Expansion-enhanced super-resolution radial fluctuations enable nanoscale molecular profiling of pathology specimens

Abstract

Expansion microscopy physically enlarges biological specimens to achieve nanoscale resolution using diffraction-limited microscopy systems¹. However, optimal performance is usually reached using laser-based systems (for example, confocal microscopy), restricting its broad applicability in clinical pathology, as most centres have access only to light-emitting diode (LED)-based widefield systems. As a possible alternative, a computational method for image resolution enhancement, namely, super-resolution radial fluctuations (SRRF)^2,3, has recently been developed. However, this method has not been explored in pathology specimens to date, because on its own, it does not achieve sufficient resolution for routine clinical use. Here, we report expansion-enhanced super-resolution radial fluctuations (ExSRRF), a simple, robust, scalable and accessible workflow that provides a resolution of up to 25 nm using LED-based widefield microscopy. ExSRRF enables molecular profiling of subcellular structures from archival formalin-fixed paraffin-embedded tissues in complex clinical and experimental specimens, including ischaemic, degenerative, neoplastic, genetic and immune-mediated disorders. Furthermore, as examples of its potential application to experimental and clinical pathology, we show that ExSRRF can be used to identify and quantify classical features of endoplasmic reticulum stress in the murine ischaemic kidney and diagnostic ultrastructural features in human kidney biopsies.

Fig. 1. Concept and broad applicability of ExSRRF to clinical and experimental tissues.

a, Schematic of the ExSRRF workflow, including (i) fluorescent molecular labelling, (ii) hydrogel embedding, (iii) tissue expansion, (iv) LED-based WF microscopy with time-stacked acquisition of the Region of Interest (ROI) and (v) computational image processing using SRRF. b, A kidney sample was labelled with pan-protein staining (inverted NHS-ester (iNHS-E)) in combination with a mitochondrial marker (apoptosis-inducing factor 1 (AIF1)). Whole-tissue scanning using LED-based WF systems generates a sample overview with spatial context and molecular information at a cellular level. ExSRRF allows the subcellular identification of hallmark mitochondrial features. c, ExSRRF resolves cytoplasmatic CD42b⁺ granules in perisinusoidal murine bone marrow, using endomucin (EMCN) and alpha-smooth muscle actin (aSMA). d, ExSRRF reveals cell heterogeneity in human glioblastoma. Active cell proliferation was defined using E3 ubiquitin protein ligase (MIB1; green), and astrocyte cytoskeletal filaments were defined using glial fibrillary acidic protein (GFAP; orange). We provide representative images of a non-proliferative astrocyte with normal cytoskeleton (MIB1⁻GFAP⁺; white arrow), a proliferative cell (MIB1⁺GFAP^–; magenta arrow) and an astrocyte with disrupted cytoskeleton (MIB1^–GFAP⁺; cyan arrow); the panels on the right provide zoomed-in views of the panels in the centre. e, Microvascular β-amyloid deposition in human brain tissue from a patient with AD, using vimentin (VMT) as a vascular reference label. f,g, Schematic of murine experimental colitis based on the administration of dextran sulfate sodium (DSS) (f), resulting in a marked loss of body weight (N = 4 mice per group) (g); the dots represent mean and the error bars represent standard deviation. h, ExSRRF revealed a progressive loss of VE-CAD clusters (cell–cell connections) in the submucosal intestinal vasculature during colitis. In the violin plots, the red lines represent medians, and the blue lines represent interquartile ranges. i, Schematic of a murine model of disrupted glomerular basement membrane (GBM) through P3h2 knockout (KO). j, GBM is labelled with laminin (LMN), and the podocyte FPs are labelled with synaptopodin (SNP). ExSRRF-resolved well-organized bilayered LMN distribution is shown under normal conditions, as well as irregular GBM protrusions and focal loss of LMN in KO mice (N = 3 mice per group). The ‘baseline’ refers to WF images acquired before expansion and without SRRF enhancement. The statistics were performed using two-tailed unpaired t-tests with Welch’s correction.

Fig. 2. Multilayered validation of ExSRRF.

a, Nanorulers are equipped with two fluorescently labelled positions at customizable distances (from 25 to 120 nm for this experiment). The resolution range is defined as the smallest distance by which the two spots are resolved without displaying overlapping PSFs. b, Comparison among WF, SRRF, ExM and ExSRRF using nanorulers (N = 3 replicates per nanoruler). c, Quantitative analysis of the percentage of non-overlapping PSFs (N = 3 replicates per nanoruler). Data are reported as mean ± standard error of the mean. d, STED confirmed ExSRRF outputs using a comparative view of the kidney SD labelled with nephrin. e,f, Correlative microscopy using STED and ExSRRF in the same sample (stained with nephrin and aSMA) revealed identical structures within the same regions. g, Lateral movement (drift) during time-stack acquisition resulted in artefact generation. h, Relative frequency of drifts after ExSRRF (N = 48). i, Performance of SRRF was directly affected by drift artefacts, which were reversed after realignment. j, Impact of drift correction measured by deltas in the mean structural similarity index measure (ΔMSSIM) and mean squared error (ΔMSE). Data are reported as mean ± standard error of the mean. k, In a murine model of renal IRI, ExSRRF revealed tight ER in control mice (N = 7) using calreticulin (CLR) as a marker of the ER lumen; the enlarged inset shows a 30 nm distance between the luminal walls. IRI (N = 7) led to dilated ER, which was confirmed by transmission electron microscopy (TEM). l, Automated image analysis (image base: N = 8 for controls and N = 9 for IRI) identified significant reductions in ER density and increases in ER spacing under ischaemic conditions; here the colour coding reflects the local spacing at a pixel level, with 0 being the shortest distance (blue) and 255 being the maximum distance (orange). The violin plots represent the median and quartiles of each distribution. In c, we used one-way ANOVA with Holm–Sidak correction for multiple comparisons. In the other panels, the statistics were performed using two-tailed unpaired t-tests with Welch’s correction.

Fig. 3. Clinical application of ExSRRF in human kidney biopsies.

a, Schematic of the kidney filter in health and disease. Cytoplasmatic projections (FPs) of podocytes and their GBM cover the surface of endothelial cells (ECs). FPs interdigitate and interconnect with their neighbours via the SD. On injury, the broadening of FPs (FP effacement) and therefore the disruption of the corresponding SD results in a leaky kidney filter with corresponding diagnostic features. b, ExSRRF identified both normal and effaced FPs using synaptopodin (SNP) (labelling FPs; orange) and collagen IV (COLIV) (labelling the GBM; cyan). c, ExSRRF resolved the SD (orange) in patients with MCD compared with control patients with immunoglobulin A nephropathy (IgAN) using nephrin. d–g, Automated image analysis provided nanoscale morphometrics (nanometrics), including SD density (N = 27 for IgAN and N = 28 for MCD for d and e) and spacing (N = 27 for IgAN and N = 28 for MCD for f and g); here the colour coding reflects the local spacing at a pixel level, with 0 being the shortest distance (blue) and 255 being the maximum distance (orange). h,i, Signature of MCD was identified using SD density (h) and spacing (i) per image. j, ExSRRF outputs were used to reproduce the perspective obtained using scanning electron micrographs; adjacent podocytes (P1 and P2 in IgAN; P3 and P4 in MCD) were pseudo-coloured with cyan or magenta to reveal the interdigitation pattern between neighbouring FPs. k, FP (N = 54 for IgAN and N = 42 for MCD) and SD (N = 52 for IgAN and N = 41 for MCD) widths were calculated per image. l, Signature of MCD was identified using the FP width per image. m, Average nanometrics per patient confirm our results (SD density, N = 5 per group; FP width, N = 3 per group). The red lines represent the mean values per condition. In this figure, "baseline" refers to WF images acquired before expansion and without SRRF. The violin plots represent the median and quartiles of each distribution. Comparisons of two groups were performed using two-tailed unpaired t-tests with Welch’s correction. The receiver operating characteristic curves were generated, the corresponding standard errors were calculated, and p values were determined from the normal distribution (two-tailed).

Extended Data Fig. 1. Quantification of expansion factor.

(a) Using the 120 nm nanorulers, we measured the distance between intensity peaks after expansion. (b) The mean distance (and SD) between 120 nm nanorulers was calculated, providing an average expansion factor of 3.7x (N = 50). The data is reported as mean ± SD. Then, we used nuclear stainings in tissues (c), the cross-sectional area of individual nuclei was calculated before and after expansion (d), providing an average expansion factor of 3.8x (N = 18).

Extended Data Fig. 2. Tissue mounting.

Exemplary images of a customized 3D-printed imaging chamber for large samples (top), and a commercial 2-well imaging chamber for small specimens (bottom).

Extended Data Fig. 3. 3D-printed imaging chamber.

Here, we provide the details of the design of the three components, including (a) frame, (b) lid, and (c) elastic cushion. (d) All 3D-printed parts are shown. (e) The frame is then fitted with a NexterionTM coverslip (see methodology), elastic cushion, and lid.

Extended Data Fig. 4. ExSRRF achieves better resolution than its individual components in tissue.

The image sequence shows a comparison between the individual components of ExSRRF, namely expansion microscopy (ExM) and SRRF, and widefield (WF) as a negative control and ExSRRF in a known structure (e.g., kidney slit diaphragm labeled with Nephrin).

Extended Data Fig. 5. Adjusting potential drift artefacts during image acquisition.

(a) The sequence shows images before and after drift correction and the impact this has on image quality and artefact generation using NanoJ-SQUIRREL. RSF: Resolution Scaling Function (b) Quantitative definition of the impact of image drift correction on one of the main NanoJ-SQUIRREL outputs (Resolution Scaled Pearson coefficient or RSP). RSE: Relative scaling error. Statistical testing was performed using a two-tailed paired t-test.

Extended Data Fig. 6. Mouse model of ischaemia reperfusion injury (IRI).

(a) Schematic representation of the experimental design. (b) Quantification of tissue damage and apoptosis (N = 7 per group). EM: electron microscopy; FFPE: formalin-fixed paraffin-embedded; FOV: field of view. Violin plots in this figure represent the median and quartiles of each distribution. Statistics were performed using two-tailed unpaired t-tests with Welch’s correction.

Extended Data Fig. 7. ExSRRF allows visualization of structures that cannot be identified using ExM alone.

(a) While ExSRRF resolves murine endoplasmic reticulum (ER) in kidney tubular cells, ExM does not provide sufficient resolution for unambiguous identification. (b) Human kidney slit diaphragm (SD) is fully resolved by STED and ExSRRF but not by ExM alone. (c) ExSRRF shows clear separation of bi-layered basement membrane (Laminin; LMN) within the murine kidney filter (podocyte foot processes labelled with Synaptopodin; SNP), but ExM does not resolve a double layered basement membrane or individual podocyte foot processes. (d) VE-cadherin (VE-CAD) clusters are fully separated by ExSRRF but not by ExM. (e) ExSRRF resolves individual intracellular CD42b⁺ granulae in murine perisinusoidal bone marrow megakaryocytes (MGK), and ExM fails to identify granulae.

Extended Data Fig. 8. ExSRRF facilitates better segmentation compared to ExM alone and STED.

We provide both visual (a) and quantitative (b) confirmation of the resolution gain provided by ExSRRF in comparison to ExM alone. SSIM: structural similarity index measure, ER: Endoplasmic reticulum (N = 17), SD: slit diaphragm (ExSRRF vs ExM: N = 52; and ExSRRF vs ExM vs STED: N = 10), **** represents P < 0.0001. Every grey dot represents one image. Violin plots represent the median and quartiles of each distribution. Statistics were performed using two-tailed unpaired t-tests with Welch’s correction when comparing two groups, and a Welch ANOVA with a Dunnett’s multiple comparisons test.

Extended Data Fig. 9. Electron microscopy images used in clinical diagnosis of minimal change disease.

Overview and enlarged panels show areas of normal foot process morphology (green) and foot process effacement (red). FP(s): foot process(es); GBM: glomerular basement membrane.

Extended Data Fig. 10. Automatic segmentation after ExSRRF in human kidney biopsies.

This sequence describes the image processing steps necessary for automatic segmentation of the kidney slit diaphragm in human kidney biopsies.