresponding author : [email protected]
Current biomedical research and clinical diagnoses rely heavilyon histological tissue sectioning to visualize phenotype and phe-notypic changes. However, two-dimensional sectioning provides avery limited representation of three-dimensional structure. Theselimitations are particularly difficult to reconcile for complex three-dimensional structures, such as neural and microvascular networks.Cancer and neurodegenerative diseases affect the surrounding tissuephenotype, introducing complex changes in tissue structure. Forexample, tumor-induced vascular endothelial growth factor (VEGF)stimulates angiogenesis, providing necessary substrates for tumorcell growth and spreading [1]. Neurodegenerative disorders suchas Alzheimer’s disease (AD) induce currently irreversible damagein vasculature and neural connectivity [2]. These changes are ex-tremely difficult to quantify with traditional histological sections,which are only 4 to 6µm thick.Three-dimensional (3D) microscopy, such as confocal, multi-photon, and light-sheet microscopy are common methods of 3Dimaging. However, these techniques are extremely time-consuming,limited to small ( ≈ Theoretical Approach
Deep Ultraviolet Optical Sectioning
The proposed instrument is inspired by a new imaging technologyknown as microscopy with ultraviolet surface excitation (MUSE)[9] that allows slide-free histology on intact tissue using fluorescentdyes. The main advantage of UV excitation is that light penetrationunder direct illumination is limited to the sample surface (10 µmor less) [10]. Many common fluorophores are excited by deep UV,including 4 (cid:48) ,6-diamidino-2-phenylindole (DAPI), Hoechst 33342(HO342), and eosin (Figure 2). Since glass optics block UV light,no excitation/emission filters or dichroic mirrors are necessary, sig-nificantly reducing the cost of optics while allowing simultaneousmulti-channel imaging using a color camera. In addition, the neces-sary deep-UV optics are inexpensive and readily available in theform of quartz lenses. a r X i v : . [ phy s i c s . b i o - ph ] M a y hree-Dimensional Microscopy by Milling with Ultraviolet Excitation — 2/7
10X Olympusobjectivemicrotomeblade 3-channelimagingexponential decay oblique excitationsamplestage 𝜃𝜃 𝑠𝑠 blade holdermicrotome a b 𝜃𝜃 𝑖𝑖 Figure 1.
MUVE imaging. (a) Side view of the MUVEinstrumentation showing (1) an automated microtome, (2)deep-ultraviolet (280 nm) source, and (3) standard microscopeobjective. (b) A planar view shows the optical train, where UVlight incident on the sample excites fluorescent labels that arecollected by the microscope objective. After imaging, a microtomeablates the tissue section and re-positions the sample for imagingthe next section.Penetration depth can be further controlled by using a higherincident angle, however there is a trade-off with illumination in-tensity. Furthermore, recent research shows that water-immersionMUSE achieves approximately 50% reduction in imaging depthcompared with air-immersion MUSE [11].The proposed approach relies on doping the sample embeddingmedium with a soluble UV-absorbant dye (UV27, Epolin). Increas-ing the dye concentration within the embedding medium reducesthe axial point-spread-function (PSF), providing optically thinnersections (Figure 3).
Serial Ablation 3D Imaging
Several approaches have been proposed for integrating block-facemicroscopy with serial ablation to enhance axial resolution andimage depth. Early studies rely on all-optical imaging and ablation[6], which is time-consuming but applicable to a wide range oftissues. Serial block face scanning electron microscopy (SBF-SEM)[13] uses a microtome blade for ablation, providing nanometer-scale resolution of samples embedded in hard polymers. Alternativeapproaches achieve similar results using focused ion beams [14].However, these methods are limited to extremely small micrometer-scale samples and lack molecular specificity.Integration of microtome sectioning with optical approaches hasbeen proposed for large-scale imaging. However, these instrumentsare extremely expensive to construct and difficult to maintain. Forexample, knife-edge scanning microscopy (KESM) [7, 15] requireshigh-precision stages and time-consuming sample protocols, whiletwo-photon tomography [8] requires expensive two-photon imagingsystems.The proposed approach, which we refer to as MUSE millingor milling with ultraviolet excitation (MUVE) relies on block-faceimaging, requiring the attachment of common microscope optics toan automated microtome. By leveraging the proposed UV-blockingapproach described above, it is possible to image formalin-fixedparaffin embedded (FFPE) samples using MUSE, while millingaway imaged sections using the microtome. This allows high- resolution three-dimensional reconstruction of complex samplesusing both fluorescent (Figure 2) and absorbing (Figure 4) dyes.
MUVE Instrumentation
Our prototype MUVE imaging system (Figure 1) is based on anHM355S motorized microtome (Thermo Fisher Scientific) capableof automated 0 . ×
23 Hz ≈
32 kpixel / s at 3 colorsper pixel, resulting in a throughput of approximately 96 kB / s. Thismicroscope was rigidly mounted to a two-axis translation stage(Thorlabs XYT1) for positioning and focusing. Materials and Methods
Tissue Collection and Labeling
Mice were euthanized using CO based on guidelines providedby the American Veterinary Medical Association (AVMA). Micewere then perfused transcardially with 20 mL of room temperaturephosphate-buffered saline (PBS) solution (pH 7.4), followed by20 mL of room temperature 10% neutral-buffered formalin (pH7.4). Perfusion with PBS and formalin removes blood from thecirculatory system and fixes the tissue.Mice were then perfused with 10 mL of undiluted India-inkat a rate of ≈ / s. We tested multiple vascular stains, includ-ing polyurethane resin (vasQtec PU4ii) and fluorescent tattoo ink(Skin Candy). Both fluorescent labels provided excellent contrastusing block-face imaging. However, vasQtec resin was degradedby alcohol during dehydration prior to perfusion (both ethanol andisoproyl alcohol were tested). While the fluorescent tattoo inks sur-vived embedding, the dyes were composed of fluorescent particles ≈ H OH).Optionally, tissue samples were also stained using a variety of com-pounds to provide cellular contrast, including DAPI, Hoechst, andeosin [9] for fluorescent imaging and thionine [7, 17] for negative-contrast Nissl staining.
Specimen Preparation and Embedding
Organs were embedded in paraffin wax for imaging. UV pene-tration was controlled by doping molten paraffin with up to 14%UV27 dye (Epolin). Similar protocols were followed for all rangesof doped paraffin infiltration. Organ sections were dehydrated hree-Dimensional Microscopy by Milling with Ultraviolet Excitation — 3/7 sinusoid binuclear hepatocytes e µm spermatids d Kupffercellhepatocytes10 µm spermatocytesalveoli a capillary b renal tubules glomerulus c red pulpcentral arteriolewhite pulp f pneumocytes granular cells nephrocytes Figure 2.
MUVE imaging of different mouse organs embedded in UV27-doped paraffin wax. (a) Singleplex imaging of mouse lungstained only with HO342. (b) Duplex imaging of mouse cerebellum perfused with India-ink and treated with DAPI. (c) Duplex imaging ofmouse kidney stained with Eosin and HO342. (d) Singleplex imaging of mouse testicle stained only with HO342. (e) Duplex imaging ofmouse liver perfused with India-ink and treated with HO342. (f) Duplex imaging of mouse spleen stained with Eosin and HO342.through a series of graded ethanols (70 to 100%) over the course of8 h, followed by clearing with xylene substitute (SIGMA A5597)for 3 h. Standard paraffin wax (Tissue-Tek Paraffin) was selectivelydoped with UV27 at 60 ◦ C, and samples were soaked in the selectedmixture for 2 h to allow infiltration. The paraffinization processwas performed with the aid of a tissue processor (Leica TP1020).Note that tissue shrinkage is always expected during paraffinizationprocedures and the degree of shrinkage can reach up to 40% involume for brain tissue. This can be potential avoided using matri-ces that have low shrinkage artifacts, such as glycol-methacrylateresins (Electron Microscopy Sciences Technovit 7100) or urethanerubbers (Smooth On Clear Flex 95). In particular, we found thatTechnovit was highly UV opaque, but significantly more difficultto mill.Other nuclear stains, such as DAPI and Hoechst (HO342), arecompatible with India ink perfusion. While these stains are sub- ject to bleaching during paraffin infiltration, we have found thatparaffinized samples can be stained with DAPI and Hoechst, withpenetration up to 1 mm after 3 days of in solution. For example, theHoechst solution was prepared by diluting the HO342 stock solu-tion (Thermo Fisher Hoechst 33342) 1:2000 in 1X PBS. This alsoallows staining of 1 to 5µm embedded tissue (Figure 2). Stainingwas performed by covering the block face with solution for 2 to3min prior to imaging.
Image Collection
Conventional microtome blades (DURAEDGE Low Profile) wereused for cutting, with a cutting angle of 10 ◦ (Figure 1b). Thesingle stroke operation mode of the microtome was used for semi-automated acquisition. Cutting velocities were randomized to pre-vent the reinforcement of artifacts such as knife chatter. However,the resting position of the microtome oscillates slightly around its hree-Dimensional Microscopy by Milling with Ultraviolet Excitation — 4/7 x - a x i s ( u m ) Confocal PSF at 500nm, 0.4NA x - a x i s ( u m ) Confocal PSF at 500nm, 0.8NA x - a x i s ( u m ) Confocal PSF at 500nm, 1.0NA x - a x i s ( u m ) MUVE Excitation PSF, A = 0.0 x - a x i s ( u m ) MUVE Emission PSF, A = 0.0 n o r m a li z e d i n t e n s i t y PSF Profiles with MUVE A = 0.0 confocal, 0.4NAconfocal, 0.8NAconfocal, 1.0NAMUVE x - a x i s ( u m ) MUVE Excitation PSF, A = 2.0 x - a x i s ( u m ) MUVE Emission PSF, A = 2.0 n o r m a li z e d i n t e n s i t y PSF Profiles with MUVE A = 2.0 confocal, 0.4NAconfocal, 0.8NAconfocal, 1.0NAMUVE x - a x i s ( u m ) MUVE Excitation PSF, A = 4.0 x - a x i s ( u m ) MUVE Emission PSF, A = 4.0 n o r m a li z e d i n t e n s i t y PSF Profiles with MUVE A = 4.0 confocal, 0.4NAconfocal, 0.8NAconfocal, 1.0NAMUVE
Figure 3.
Monte-Carlo simulations of confocal and MUVE point-spread-functions using coupled-wave theory for absorbance in alayered homogeneous substrate [12]. All simulations show x -polarized coherent light propagating from left to right and intensities arenormalized for each image. Contours indicate (from darkest to lightest) 1%, 10%, and 30% thresholds of maximum intensity. (red)Confocal PSFs for imaging in idealized (i.e. cleared) samples using 0.4NA (left), 0.8NA (center), and 1.0NA (right) objectives. In MUVEimaging, exponential absorbance of the excitation is the dominant factor describing the axial PSF. (purple) Incident deep-UV light isshown incident on a sample using a low-NA ( ≈ Results
Point Spread Function Characterization
The lateral resolution of MUVE is diffraction limited, and similar tofluorescence microscopy is determined by the emission wavelengthand objective numerical aperture (NA). We verified the lateral reso-lution using a USAF 1951 resolution test target (Edmund Optics). Images for this paper were acquired using a 40X Nikon objective(0.6NA). The horizontal construction of our MUVE prototype pro-hibited the use of immersion objectives, however previous work hasalready demonstrated MUSE compatibility with water-immersionoptics [11].MUVE axial resolution is dominated by the exponential ab-sorbance of the embedding medium (Figure 3). The presentedprototype provides axial resolution beyond the diffraction limit dueto limitations in the NA of air objectives. Further studies will berequired to determine the practical PSF in other imaging media.MUVE resolution benefits were validated by imaging a phan-tom composed of 1 to 5µm fluorescent green beads (CosphericPolyethylene Microspheres), (em. 515 nm) which were diluted1000-fold into UV27-doped paraffin wax. We compared MUVEwith wide-field fluorescence microscopy (Nikon Eclipse TI-E In-verted Microscope) using conventional excitation at 390 nm (DAPIexcitation). The MUVE axial PSF shows a notable improvementover the traditional pattern of the wide-field microscope (Figure 5).The benefits of the MUVE PSF come in two forms: (1) physical hree-Dimensional Microscopy by Milling with Ultraviolet Excitation — 5/7 µm a b cd e f Figure 4.
Advantages of MUVE imaging as 3D microscopy. (a)Block-face imaging of paraffin-embedded brain (top) andUV27-doped paraffin-embedded brain (bottom) using a wide-fieldfluorescence microscope with DAPI excitation (390 nm).Near-visible penetration in tissue is large in both cases, making itimpossible to reconstruct the 3D structure of microvessels. (b)Block-face imaging of paraffin-embedded brain (top) andUV27-doped paraffin-embedded brain (bottom) using MUVE(same regions as shown in (a)). Deep-UV penetration in tissue issignificantly shorter than that of near-visible and UV27 infiltrationfurther reduces the excitation volume. (c) Isosurface rendering ofparaffin-embedded brain (top) shows uneven vessel surfacereconstruction whereas isosurface rendering of UV27-dopedparaffin-embedded brain (bottom) shows sharp and smooth vesselsurface reconstruction.ablation results in a truncated asymmetric emission spot, since pre-vious layers of the sample have been removed, and (2) absorbanceof the doped embedding medium dominates the penetrating half ofthe PSF. This allows reconstruction of elements (i.e. lower parts ofspheres) obstructed using optical sectioning.
Block-Face Imaging of Fluorescent Samples
Mouse organs, including brain, kidney, liver, lung, spleen, andtesticle were embedded in UV27-doped paraffin wax and stainedafter embedding with DAPI. MUVE axial resolution sufficient toresolve individual cell bodies and their chromatin distributions.For example, two types of pneumocytes are distinguishable in thelung image (Figure 2a) and Kupffer cells and hepatocytes are alsodistinguishable in the liver (Figure 2d). Cerebellar neurons withinthe granular are also clear to determine, along with their chromatinstructure (Figure 2c). The use of oblique UV illumination revealstissue topographical information with enhanced contrast, consistentwith previously published MUSE images [9]. For example, surfaceprofiles of kidney renal tubules are visible using eosin (Figure 2e).The effectiveness of UV27 doping is shown using conventionalmicroscopy (Figure 4a,d), MUSE (Figure 4b,e) and MUVE recon-structions (Figure 4c,f). Direct comparisons in light penetrationand reconstruction behaviors between conventional paraffin em-bedding, which introduces some UV absorption, and UV27-doped µm -10 -5 0 5 10 I n t e n s i t y ( a . u . ) Axial Position (µm) isocontour holes µm Figure 5.
Central profiling of microspheres using a wide-fieldfluorescence microscope (blue) and MUVE (green). The axialmeasurements of micro-beads ( ≈ . ≈
500 µm) showing imaging artifacts such as shadows(indicated by black arrows) involved in optical sectioningmicroscopy.paraffin-embedding tissues. For instance, the 3D reconstructionof paraffin-embedded tissue shows large and rough microvascularstructures whereas the 3D reconstruction of UV27-doped paraffin-embedded tissue shows fine and smooth capillaries.
Microvascular Imaging
UV excitation is compatible with absorbing (negative) stains, wherecontrast is provided by exciting auto-fluorescence in the surround-ing tissue and embedding compound. This is particularly usefulfor microvascular reconstructions using India-ink (Figure 6), whichmitigates the need for expensive fluorescent alternatives such aslectins [19]. This data set was imaged with a lateral sample spacingof 0 .
37 µm and a 2 . hree-Dimensional Microscopy by Milling with Ultraviolet Excitation — 6/7 µm µm b c d µm µm a cut Figure 6.
Coronal MUVE imaging of mouse midbrain stained with India-ink. (a) Volume rendering of the entire data set(389 × × × × Microvascular and Nuclear Imaging
Finally, we investigate combination staining of both microvascula-ture and nuclei in the brain using thionine with India ink perfusion(Figure 8). A region of the mouse thalamus was imaged at a 0 .
37 µmlateral resolution with 1 . Discussion and Future Work
We have introduced a high-throughput imaging methodology formultiplex imaging of large-scale samples at sub-micrometer reso-lution at low cost. MUSE milling is capable of imaging densely-interconnected microvascular networks, opening the door to simpleacquisition and quantification of capillary changes common duringdisease progression [21] and guide the fabrication of in vitro diseasemodels [22]. The proposed technique is compatible with a widerange of existing objectives, and can be integrated into immersion-based imaging systems to provide lateral resolution equivalent toexisting fluorescence techniques. While MUSE milling eliminatesconstraints on sample depth, additional work on UV doping ofembedding compounds will be necessary to further reduce andquantify axial resolution. Finally, the proposed method providescomparable speed to 2D fluorescence imaging, and was able toproduce a deep microvascular network ( ≈ Acknowledgments
We would like to thank Pavel Govyadinov at University of Houstonfor his help on microvascular segmentation. This work was fundedin part by the National Science Foundation I/UCRC BRAIN Cen-ter
References [1]
Goel, H. L. & Mercurio, A. M. VEGF targets the tumour cell.
NatureReviews Cancer , 871–882 (2013). [2] Bell, R. D. & Zlokovic, B. V. Neurovascular mechanisms andblood–brain barrier disorder in alzheimer’s disease.
Acta Neuropatho-logica , 103–113 (2009). [3]
Helmchen, F. & Denk, W. Deep tissue two-photon microscopy.
NatureMethods , 932–940 (2005). [4] Vettenburg, T. et al.
Light-sheet microscopy using an airy beam.
Nature Methods , 541–544 (2014). [5] Rocha, M. D. et al.
Tissue clearing and light sheet microscopy:Imaging the unsectioned adult zebra finch brain at cellular resolution.
Frontiers in Neuroanatomy (2019). [6] Tsai, P. S. et al.
All-optical histology using ultrashort laser pulses.
Neuron , 27–41 (2003). hree-Dimensional Microscopy by Milling with Ultraviolet Excitation — 7/7 µm radius ( µm ) Figure 7.
Mouse cerebral cortex microvasculature (top) andsegmentation (bottom). The smallest microvessel (rendered inblue) acquired in this data set is ≈ ≈
148 µm. [7]
Mayerich, D., Abbott, L. & McCormick, B. Knife-edge scanning mi-croscopy for imaging and reconstruction of three-dimensional anatom-ical structures of the mouse brain.
Journal of Microscopy , 134–143 (2008). [8]
Ragan, T. et al.
Serial two-photon tomography for automated ex vivomouse brain imaging.
Nature methods , 255 (2012). [9] Fereidouni, F. et al.
Microscopy with ultraviolet surface excitation forrapid slide-free histology.
Nature Biomedical Engineering , 957–966(2017). [10] Meinhardt, M., Krebs, R., Anders, A., Heinrich, U. & Tronnier, H.Wavelength-dependent penetration depths of ultraviolet radiation inhuman skin.
Journal of Biomedical Optics , 044030 (2008). [11] Yoshitake, T. et al.
Rapid histopathological imaging of skin andbreast cancer surgical specimens using immersion microscopy withultraviolet surface excitation.
Scientific Reports , 4476 (2018). [12] Davis, B. J., Carney, P. S. & Bhargava, R. Theory of midinfraredabsorption microspectroscopy: I. homogeneous samples.
Analyticalchemistry , 3474–3486 (2010). [13] Denk, W. & Horstmann, H. Serial block-face scanning electron mi-croscopy to reconstruct three-dimensional tissue nanostructure.
PLoSbiology , e329 (2004). [14] Knott, G., Marchman, H., Wall, D. & Lich, B. Serial section scanningelectron microscopy of adult brain tissue using focused ion beammilling.
Journal of Neuroscience , 2959–2964 (2008). µm ab c cellsvessels 10 µm Figure 8.
Coronal MUVE imaging of mouse thalamus stainedwith India-ink and thionin. (a) Tissues are dark red while cellnuclei are dark brown (arrow) and vessels are black (dashed arrow)under UV illumination (left), providing enough contrast to segmentboth the cellular and vascular structures (right). (b) Volumerendering of the entire data set shows the density and organizationof microvasculature with surrounding cellular details. (c) Volumerendering of a small region (100 × × [15] Li, A. et al.
Micro-optical sectioning tomography to obtain a high-resolution atlas of the mouse brain.
Science , 1404–1408 (2010). [16]
Mayerich, D. et al.
Fast macro-scale transmission imaging of mi-crovascular networks using kesm.
Biomedical optics express , 2888–2896 (2011). [17] Choe, Y. et al.
Specimen preparation, imaging, and analysis protocolsfor knife-edge scanning microscopy.
JoVE (Journal of VisualizedExperiments) e3248 (2011). [18]
Evangelidis, G. D. & Psarakis, E. Z. Parametric image alignment usingenhanced correlation coefficient maximization.
IEEE Transactions onPattern Analysis and Machine Intelligence , 1858–1865 (2008). [19] Porter, G., Palade, G. & Milici, A. Differential binding of the lectinsgriffonia simplicifolia i and lycopersicon esculentum to microvascu-lar endothelium: organ-specific localization and partial glycoproteincharacterization.
European journal of cell biology , 85–95 (1990). [20] Govyadinov, P. A., Womack, T., Eriksen, J. L., Chen, G. & Mayerich,D. Robust tracing and visualization of heterogeneous microvascu-lar networks.
IEEE Transactions on Visualization and ComputerGraphics , 1760–1773 (2019). [21] Castillo-Carranza, D. L. et al.
Cerebral microvascular accumulationof tau oligomers in alzheimer’s disease and related tauopathies.
Agingand Disease , 257–266 (2017). [22] Guo, J. et al.
Accurate flow in augmented networks (AFAN): anapproach to generating three-dimensional biomimetic microfluidicnetworks with controlled flow.
Analytical Methods11