Reviving formalin-fixed, paraffin embedded (FFPE) tissues for on-slide and multiscale correlative microscopy Authors: Zenan Chen, Chad Moore, Samson Dowland, John DeWitt, Douglas J Taatjes, Gregory Milburn, Kenneth S Campbell, Amy Li, Filip Braet, Gerald J Shami Published: 2026 ABSTRACT AbstractCorrelative microscopy integrates complementary imaging modalities to enable structural analysis of biological samples across multiple length scales. Here, we present two correlative workflows for formalin-fixed, paraffin-embedded (FFPE) human tissues that extend the utility of archived clinical specimens for high-resolution imaging. The first, an on-slide correlative light and electron microscopy (CLEM) workflow, enables direct integration of brightfield light microscopy with scanning electron microscopy to resolve ultrastructural features from standard histological sections. The second, a multiscale three-dimensional workflow, combines X-ray micro-computed tomography with serial block-face scanning electron microscopy to enable targeted, volumetric ultrastructural analysis within a defined tissue context. Using these approaches, we demonstrate that FFPE tissues retain sufficient structural integrity to resolve cellular and subcellular architecture across liver, lung, brain, and heart specimens. Finally, we apply artificial intelligence-assisted image processing and segmentation to enable scalable, quantitative analysis of large-volume datasets. Together, these workflows establish FFPE specimens as a versatile and accessible resource for multiscale correlative imaging in both research and clinical settings.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-56823-9.Keywords: Formalin-fixed paraffin embedded, Correlative microscopy, Correlative light and electron microscopy, Volume electron microscopy, Artificial intelligence, HistopathologySubject terms: Biological techniques, Optics and photonics Introduction Formalin fixation and paraffin embedding (FFPE), combined with chromogenic staining and brightfield light microscopy (BFLM), remains the gold standard for histological assessment of tissues in both biomedical research and clinical diagnostics1–3. This workflow enables robust evaluation of cellular morphology and tissue architecture from micron-to-millimetre scales, supports standardised and automated tissue processing, and is compatible with a broad range of chromogenic and immunohistochemical stains4. Despite these advantages, FFPE-based imaging is largely restricted to the collection of two-dimensional (2-D) data of structures larger than ~ 250 nm, limiting spatial interpretation of complex tissue architecture and precluding detailed ultrastructural analysis5. To address these limitations, correlative microscopy (CM) approaches have been developed to integrate complementary imaging modalities, leveraging their respective strengths to provide structural and functional insights into biological samples6–12. For example, CM can integrate the non-destructive three-dimensional (3-D) imaging capability of X-ray micro-computed tomography (micro-CT)13, the molecular and staining versatility of light microscopy14, and the nanometre-scale resolution of electron microscopy15,16. By integrating data across spatial scales and contrast mechanisms, CM enables comprehensive characterisation of biological specimens, encompassing structural organisation, cellular interactions and molecular information17,18. Clinicopathological FFPE specimens represent an underutilised yet highly valuable resource for such multiscale analyses, given their widespread availability and the substantial structural and biochemical information they retain19. In this study, we build upon and extend existing CM approaches for paraffin sections and reprocessed FFPE human tissues. The first, an on-slide CLEM workflow, integrates BFLM with inverted backscattered scanning electron (BSE) microscopy, enabling direct correlation of chromogenic staining with ultrastructural detail. The second, a multiscale 3-D workflow, integrates large-volume, tissue-scale imaging via micro-CT with high-resolution volumetric ultrastructural analysis using serial block-face scanning electron microscopy (SBF-SEM). Collectively, these workflows demonstrate that FFPE specimens retain substantial untapped information, establishing their utility for advanced imaging applications in both biomedical research and clinical diagnostics. Materials and methods Sample selection A variety of human FFPE tissue samples, including liver, lung, heart and brain were obtained from international biobanks at the University of Vermont and the University of Kentucky. These diverse tissue types were selected to evaluate the applicability of the two workflows employed in this study. All procedures were conducted in accordance with the ethical standards of the Declaration of Helsinki20. On-slide CLEM workflow Sample preparation and light microscopy FFPE tissue blocks were sectioned at 5 μm thickness using a microtome and mounted onto glass slides. Sections were dewaxed in xylene (3 × 5 min) and rehydrated through a graded ethanol series (100%, 90%, 70%; 2 min each). Sections were stained using either toluidine blue or hematoxylin and eosin (H&E) following standard protocols. For toluidine blue staining, sections were incubated in 0.5% toluidine blue and 1% sodium borate for 2 min. For H&E staining, sections were immersed in Mayer’s haematoxylin for 2 min, rinsed in running tap water, blued in tap water and stained with eosin Y for 2 min. Following staining, slides were rinsed in double-distilled water (ddH2O) (3 × 2 min) and mounted in ddH2O. Whole-slide imaging was performed using a Pannoramic MIDI II Rx slide scanner (3Dhistech) equipped with a 20×/0.8 NA objective to capture the entire tissue section (Fig. 1). Schematic overview of the on-slide CLEM workflow. (A) Formalin-fixed, paraffin embedded tissue; (B) 5 μm-thick sections are cut using a microtome and mounted onto glass slides; (C) sections are dewaxed, rehydrated and stained with chromogenic dyes (e.g. toluidine blue, H&E); (D) sections are mounted in water, coverslipped and imaged using a slide scanner; (E) coverslips are removed; sections are post-stained with uranyl acetate and lead citrate, then carbon-coated; (F) the same region of interest imaged in (D) is relocated and imaged using SEM. Schematic created in BioRender.21. Scanning electron microscopy Following BFLM imaging, the coverglass was gently removed and the slide was dried on a heating block (55 °C). Tissue sections were stained with 2% aqueous uranyl acetate for 5 min in darkness, rinsed with warm (60 °C) ddH₂O, then stained with lead citrate for 5 min. Slides were again rinsed with warm ddH₂O and dried on a heating block (55 °C). To minimise electrostatic charging during imaging, slides were coated with a 10 nm layer of carbon using a carbon sputter coater. An aluminium stub was affixed to the underside of each slide using double-sided conductive carbon tape, and a strip of silver electrodag was applied from the top surface of the slide to the stub, to establish an electrically conductive path. BFLM images of the same slides were viewed using the open-source software QuPath (version 0.5.1), and prominent morphological features (e.g. large blood vessels, bronchioles) were used as landmarks to relocate regions of interest (ROIs) across light and electron imaging modalities. High-resolution inverted BSE micrographs were acquired at an accelerating voltage of 3 kV under high vacuum (HV) using a Zeiss Gemini field-emission SEM (FE-SEM) (Fig. 3), or at 5 kV using a Phenom XL desktop SEM (Thermo Scientific) under HV (Fig. 4). Images were aligned and overlaid in Adobe Photoshop using morphological landmarks as reference points (Fig. 1). On-slide CLEM workflow in liver tissue. (A) Low-magnification toluidine blue-stained liver section showing overall tissue architecture. (B) Higher-magnification light micrograph of the ROI in (A), highlighting components of the portal triad, including a branch of the hepatic portal vein (hPV), hepatic artery (hPA), and bile duct (hBD). (C) Corresponding BSE image of the ROI shown in (A). (D) High-magnification BSE image of a hepatic parenchymal cell, revealing the nucleus (N), endoplasmic reticulum (ER), mitochondria (M) and glycogen (G). (E, F) Comparative BSE and SE2 micrographs demonstrating differential contrast for improved discrimination of cellular structures. Scale bars: (A) 300 μm; (B, C) 100 μm; (D) 2 μm; (E, F) 50 μm. On-slide CLEM workflow in lung tissue. (A) BFLM image of H&E-stained human lung. (B) Corresponding BSE image of the same region in (A). (C) Overlay of (A) and (B). (D) Higher magnification BFLM image of the ROI indicated by the yellow bounding box in (A). (E) Corresponding BSE of the same region in (D). (F) Overlay of (D) and (E). Scale bars: (A-C) 100 μm; (D-F) 50 μm. Multiscale 3-D workflow Sample preparation For each sample, a 5 × 5 × 5 mm tissue block was excised from FFPE material using a single edge razor blade and further sectioned into 1 × 1 × 1 mm blocks. Tissues were deparaffinised in xylene (4 × 15 min) and gradually rehydrated through a graded ethanol series: 100% ethanol (2 × 15 min), followed by 90% and 70% ethanol for 10 min each. Samples were processed using a modified en bloc rOTO staining protocol for SBF-SEM22. Tissues were rinsed in 0.1 M Sorensen’s Phosphate Buffer pH 7.4 (SPB) (2 × 15 min) and then post-fixed in 2% osmium tetroxide and 1.5% potassium ferrocyanide in 0.1 M SPB pH 7.4 for 1 h in darkness. Samples were washed in ddH2O (3 × 5 min) and placed in 1% aqueous thiocarbohydrazide solution for 20 min. Tissues were washed in ddH2O (3 × 5 min) and stained with 2% aqueous osmium tetroxide for 30 min in darkness. Samples were washed in ddH2O (3 × 5 min) and placed in 1% aqueous uranyl acetate in darkness at 4 ˚C overnight. Samples were washed in ddH2O (3 × 5 min) and stained in Walton’s lead aspartate for 30 min at 60 ˚C. Samples were washed in ddH2O (3 × 5 min) and then dehydrated with 30%, 50%, 70%, 90% ethanol (1 × 5 min), 100% ethanol (2 × 5 min) and 100% acetone (2 × 10 min). Samples were gradually infiltrated and embedded in hard grade Procure 812 resin and polymerised at 60 ˚C for 48 h. After polymerisation, resin blocks were trimmed using an ultramicrotome (Leica EM UC7) and a semithin survey section stained with toluidine blue was examined to assess the quality of sample preparation before proceeding with micro-CT imaging. X-ray micro-CT imaging (micro-CT) and precision trimming Samples were imaged using a Skyskan 2214 system (Bruker micro-CT, Kontich, Belgium) operated at 50 kV over a 180° rotation angle. An initial low-resolution scan was acquired to obtain a global overview of the tissue block. A total of 506 projection images were collected (2,024 × 2,024 pixels; voxel size = 4 μm; frame averaging = 3; exposure time = 750 ms), with no filter applied. Projections were acquired at an angular increment of 0.4° using the instrument’s computer-controlled rotation stage. Based on the spatial coordinates from the original scan, the block was precision trimmed using an ultramicrotome to expose a volume of interest. A second high-resolution scan was then acquired at 40 kV (931 projections; 2,024 × 2,024 pixels; voxel size = 0.6 μm; frame averaging = 10; exposure time = 2,472 ms), with no filter applied. Projections were acquired at an angular increment of 0.2°. Datasets were reconstructed using NRecon and registration was performed using Dataviewer (part of the Bruker 3D.Suite, version 2.2.0.6). Serial block-face scanning electron microscopy (SBF-SEM) Following micro-CT imaging, samples were mounted on an aluminium pin with cyanoacrylate glue and sputter coated with 10 nm gold. SBF-SEM was conducted on a Zeiss Gemini FE-SEM equipped with a Gatan 3View2XP system and an OnPoint BSE detector. Images (8,192 × 8,192 pixels; XY pixel size = 10 nm; Z pixel size [slice thickness] = 100 nm) were acquired at 3 kV using variable pressure mode. Image processing, segmentation and quantitative analysis of SBF-SEM data SBF-SEM datasets were normalised and binned using FIJI23and automatically aligned using the StackReg plugin24. Artificial intelligence (AI)-enabled denoising was performed using the Noise-2-Void algorithm implemented in Dragonfly 3D World (Comet Technologies Canada Inc., Montreal, Canada)25. Manual segmentation was performed in 3dmod (IMOD image processing suite)26by tracing features across serial sections. Quantitative measurements were extracted using the ‘Get Info’ function in 3dmod. For large-volume datasets, segmentation was performed using a hybrid approach combining manual annotation and AI-assisted segmentation. Deep learning models were trained on a per-dataset basis using the Dragonfly “Deep Learning Tools”, which leverage U-Net convolutional neural networks27. Representative structures (e.g. surfactant granules, nuclei and blood vessels) were manually annotated to generate training datasets. Segmentation outputs were subsequently validated by manual inspection to validate accuracy and consistency (see Fig. 2). Schematic overview of the multiscale 3-D workflow. (A) FFPE tissue; (B) resected tissue block (~ 5 × 5 × 5 mm); (C) chemical fixation, en bloc staining and resin embedding; (D) micro-CT imaging of the tissue block; (E) precision trimming by ultramicrotomy guided by micro-CT; (F) SBF-SEM imaging of the targeted region. Schematic created in BioRender.28. Results On-slide CLEM workflow We first assessed the feasibility of reprocessing FFPE tissues for CM (Fig. 3). Toluidine blue stained liver tissue examined using BFLM (Fig. 3A, B) revealed well-preserved tissue architecture, including the components of the portal triad (portal vein, hepatic artery and bile ducts). At intermediate magnification, chords of hepatic parenchymal cells radiating around larger vessels were clearly distinguishable and were separated by the intervening hepatic sinusoids. Relocation of the same region of interest previously examined by BFLM (Fig. 3B) and subsequently examined with EM (Fig. 3C), revealed markedly enhanced contrast, thereby facilitating greater differentiation of cellular structures, including plasma membranes, nuclei and connective tissue components. High magnification imaging using EM revealed subcellular features including heterochromatin and euchromatin organisation within the nucleus, glycogen rosettes, rough endoplasmic reticulum and mitochondria (Fig. 3D). BSE and secondary electron (SE2) imaging modes were also compared to evaluate what additional information each could provide. Inverted BSE micrographs revealed higher contrast and ultrastructural detail, with minimal depth information (Fig. 3D, E). Conversely, SE2 micrographs revealed lower image contrast, however enhanced topological and depth-related information (Fig. 3F). The combined use of both imaging modes yielded complementary structural insights that were not attainable when either mode was used in isolation. The on-slide CLEM workflow was also applied to a human lung sample derived from a patient with hypertensive and atherosclerotic cardiovascular disease. BFLM revealed a large immune infiltrate at the interalveolar septum, surrounded by blood vessels and alveolar air spaces (Fig. 4A). This region was relocated and imaged using EM (Fig. 4B) and successfully overlaid (Fig. 4C). The ease of the correlative alignment was expedited due to the minimal tissue distortion (i.e. shrinkage or swelling) provided by the glass slide substrate. High magnification EM investigation was performed on a region enriched in leukocytes, which were identified based on the accumulation of carbon deposits within their cytoplasm. Multiscale 3-D workflow Micro-CT was performed on a reprocessed FFPE lung biopsy taken from a patient with organising pneumonia. An average intensity projection image of the 3-D volume revealed an intricate network of radio-dense, anastomosing vascular structures separated by radio-lucent airspaces (Fig. 5A, E; Supplementary Material 1). Gaps or ruptures in the vasculature were also observed, likely reflecting pneumonia-associated complications. From the same resin block, a region of interest was trimmed (Fig. 5A-B, F-H) and examined by SBF-SEM (Fig. 5C-D, I-J). Manual and AI-assisted segmentation were combined to reconstruct cell morphology, tissue organisation and extract morphometric data (Figs. 5 and 6 and Supplementary Material 2). Multiscale 3D workflow applied to human lung tissue. (A) Low-magnification micro-CT volume of the lung sample within the resin block, showing an intricate network of radio-dense, anastomosing vascular structures separated by radio-lucent airspaces (yellow bounding box). The inset (white bounding box) shows the registered higher-resolution volume (green) acquired following precision trimming. (B) 2-D micro-CT XY slice corresponding to the green rendered volume in (A). (C) SBF-SEM block-face image of the same region shown in (B). White arrowheads indicate features visible in both datasets, confirming spatial correspondence of the ROI. (D) BSE image of the ROI indicated in (C), revealing pulmonary ultrastructure, including the blood–air barrier formed by type I pneumocytes and pulmonary endothelial cells separated by a shared basement membrane. (E) Volume rendering of the low-magnification micro-CT dataset indicated in (A) (yellow bounding box). (F–H) Volume renderings of the high-resolution micro-CT dataset indicated by the white bounding boxes in (A, E) and corresponding to the XY slice in (B). Colour legend: alveolar septum (blue), airspace (green), vasculature (red). (I, J) 3-D renderings of the SBF-SEM dataset corresponding to the ROI in (C) and low-magnification XY image in (D). The alveolar septum (bright blue) contains pulmonary capillaries (cream/orange), within which red blood cells (red) are visible in the lumen. Alveolar macrophages (dark blue) are distributed throughout the tissue and contain digestive vacuoles (orange). Scale bars: (B, C, F–H) 200 μm; (D, I, J) 20 μm; (E) 1 mm. Morphological and morphometric features of human lung cells. Representative SBF-SEM images of (A) pulmonary endothelial cells (PECs), (C) type I pneumocytes (T1Ps), (E) type II pneumocytes (T2Ps), and (G) alveolar macrophages (AMs). Corresponding 3-D renderings of segmented cells, showing the plasma membrane and nucleus, are presented in (B, D, F, H). Proteolytic granules within the cytoplasm of alveolar macrophages are visible in (G). (I–N) Quantitative morphometric analyses of cell surface area (I), volume (J), surface area-to-volume ratio (K), maximum cell length (L), minimum cell width (M), and aspect ratio (N) for PECs, T1Ps, T2Ps, and AMs. Data are presented as mean ± SD with individual data points overlaid. Scale bars: 5 μm. SBF-SEM revealed the cellular components of the blood–air barrier, including type I (T1P) and type II pneumocytes (T2P), endothelial cells (PEC) and the basement membrane on which these cells rested. Both T1Ps and PECs exhibited squamous morphologies, as reflected by their high aspect ratios (2.44 ± 0.39 and 4.57 ± 0.72, respectively). Likewise, T1Ps and PECs displayed high surface area-to-volume ratios (1.36 ± 0.22 and 1.95 ± 0.41, respectively), which is commensurate with their function in facilitating gas exchange between the blood-air barrier. Within the interstitial space, numerous leukocytes, including pulmonary alveolar- and foamy macrophages were visible; the latter were identified by their lipid-rich subcellular granules. Collectively, these cells retained their characteristic morphology, however displayed variable fixation quality at the subcellular scale. Notably, lipid-containing structures, such as lipid droplets in foamy macrophages and surfactant-containing vesicles in type II pneumocytes appeared extracted (Fig. 6E), whereas proteolytic granules in alveolar leukocytes, including alveolar macrophages, were well preserved (Fig. 6G). Having validated the workflow and its expedience in generating high-resolution EM information, SBF-SEM was performed on human brain and heart samples. The brain biopsy was derived from a patient with hypertension and atherosclerotic cardiovascular disease. 3-D renderings were generated of the vascular network, neurons and glial cells (Fig. 7, Supplementary Material 3). Given the lipid-rich composition of neural tissue, significant extraction of structures including myelin was evident, however axonal and dendritic processes remained observable and could be traced throughout the volume. Neuronal somata were large and conspicuous, measuring 13.83 μm ± 1.86 μm in diameter, and contained a centrally located, highly euchromatic nucleus (Fig. 7I). Nissl substance was well preserved and distributed throughout the perikaryon, appearing as granular, electron-dense material. At higher magnification, components of the blood-brain barrier were evident, including endothelial cells (Fig. 7H), pericytes (Fig. 7G) surrounding the endothelium, and astrocytes (Fig. 7F), which displayed a highly ruffled cytoplasm that extended into long protoplasmic processes that contacted endothelial cells (Fig. 7E). SBF-SEM and 3-D reconstruction of human brain tissue. (A) BSE image of human brain tissue, revealing neurons and glial cells. Despite extraction of lipid-rich structures such as myelin, dendritic and axonal processes, these structures remain discernible. (B–D) 3-D rendering of brain tissue showing capillaries formed by endothelial cells (red; nuclei, yellow), surrounded by pericytes (lilac; nuclei, green), with astrocytes (blue/grey; nuclei, cyan) in close association. Neuronal somata (blue; nuclei, magenta) are also visible and contain granular Nissl substance (orange). (E–H) Reveals high-magnification views of the cellular components of the blood–brain barrier. Astrocytes (blue/grey; F) extend cytoplasmic processes that terminate as endfeet on pericytes (lilac; G), which in turn envelop endothelial cells (red; H). (I) 3-D rendering of a neuronal soma (blue), showing the nucleus (magenta) and Nissl substance (orange) within the perikaryon. Scale bars: (A–D) 30 μm; (E–I) 10 μm. The heart sample was obtained from a patient with non-ischemic heart disease. At low magnification, the myocardium displayed pronounced structural alterations, including separation and distortion of cardiomyocytes, as well as shrinkage artefacts. Individual cardiomyocytes were detached from neighbouring cells and from the surrounding interstitium, within which blood vessels were embedded (Fig. 8A). AI-assisted modelling of the dataset revealed an intricate vascular network, composed of endothelial cells and interspersed fibroblasts distributed throughout the tissue. Cardiomyocytes were observed as highly elongated cells, within which contractile units, bounded by Z-discs, were clearly discernible. The cardiomyocyte cytoplasm contained numerous lipid droplets and abundant deposits of glycogen and lipofuscin, the latter of which is associated with cellular ageing and disease (Figs. 8B–F, Supplementary Material 4). SBF-SEM and 3-D reconstruction of human heart tissue. (A) Representative SBF-SEM image of human heart tissue showing cardiomyocytes in longitudinal profile. Individual myofibrils are clearly resolved, with separation between adjacent cardiomyocytes and the surrounding interstitial space. At this magnification, lipofuscin and glycogen deposits, appearing as electron-dense granules, as well as Z-discs, are readily identifiable within cardiomyocytes. (B) 3-D rendering of cardiomyocytes (brown) and Z-discs (purple). (C-F) 3-D renderings highlighting additional tissue and subcellular features, including fibroblast nuclei (cobalt blue), cardiomyocyte nuclei (green), lipid droplets (cyan), vasculature (red), and endothelial nuclei (yellow). Scale bars: (A-F) 20 μm. Discussion In this study, we present two correlative imaging workflows for FFPE tissues that extend existing methodologies and enable integrated structural analysis of archived human specimens across tissue and subcellular scales. The extended analysis of FFPE samples offers many advantages, including enabling the direct investigation of disease pathogenesis in clinical specimens29, the abundance, availability and stability of archived specimens30,31, reduced logistical barriers associated with fresh tissue collection32 and compatibility with complementary analytical techniques such as genomics30and proteomics33. In the on-slide CLEM workflow, chromogenic staining (e.g., toluidine blue, H&E) was first performed, followed by BFLM, enabling clear visualisation of tissue architecture and differentiation of cell types based on colour contrast (Figs. 3A and B and 4A and D). ROIs were subsequently correlated with high-resolution FE-SEM (Fig. 3C-F). By leveraging a contemporary low-kV BSE detector, ultrastructural information could be resolved directly from 5 μm-thick paraffin sections, including chromatin, mitochondria, endoplasmic reticulum, and glycogen (Fig. 3D). These results extend prior studies by Inaga, et al.34, Kawasaki, et al.7 and Sawaguchi, et al.12, who applied similar workflows using low-vacuum SEM on thick paraffin sections. In contrast, our approach employs low-kV BSE imaging under high-vacuum conditions, enabling improved contrast and resolution from conventionally prepared FFPE material. We further validated the integration of a desktop SEM in this workflow, which similarly revealed ultrastructural detail not accessible by BFLM alone. This enhances translational applicability, particularly in clinical environments where desktop SEM systems are increasingly adopted. Although ultrastructural preservation was not equivalent to that achieved with samples prepared specifically for electron microscopy, key subcellular features remained resolvable. This enables interrogation of pathological alterations, inter-organelle relationships, and morphometric parameters in human clinical specimens. Notably, the workflow can also be applied to existing histological slides, extending its utility to archived material without requiring additional sample preparation. Examination of FFPE samples by EM also enables the use of multiple detectors, such as SE2 and BSE, to acquire complementary structural information from the same region of interest. SE2 imaging provided enhanced topographical and depth information (Fig. 3F), particularly for structures such as erythrocytes. This capability enables morphological classification of erythrocytes (e.g., biconcave, acanthocytic, and echinocytic forms), which is useful for supporting the diagnosis and study of conditions including alcoholic cirrhosis as well as many blood disorders35–37. BSE imaging provided significantly improved contrast, generating TEM-like images that enabled clear visualisation of subcellular structures. The on-slide CLEM workflow also offers a rapid and accessible approach for correlating millimetre-scale tissue context with ultrastructural detail. Whole-slide imaging was automated using a slide scanner, while relocation and overlay of light- and electron-microscopy images – guided by prominent morphological landmarks – was straightforward due to minimal tissue distortion between modalities. Direct imaging on glass slides eliminated the need for resin infiltration, embedding, and ultrathin sectioning, reducing sample preparation from several days to hours. The workflow is also compatible with desktop scanning electron microscopes, which are more compact, cost-efficient, and user-friendly, thereby enhancing accessibility of CLEM workflows, particularly in diagnostic settings. Attempts to further extend the workflow through integration of energy-dispersive X-ray spectroscopy (EDS), an analytical technique compatible with SEM, were explored (Supplementary Material 5). While preliminary analyses suggested the potential for elemental characterisation within the same correlative workflow, signal-to-noise and elemental specificity were insufficient for reliable interpretation under the current imaging and sample preparation conditions. Further optimisation of acquisition parameters, together with the use of samples containing higher atomic number elements more amenable to EDS detection, will likely be required to fully realise the potential of this approach within the workflow38. Having established the utility of FFPE samples for CM, we next extended this approach to a multiscale 3-D workflow that integrates imaging across length scales. Tissues were prepared using a modified en bloc staining protocol to generate contrast compatible with both micro-CT and SBF-SEM22,39. While micro-CT can be used, as demonstrated here, to guide ROI selection and targeted trimming, its primary value lies in providing µm-to-mm tissue-scale information in a fully non-destructive manner, enabling volumetric assessment of tissue architecture prior to higher-resolution EM analysis. This approach offers several advantages in that it preserves the integrity of the sample for downstream workflows, allows rapid identification of regions of interest at the organ or tissue scale and provides contextual information that complements ultrastructural datasets. A key limitation, however, is the limited availability of selective stains for highlighting specific tissue features, necessitating a strong understanding of the sample to accurately interpret the data. Following micro-CT, SBF-SEM provided higher-resolution ultrastructural information, including the alveolar septum, which comprises the blood–air barrier and is formed by the alveolar epithelium, capillary endothelium, and a shared basement membrane. Notably, it was possible to resolve the extremely thin cytoplasmic processes (0.2–0.4 μm) through which gas exchange occurs across the blood–air barrier. These processes, formed by cytoplasmic extensions of endothelial cells and type I pneumocytes, are below the resolution limits of BFLM and micro-CT. Our observation of foamy alveolar macrophages in this sample was also consistent with previous reports describing the presence of these cells in pathologies associated with broncho-bronchiolar obstruction, such as organising pneumonia40. Notably, while similar micro-CT-guided correlative workflows have been reported41, our implementation extends these approaches through integration with reprocessed FFPE material and compatibility with heavy metal staining optimised for volume EM. This enables enhanced contrast for both micro-CT imaging and SBF-SEM. Similarly, SBF-SEM was applied to brain and heart FFPE samples to evaluate the broader applicability of the workflow independent of micro-CT guidance. In brain tissue, neuronal somata, dendritic and axonal processes, and glial networks were readily identifiable, enabling assessment of cellular organisation and intercellular connectivity. Notably, myelin extraction was pronounced, likely reflecting the absence of lipid-preserving fixatives in the FFPE workflow. Heart tissue showed good overall preservation, with vascular structures and capillary networks remaining clearly discernible, supporting analysis of tissue-level organisation and cellular interactions. Importantly, AI-assisted 3-D reconstruction revealed structural features not readily apparent in 2-D EM, including the spatial relationships between cardiomyocytes, interspersed vasculature, and stromal populations such as fibroblasts and endothelial cells. This added spatial context enhances interpretation of tissue architecture and is particularly informative for assessing pathological change. Subcellular features, including lipofuscin – a well-established marker of cellular ageing42 – could also be reliably modelled. The application of deep learning-based image analysis provided substantial gains in scalability and analysis throughput of lung, brain, and heart datasets. Once trained, the software was able to automatically identify numerous and complex structures of interest across large-volume datasets, substantially reducing analysis time and enabling the visualisation of well-defined cellular and subcellular features that would be impractical to reconstruct using conventional manual tracing approaches. However, the AI-based segmentation approach used in this study is sensitive to variability in sample preparation and, consequently, data quality. This limitation is not unexpected, given that the specimens were originally processed for routine histopathological assessment and exhibited suboptimal ultrastructural preservation and preparation-related artefacts. Accordingly, manual segmentation was also employed to enable expert-guided delineation of fine structural features, particularly closely apposed membranes in highly squamous cells (e.g., PECs and T1Ps). As training datasets expand and algorithms mature, the accuracy and generalisability of AI-driven segmentation are expected to improve, enabling broader adoption and further alleviating the substantial time constraints associated with conventional approaches. When interpreting the data presented here, several limitations inherent to FFPE processing should be considered. For instance, fixation in buffered formalin – standard for FFPE samples – is suboptimal for ultrastructural preservation, particularly due to the absence of glutaraldehyde43. Delays in fixation – common in clinical settings – can result in autolysis44, while prolonged fixation (> 24 h at room temperature) can promote tissue shrinkage45. The omission of lipid-stabilising agents (e.g. osmium tetroxide) and the use of organic solvents (e.g., alcohol, xylene) further contribute to the loss of lipid-rich structures, including cell membranes, lipid droplets, pulmonary surfactant, and myelin46. High temperatures during paraffin embedding exacerbate ultrastructural degradation and shrinkage. Collectively, these factors introduce artefactual changes that complicate subcellular interpretation and demand considerable expertise. Despite these limitations, we have demonstrated that FFPE samples can still provide valuable structural insights using our correlative workflows, even when prepared under suboptimal conditions. Conclusion This study demonstrates that FFPE tissues retain sufficient structural integrity to enable meaningful insights to be gained when examined using CM workflows that combine micro-CT, BFLM and EM. By applying these complimentary modalities, we characterised human tissues across multiple length scales, providing insights into tissue organisation, intercellular contacts and subcellular features. These findings highlight the potential of archived clinical FFPE samples as a valuable resource for multimodal analysis, without the constraints of fresh sample availability. By integrating high-resolution imaging with conventional workflows, this approach opens new avenues for exploring the structural basis of human health and disease. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5 Acknowledgements The authors acknowledge Microscopy Australia (https://ror.org/042mm0k03), enabled by NCRIS. We thank Dr. Matthew Foley for assistance with micro-CT acquisition and registration, and Sydney Microscopy and Microanalysis (https://ror.org/02441px27) at the University of Sydney. We also acknowledge the Charles Perkins Centre Histology Facility (https://ror.org/00zqx9e28). We are grateful to Megan Crabtree (University of Kentucky) and Dr. Yingying Su (University of Sydney) for assistance with the transport of tissue samples between laboratories. Author contributions Z.C. and G.J.S. were responsible for the acquisition of experimental data and segmentation of volumetric datasets. S.D. was responsible for the acquisition of BFLM data. G.J.S. and F.B. were equally responsible for the experimental conception and design, as well as manuscript formulation. K.S.C., A.L., J.D. and D.T. curated tissue samples, as well clinical information associated with tissue samples. G.M. and A.L. contributed to methodology of heart tissue processing. F.B. provided funding and access to resources. All authors have reviewed, edited, and approved the manuscript. Funding F.B. and G.J.S. are grateful to the Australian Research Council for providing support through the Linkage Infrastructure, Equipment and Facilities (LIEF) Scheme (LE110100203). K.C. thanks the National Institutes of Health (HL163977) and American Heart Association (23TPA1074093) for support of this work. Data availability All data supporting the findings of this study are available within the paper and its Supplementary Information. Experimental data is available upon request from the corresponding author G.J.S. (gerald.shami@sydney.edu.au). Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5 Data Availability Statement All data supporting the findings of this study are available within the paper and its Supplementary Information. Experimental data is available upon request from the corresponding author G.J.S. (gerald.shami@sydney.edu.au).