Abstract

   Hydrogel-based tissue expansion combined with mass spectrometry (MS)
   offers an emerging spatial proteomics approach. Here, we present a
   filter-aided expansion proteomics (FAXP) strategy for spatial
   proteomics analysis of archived formalin-fixed paraffin-embedded (FFPE)
   specimens. Compared to our previous ProteomEx method, FAXP employed a
   customized tip device to enhance both the stability and throughput of
   sample preparation, thus guaranteeing the reproducibility and
   robustness of the workflow. FAXP achieved a 14.5-fold increase in
   volumetric resolution. It generated over 8 times higher peptide yield
   and a 255% rise in protein identifications while reducing sample
   preparation time by 50%. We also demonstrated the applicability of FAXP
   using human colorectal FFPE tissue samples. Furthermore, for the first
   time, we achieved bona fide single-subcellular proteomics under image
   guidance by integrating FAXP with laser capture microdissection.

   Subject terms: Proteomic analysis, Tumour heterogeneity, Proteomics
     __________________________________________________________________

   Hydrogel-based tissue expansion proteomics represents an emerging
   spatial proteomics approach. Here, the authors develop the filter-aided
   expansion proteomics (FAXP) strategy, enabling proteomic analysis of
   single cells and nuclei in formalin-fixed paraffin-embedded (FFPE)
   tissue sections.

Introduction

   Spatial proteomics can be achieved by affinity-based methods or mass
   spectrometry (MS)-based proteomics^[42]1,[43]2. The latter enables
   comprehensive quantitative detection of proteins without the need for
   affinity reagents, the applicability and reliability of which largely
   rely on their specificity and stability. In addition, the
   antibody-based method requires intact reservation of the sample
   epitopes, posing a great challenge for clinical specimens. Two primary
   MS-based spatial proteomics methodologies have emerged. One concerns MS
   imaging (MSI)^[44]3, utilizing matrix-assisted laser
   desorption-ionization (MALDI), which converts proteins or peptides from
   regions of interest (ROI) into gas-phase ions for subsequent MS
   analysis^[45]4,[46]5. However, compared to the liquid chromatography
   (LC)-based fractionation coupled with electrospray ionization (ESI) MS
   method, MALDI-based MSI characterizes a limited number of peptides and
   proteins. Recently, with rapid improvement of the LC-MS instruments,
   even minute tissue samples dissected from laser capture microdissection
   (LCM) could be effectively analyzed^[47]6–[48]11. The most prominent
   approach, deep visual proteomics (DVP), combines AI-driven image
   analysis, automated LCM, and high-sensitivity mass spectrometry^[49]6.
   Its extension to single-cell analysis (scDVP)^[50]8 addresses the
   complexity of analyzing heterogeneous tissues at single-cell
   resolution. Coupled with other techniques, such as
   microfluidics^[51]12,[52]13, 3D-printed devices^[53]14, or tissue
   clearing^[54]15, LCM has achieved various applications in
   high-throughput and 3D proteome analysis.

   Inspired by expansion microscopy^[55]16–[56]19, tissue expansion
   technology has recently been integrated with MS-based
   proteomics^[57]20,[58]21. Coupled with antibodies, physical tissue
   expansion has recently been integrated with multiplexed ion beam
   imaging and imaging mass cytometry platforms to enable the
   visualization of over 40 protein markers in archived tissue
   sections^[59]22. More recently, tissue expansion has been coupled to
   MALDI-based MS imaging^[60]23.

   In the expansion proteomics (ProteomEx) workflow, paraformaldehyde
   (PFA)-fixed tissue sections are treated with N-succinimidyl acrylate
   (NSA) to modify protein primary amino groups with an acryl group,
   enabling co-polymerization with hydrogel monomers^[61]21. Subsequently,
   the treated tissue undergoes infusion with hydrogel monomers,
   initiating an in situ polymerization reaction to create a
   tissue-hydrogel composite. To achieve isotropic expansion, the
   resulting tissue-hydrogel composite undergoes mechanical homogenization
   in a sodium dodecyl sulfate (SDS)-containing buffer at elevated
   temperatures (95 °C). Following the tissue expansion, the expanded
   sample is manually dissected to isolate a small ROI (below ~200 µm in
   size). These excised samples then undergo in-gel reduction/alkylation
   and tryptic digestion to extract peptides for subsequent LC-MS/MS
   analysis. Crucially, the chemical treatment of biological tissue does
   not introduce any chemical modifications and yields proteomics data
   identical to conventional sample preparation methods^[62]21. Moreover,
   ProteomEx operates without the need for sophisticated or custom
   equipment, relying solely on readily available reagents and accessories
   for sample preparation. However, limitations persist in the original
   ProteomEx protocol, with lateral and volumetric resolution capped at
   ~160 µm and ~0.61 nL, respectively^[63]21. Besides, the sample
   treatment usually takes ~54 h (excluding MS analysis). There are
   multiple manual steps that hinder the parallel processing of multiple
   samples, impeding large-scale analyses. Additionally, the applicability
   of ProteomEx to other types of archived tissue samples and their
   relevant tissue expandability has not been verified.

   In this work, we present a filter-aided expansion proteomics (FAXP)
   approach, inspired by sample preparation methods such as filter-aided
   sample preparation^[64]24, One-Tip^[65]25, and simple and integrated
   spintip-based proteomics technology (SISPROT)^[66]26, which utilize
   filter membranes^[67]24 or tips^[68]25,[69]26 to achieve clean and
   comprehensive proteome profiling. Unlike One-Tip^[70]25 and
   SISPROT^[71]26, FAXP integrates optimized in-gel digestion^[72]27 into
   self-assembled C18 filter tips, enabling high-throughput processing of
   tissue-hydrogel composites without sample size limitations. Compared to
   ProteomEx, FAXP offers enhanced lateral and volumetric resolution by
   2.2- and 14.5-fold, respectively, and reduces processing time by 50%.
   For the first time, FAXP integrated with LCM enables precise isolation
   and analysis of a single mouse hepatocytic nucleus, identifying an
   average of 2368 proteins and showcasing its capability in subcellular
   proteomics. Moreover, FAXP is compatible with archived clinical FFPE
   samples and exhibits robust reproducibility. These advancements arise
   from improvements in tissue expansion, filter-aided in-gel digestion,
   and robotic integration. Using FAXP, we successfully characterize
   proteins associated with colorectal tumor progression from FFPE
   samples.

Results

Optimization of the tissue expansion procedures

   Our initial focus was on expediting several time-consuming steps in the
   ProteomEx protocol, namely protein anchoring, hydrogel embedding,
   homogenization, and Coomassie staining (Fig. [73]1a). These
   optimizations collectively reduced the duration of these steps from
   40.5 to 11.2 h (Fig. [74]1b) without compromising peptide and protein
   identifications (Supplementary Fig. [75]1). We found that autoclaving
   the tissue-hydrogel composite at 105–121 °C, which was intended for
   rapid homogenization, resulted in softer hydrogels post-expansion
   compared to the original treatment conditions. Since softer hydrogels
   were more difficult to handle during the dissection step and could
   potentially introduce larger distortions to the expanded tissue, we
   increased the hydrogel cross-linker ratio by 17.5 times (Supplementary
   Table [76]1) to harden it. This adjustment robustly enhanced the
   sturdiness of the hydrogel, consistent with a previous report^[77]28.

Fig. 1. Comparative analysis of filter-aided expansion proteomics (FAXP) and
ProteomEx.

   [78]Fig. 1
   [79]Open in a new tab

   a Workflow of FAXP: The FAXP process initiates with formalin-fixed
   paraffin-embedded (FFPE) tissue block sectioning and dewaxing, followed
   by chemical treatment and embedding into a hydrogel. Subsequent steps
   involve autoclave-based homogenization, staining, expansion, imaging,
   and manual or laser capture microdissection. Filter-aided in-gel
   digestion occurs under robotic automation for DIA-MS, followed by
   proteomic measurement. (LEF linear expansion factor, which is
   calculated by comparing the expanded sample size to the original
   unexpanded sample size.) b Comparison of processing timelines:
   Timelines for each step in FAXP (red) and ProteomEx (blue) workflows
   are represented. Hands-on time is not delineated due to similarities
   between both workflows for small sample amounts. ProteomEx involves
   reduction and alkylation on individual punches during conventional
   in-gel digestion, while FAXP executes these steps on the entire gel.
   FAXP showcases faster processing times owing to shorter durations for
   various steps compared to ProteomEx. The FAXP data presented are
   derived from 5-μm-thick mouse liver FFPE samples. Source data are
   provided as a Source Data file. Figure 1a, created with BioRender.com,
   released under a Creative Commons Attribution-NonCommercial-NoDerivs
   4.0 International license.

   To assess the impact of the modified hydrogel composition on tissue
   expansion, we conducted a quantitative analysis of the linear expansion
   factor and isotropy of expansion at both the cellular and whole tissue
   slide levels using a mouse liver FFPE slide. Protein anchoring time was
   selected as a variable in our experiments (Fig. [80]2a–e). At the
   cellular level, employing a 1-h anchoring condition yielded linear
   expansion factors (LEF) of the cell nucleus, as visualized via DAPI
   staining, comparable to those measured at the whole expanded tissue
   level (LEF = 4.42 ± 0.31 vs. 4.69 ± 0.17 for individual nucleus vs.
   whole expanded tissue, respectively; Fig. [81]2e), suggesting a
   consistent and uniform expansion pattern extending from the cell
   nucleus to the entire tissue structure. However, prolonged anchoring
   time (2 h) resulted in reduced expansion factors (LEF = 3.10 ± 0.58,
   p = 4.50e-6) at the cellular level compared to those measured at the
   whole expanded tissue level (Fig. [82]2e). Subsequently, we performed
   the anchoring for 1 h for all subsequent experiments. We then assessed
   the root-mean-square (RMS) measurement error in feature-length
   measurements after non-rigid registration of pre- and post-expansion
   tissue across length scales of up to 1500 µm, using 10-µm-thick mouse
   liver FFPE slides (n = 4). Our analysis revealed that the RMS errors
   were ~5.5% of the measurement distance, slightly improved compared to
   that of the original protocol (Fig. [83]2f, g). Comparable RMS errors
   were also observed for tissues with rich extracellular matrix, such as
   the kidney (n = 3; Supplementary Fig. [84]2).

Fig. 2. Optimization of gel-making procedures and comparative proteomic
analysis.

   [85]Fig. 2
   [86]Open in a new tab

   DAPI staining of a a mouse liver formalin-fixed paraffin-embedded
   (FFPE) slide after a 1-h anchoring before expansion (n = 1), b
   post-expansion tissue-hydrogel composite following 1-h anchoring
   (n = 1), c a mouse liver FFPE slide after a 2-h anchoring before
   expansion (n = 1), and d post-expansion tissue-hydrogel composite
   following 2-h anchoring (n = 1). Magnified views highlighting nuclei
   with their measured diameters are included (a–d). Scale bars are
   provided within each image. e Comparison of macro-level colorimetric
   staining (n = 5 biological replicates, representing different FFPE
   slides) and micro-level DAPI nuclear staining under 1- and 2-h
   anchoring conditions (n = 10 biological replicates, representing
   different nuclear diameters after expansion compared to before
   expansion). P values are estimated by Welch’s t-test (two-sided). Data
   were presented as mean values ± standard deviation. f Bright-field
   images of a mouse liver FFPE slide before expansion (top) and after
   expansion (bottom), stained with Coomassie Brilliant Blue. The linear
   expansion factor (LEF) was calculated by comparing the dimensions of
   the expanded sample to the unexpanded sample (n = 4). g Comparing
   root-mean-square (RMS) measurement length errors pre- and
   post-expansion in liver slide images (n = 4); an example shown in (f).
   The blue line represents the mean, and the shaded area indicates the
   SD. h Study design for comparative proteomic analysis involving four
   reduction and alkylation conditions compared to in-gel digestion (n = 4
   biological replicates), used under ProteomEx in red^[87]21, as the
   control group. In this study, mouse liver FFPE slides from three
   independent biological samples undergo conditions: 10 mM dithiothreitol
   (DTT)/55 mM iodoacetamide (IAA) in blue, 20 mM Tris (2-carboxyethyl)
   phosphine (TCEP)/55 mM IAA in green, 45 mM TCEP/100 mM IAA in purple,
   and 20 mM DTT/55 mM IAA in orange. One-quarter of each gel is used per
   condition (n = 3 biological replicates). After in-gel digestion, two
   5-mm punches from each condition were pooled, and MS data were
   collected using 100 ng per injection with two injections per condition.
   i Comparative proteomic analysis for the reduction and alkylation
   conditions outlined in Fig. 2h, including identified peptides and
   proteins, and carbamidomethyl modification site percentages. P values
   are estimated by Welch’s t-test (two-sided). For all boxplots, center
   lines show the medians; box limits indicate the 25th and 75th
   percentiles; whiskers extend to the most extreme data points within 1.5
   times the interquartile range from the 25th and 75th percentiles (IQR).
   Data were presented as dots and points beyond whiskers are outliers.
   Source data are provided as a Source Data file. Fig. 2h, partially
   created with BioRender.com, released under a Creative Commons
   Attribution-NonCommercial-NoDerivs 4.0 International license.

   Subsequently, we expedited the ProteomEx timeline and optimized peptide
   recovery from microdissected tissue-hydrogel samples by targeting the
   most laborious and unreliable steps associated with multiple manual
   procedures. We started with testing whether the reduction/alkylation
   steps, standard treatment to increase the yield of cysteine-containing
   peptides^[88]29, could be performed on the whole tissue slice either
   before hydrogel embedding or after tissue-hydrogel composite
   homogenization using mouse brain slides (n = 3; Supplementary
   Fig. [89]3 and Supplementary Table [90]2). Compared to the original
   in-gel digestion protocol, reduction and alkylation performed after
   homogenization step on whole tissue section resulted in slightly
   increased peptide and protein identifications (by 22.8 and 13.0%,
   respectively) and improved distribution of peptide lengths while
   yielding similar carbamidomethyl modification rates and missed cleavage
   rates (Supplementary Fig. [91]3). These results show that performing
   reduction and alkylation on whole tissue-hydrogel composites after
   homogenization simplifies the protocol by reducing manual steps during
   in-gel digestion.

   Then, we tested four reduction/alkylation conditions—10 mM
   dithiothreitol (DTT)/55 mM iodoacetamide (IAA), 20 mM Tris
   (2-carboxyethyl) phosphine (TCEP)/55 mM IAA, 45 mM TCEP/100 mM IAA, and
   20 mM DTT/55 mM IAA—on entire tissue-hydrogel composites after
   homogenization using 10-µm-thick mouse liver FFPE slides (n = 3),
   comparing them to the in-gel digestion used in ProteomEx with one
   10-µm-thick mouse liver FFPE slide (Fig. [92]2h). For the test groups,
   each condition was applied to a quarter of the hydrogel (n = 3). We
   found that 20 mM DTT/55 mM IAA resulted in more stable proteomic
   identification and better alkylation performance (Fig. [93]2i).

   Next, we optimized in-gel proteolytic digestion conditions using mouse
   brain slides (n = 3; Supplementary Table [94]2) utilizing (1) one round
   of trypsinization (3 h); (2) two rounds of trypsinization (4 h + 12 h);
   (3) combination of LysC and trypsinization (4 h + 12 h). All conditions
   yielded a comparable number of identified peptides and proteins
   (Supplementary Fig. [95]4). However, the single-round trypsin digestion
   exhibited slightly higher missed cleavage rates compared to
   “LysC + Trypsin” digestion (Supplementary Fig. [96]4c). In conclusion,
   in-gel proteolytic digestion can be carried out in a wide range of
   conditions while extended time decreased missed cleavage rates.

Adoption of filter-aided in-gel digestion

   The ProteomEx method originally posed challenges in achieving high
   lateral and volumetric resolutions, limited to around 160 µm and
   0.61 nL, respectively. This was mainly due to the difficulty in
   handling hydrogel samples under 1 mm in diameter. Specifically, these
   smaller hydrogel pieces tend to shrink through successive buffer
   exchanges, leading to sample loss during pipetting. To address this, we
   proposed immobilizing the tiny hydrogel pieces onto a membrane placed
   within a pipette tip. We tested the feasibility of this filter-aided
   in-gel digestion method, also known as “in-tip digestion”, compared to
   the original in-gel digestion using 10-µm-thick mouse liver FFPE slides
   (n = 6 punches for each condition; Fig. [97]3a). Utilizing timsTOF Pro
   data-dependent acquisition (DDA) MS (50-min effective LC gradient)
   analysis for peptides generated from a 1.55 nL tissue-hydrogel
   composite, in-tip digestion yielded a remarkable increase in peptide
   and protein detections (by 96.6 and 31.1%, respectively; Fig. [98]3b).
   Additionally, this method enhanced protein digestion efficiency, with a
   missed cleavage rate of approximately 17.0%, lower than the 23.8%
   observed in in-gel digestion (Fig. [99]3c and Supplementary
   Fig. [100]5a). We also observed an unbiased identification across a
   series of physicochemical properties, like peptide hydrophobicity
   (Fig. [101]3d and Supplementary Fig. [102]5b) and increased detection
   of longer peptides (Fig. [103]3e), suggesting improved peptide
   coverage. Moreover, 74.9% of peptides and 97.5% of proteins identified
   through in-gel digestion were also detected via in-tip digestion. The
   in-tip method also identified 56.4% more peptides and 20.5% more
   proteins (Fig. [104]3f).

Fig. 3. Comparative proteomic analysis of sample preparation methods applied
to mouse liver FFPE slides.

   [105]Fig. 3
   [106]Open in a new tab

   (a) Comparative proteomic study between conventional in-gel digestion
   (in-gel; blue) and filter-aided in-gel digestion (in-tip; red) using
   2000-µm punches (n = 6 biological replicates) from mouse liver
   formalin-fixed paraffin-embedded (FFPE) slides (LEF linear expansion
   factor, which is calculated by comparing the expanded sample size to
   the original unexpanded sample size). b Comparison of identified
   peptides and proteins between in-gel and in-tip digestions (n = 6
   biological replicates outlined in Fig. 3a). P values are estimated by
   Welch’s t-test (two-sided). c Percentage of identified peptides with
   zero missed cleavage sites using in-gel and in-tip digestions (n = 6
   biological replicates outlined in Fig. 3a). P values are estimated by
   Welch’s t-test (two-sided). For both Fig. 3b, c, all boxplots show
   center lines representing the medians; box limits indicate the 25th and
   75th percentiles, and whiskers extend to the most extreme data points
   within 1.5 times the IQR. Data were presented as dots, with points
   beyond whiskers representing outliers. d Density distribution of
   isoelectric points for identified peptides in both digestion methods. e
   Distribution of peptide lengths as shown in (b) in in-gel and in-tip
   digestions. f Venn diagram illustrating the overlap of identified
   peptides and proteins between in-gel and in-tip digestions. g
   Coefficient of variation (CV) showing quantified peptide and protein
   abundance variations between in-gel and in-tip digestions. h Pearson’s
   correlation analysis for protein quantification between in-gel and
   in-tip digestions. i Log2-transformed mean protein intensities ranked
   by abundance, highlighting liver-specific proteins identified in both
   digestion methods. j Enriched protein subcellular locations and types
   presented in chord diagrams. Source data are provided as a Source Data
   file. Figure 3a, partially created with BioRender.com, released under a
   Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International
   license.

   The in-tip digestion approach also led to more reliable quantification
   outcomes, evidenced by considerable improvements in the coefficient of
   variance (below 0.2 for both peptides and proteins; Fig. [107]3g) and
   Pearson’s correlation (close to 1.0 for both peptide and protein
   levels; Fig. [108]3h and Supplementary Fig. [109]5c). Furthermore, the
   in-tip digestion method identified 26.7% additional liver-specific
   proteins than that by the in-gel digestion method (Fig. [110]3i), as
   curated by the UniProt database. Pathway enrichment analysis of the
   proteins identified by both methods showed a similar distribution of
   subcellular localizations and protein types (Fig. [111]3j).
   Collectively, these findings underscore that the filter-aided in-gel
   digestion not only facilitates sample handling but also substantially
   enhances proteome coverage and reproducibility.

FAXP enhances the throughput, resolution, and sensitivity of spatial
proteomics

   Combining filter-aided in-tip digestion with an accelerated tissue
   expansion workflow not only enhanced proteome coverage but also
   increased sample preparation throughput, spatial resolution, and
   sensitivity. To assess the impact of in-tip digestion on the lateral
   and volumetric resolution of ProteomEx, we investigated the
   volume-dependent limit of tissue micro-sampling using the optimized
   workflow, with peptide recovery performed using either in-gel or in-tip
   protocols using 10-µm-thick mouse liver FFPE consecutive slides (n = 2;
   Fig. [112]4a). Manual dissection of tissue microsamples with volumes
   ranging from 0.042 to 3.054 nL (equivalent to lateral resolutions from
   73 to 624 µm; n = 7 punches each size per protocol) revealed that
   in-tip digestion, under both DDA-MS mode and data-independent
   acquisition (DIA) MS mode, identified a range of 5196 to 28,411
   peptides and 19,499 to 48,330 peptides, respectively (Fig. [113]4b).

Fig. 4. Detection range and limitation.

   [114]Fig. 4
   [115]Open in a new tab

   a Study design for comparative proteomic analysis across various tissue
   volumes using in-gel (blue) and in-tip (red) digestions. Peptides from
   different diameters are analyzed using data-dependent acquisition mass
   spectrometry (DDA-MS; n = 3 technical replicates) and data-independent
   acquisition mass spectrometry (DIA-MS; n = 3 technical replicates). b
   Number of identified peptides and proteins for varied tissue volumes
   using in-gel and in-tip digestions as shown in (a), including in-gel
   DDA-MS (light blue), in-gel DIA-MS (blue), in-tip DDA-MS (orange), and
   in-tip DIA-MS (red). n = 3 technical replicates for each method and
   tissue volume. Data were presented as mean values ± standard deviation.
   c Asialoglycoprotein receptor 1 (ASGR1; yellow) and nuclear (blue)
   staining image of a mouse liver formalin-fixed paraffin-embedded (FFPE)
   slice before gel-making (n = 1). d ASGR1 and nuclear staining image of
   a mouse liver FFPE slice after gel-making (n = 1). e ASGR1 and nuclear
   staining image of a mouse liver FFPE slice after drying onto laser
   capture microdissection frame slide (n = 1). Scale bars are provided. f
   Comparison of proteins between single nucleus (n = 5 biological
   replicates; blue) and single mononuclear hepatocyte (n = 6 biological
   replicates; magenta). P value is estimated by Welch’s t-test
   (two-sided). For all boxplots, center lines show the medians; box
   limits indicate the 25th and 75th percentiles; whiskers extend to the
   most extreme data points within 1.5 times the IQR. Data were presented
   as dots and points beyond whiskers are outliers. g Heatmap showing
   Pearson’s correlations for protein quantification between each sample
   pair from a single mouse nucleus and mononuclear hepatocyte shown in
   (f). The color bar represents Pearson’s correlation coefficient. h
   Ridge plot displaying protein expression distribution in enrichment
   analysis across major cellular compartments (left) and Reactome
   pathways (right) between single mononuclear hepatocyte and nucleus from
   (f). Selected items with FDR <0.05. Enrichment analysis was performed
   using fast gene set enrichment analysis with a Benjamini–Hochberg
   adjustment for multiple comparisons. Source data are provided as a
   Source Data file. Figure 4a, partially created with BioRender.com,
   released under a Creative Commons Attribution-NonCommercial-NoDerivs
   4.0 International license.

   In-tip digestion consistently outperformed in-gel digestion in terms of
   peptide and protein identification in both DDA and DIA modes,
   especially for tissue samples with a volume less than 3.054 nL.
   Notably, in-tip digestion reached a peak for tissue volumes ranging
   between 0.763 and 1.357 nL, beyond which no further increase was
   observed. On the other hand, in-gel digestion encountered technical
   constraints, limiting its ability to process tissue volumes below
   0.339 nL. For a tissue volume of 0.339 nL, in-tip digestion
   outperformed in-gel digestion by over 800 and 250% in peptide and
   protein identification, resulting in the identification of 2204 and
   4844 proteins in DDA and DIA mode, respectively.

   Remarkably, in-tip digestion demonstrated compatibility with
   tissue-hydrogel composites as small as 0.042 nL in volume, achieving a
   lateral resolution of 73 µm on 10 µm-thick sections, equivalent to
   eight murine hepatocytes^[116]30. These findings illustrate that in-tip
   digestion improves both lateral and volumetric resolution by 2.85- and
   8.07-fold, respectively, while providing 8.07/2.55-fold higher
   peptide/protein identification compared to the ProteomEx protocol with
   in-gel digestion under its lower limit tissue volume using DIA mode.

Coupling FAXP with LCM to study single nucleus proteome

   We then tried to integrate FAXP and LCM to analyze the proteome of a
   single nucleus, which can hardly be isolated and collected for
   downstream proteomics analysis using LCM directly from FFPE tissues
   (Supplementary Fig. [117]6a). The diameter of a typical mouse liver
   hepatocyte nucleus is about 8 µm (Fig. [118]2a). After expansion with
   FAXP, the diameter increased to about 35 µm, which could be effectively
   isolated for proteomic analysis (Supplementary Fig. [119]6b). Without
   expansion, LCM can produce thick laser-burning curves around the
   nucleus, potentially causing damage or contamination from the
   surrounding cytoplasm. This also makes it challenging to isolate the
   nucleus intact for downstream proteomic analysis. In contrast, with
   FAXP, the nucleus expands linearly by about four times, resulting in a
   volume increase of approximately 64 times. This expansion makes
   isolation by LCM significantly easier. Additionally, when surrounded by
   hydrogel, the weight increases, simplifying further processing for
   proteomic analysis.

   We benchmarked this strategy to study mouse hepatic nuclei and cells
   using one 10-µm-thick mouse liver FFPE slide (Fig. [120]4c, d). After
   attaching the tissue-hydrogel composite onto the LCM frame slide
   (Fig. [121]4e), nuclei were successfully isolated, and hepatocytes were
   identified using asialoglycoprotein receptor 1 (ASGR1), a hepatocytic
   plasma membrane marker^[122]31. The proteome of these isolated nuclei
   and hepatocytes was then analyzed using an Orbitrap Astral mass
   spectrometer. Following quality control, which involved removing one
   low-correlation single nucleus sample and two outliers from the
   mononuclear cell group (Supplementary Fig. [123]7), an average of 2368
   and 3312 proteins were identified from single nuclei and single
   mononuclear cells, respectively (Fig. [124]4f). The proteome
   correlations were strong, exceeding 0.86 among nuclei and over 0.92
   among mononuclear hepatocytes (Fig. [125]4g). Protein expression
   analysis showed that nuclear-associated terms like nucleoplasm,
   chromosome, and nucleolus were predominantly expressed in nuclei,
   whereas terms related to mitochondrion, endoplasmic reticulum, Golgi
   apparatus and peroxisome were more prominent in cells (Fig. [126]4h).
   Pathway analysis revealed that the citric acid (TCA) cycle and
   metabolism were highly activated in cells, while pathways like gene
   expression and mRNA splicing were more active in nuclei (Fig. [127]4h).
   We used ASGR1 in conjunction with nuclear staining to differentiate
   between mononuclear and binuclear hepatocytes. The correlation between
   these cell types exceeded 0.9 (Supplementary Fig. [128]8a). Histone
   expression levels were consistent with the previous reports^[129]8
   (Supplementary Fig. [130]8b). Additionally, we identified 129 proteins
   exclusive to binucleated cells and 37 proteins unique to mononucleated
   cells (Supplementary Fig. [131]8c). These findings demonstrate FAXP’s
   effectiveness in accurately analyzing single cells and nuclei.

Application of FAXP to understand intra-tissue variability of colorectal
tumors

   Next, we tried to apply the FAXP method to understand the intra-tissue
   variability of colorectal cancer (CRC). There were three patients, with
   the first having three consecutive FFPE slides, while each of the other
   two patients had one slide (Fig. [132]5a). Four regions, including
   normal (N), low-grade dysplasia (L), high-grade dysplasia (H), and
   carcinoma (C), were delineated from these samples based on hematoxylin
   and eosin (H&E) staining by an expert pathologist. Firstly, we
   validated the efficacy of the FAXP approach in achieving isotropic
   expansion of archived clinical samples (Fig. [133]5b). Of note, the RMS
   error for CRC tissue was found to be ~6% of the measurement distance
   (Fig. [134]5c), falling within the expected range based on mouse tissue
   expansion. Next, we manually performed a visual-guided microdissection
   of the CRC tissue-hydrogel composites from three distinct patients
   (referred to as P1, P2, and P3) and collected samples as outlined in
   Fig. [135]5a and Supplementary Table [136]3). A total of 131
   microdissected tissue samples were processed using FAXP, leading to the
   identification of 6751 proteins. The systematic data analysis revealed
   that the number of identified proteins progressively increased with the
   severity of the cancer stage. Specifically, the numbers of protein
   identifications in L, H, and C were 6.67, 17.36, and 19.74% higher than
   that of the N samples based on median values, respectively
   (Fig. [137]5d), consistent with the previous reports^[138]32. To assess
   variations, we classified the identified proteins by disease region, MS
   batch, patient, and slide (Fig. [139]5e). Our data showed consistent
   protein identification across batches, patients, and slides
   (Fig. [140]5e and Supplementary Fig. [141]9), demonstrating the
   reproducibility and reliability of the FAXP workflow. Principal
   variance components analysis (PVCA)^[142]33 indicated that the primary
   source of variability was due to different regions (48.94%), compared
   to patients (8.89%) and slides (0.7%), underscoring that regional
   differences are the main contributors to variability (Supplementary
   Fig. [143]9f).

Fig. 5. Application on colorectal cancer patient samples.

   [144]Fig. 5
   [145]Open in a new tab

   a Study design overview: Analysis of colorectal cancer (CRC) samples
   from three patients (P1, P2, P3), including sequential slides from the
   first patient, to assess protein expression differences. Disease
   progression subtypes are labeled as follows: N normal (blue), L
   low-grade dysplasia (yellow), H high-grade dysplasia (orange), C
   carcinoma (magenta). Haematoxylin and Eosin (H&E) staining was
   performed on formalin-fixed paraffin-embedded (FFPE) slides. Mass
   spectrometry (MS) analysis was conducted on 1500-µm punches, followed
   by timsTOF Pro data-independent acquisition mass spectrometry (DIA-MS)
   and data analysis. Scale bars = 1000 µm. b Bright-field images of human
   CRC slides before (top) and after (bottom) expansion. Tissue-hydrogel
   composites were stained with Coomassie Brilliant Blue (n = 3). c
   Quantitative comparison of root-mean-square (RMS) measurement length
   error between original and expanded images (n = 3); an example is shown
   in (b). The blue line represents the mean, and the shaded area
   indicates the standard deviation. d Comparison of identified proteins
   across different subtypes. n = 26 biological replicates for N, n = 31
   biological replicates for L, n = 26 biological replicates for H, and
   n = 48 biological replicates for C. These biological replicates
   represent different regions from multiple patients (as detailed in
   Supplementary Table [146]3). P values are estimated using Welch’s
   t-test (two-sided) with a Benjamini–Hochberg adjustment. For all
   boxplots, center lines show the medians; box limits indicate the 25th
   and 75th percentiles; whiskers extend to the most extreme data points
   within 1.5 times the IQR. Data were presented as dots and points beyond
   whiskers are outliers. e Heatmap displaying protein abundance (z-score)
   with annotations for subtype regions, batch numbers (b1–b6), patient
   IDs (P1–P3), and slide numbers (1–3). Colors corresponding to each
   variable are defined in the legend. Source data are provided as a
   Source Data file. Figure 5a, partially created with BioRender.com,
   released under a Creative Commons Attribution-NonCommercial-NoDerivs
   4.0 International license.

   We then compared the proteomes between different pairs of tissue
   regions (Fig. [147]6a). The UMAP plot effectively distinguished all CRC
   samples based on malignancy, particularly differentiating between
   normal and diseased regions (Fig. [148]6b). The number of
   differentially expressed proteins (DEPs) increased with disease
   progression, reflecting greater divergence between disease stages
   (Fig. [149]6c). Advanced-stage tumors exhibited a higher number of
   DEPs. We further employed Ingenuity Pathway Analysis (IPA) to explore
   DEPs across different malignancy stages compared to normal regions.
   This analysis revealed potential clinical biomarkers (Fig. [150]6d).
   Additionally, using the ADCdb^[151]34, we identified potential
   antibody-drug conjugate (ADC) targets across different malignancy
   stages (Fig. [152]6e). Targets were generally present in all malignant
   stages compared to normal, with CEACAM5, PTGFRN, and ITGA3 consistently
   identified in every comparison. Besides, pronounced alterations in DEP
   expression were observed in advanced CRC stages compared to less severe
   stages (Supplementary Fig. [153]10a). We identified 19 DEPs associated
   with disease progression by comparing different malignant stages
   (Supplementary Fig. [154]10b). Notably, CEACAM5, a potential ADC target
   (Fig. [155]6e) and an oncogene associated with tumor progression and
   resistance to anoikis in colorectal carcinoma cells^[156]35, exhibited
   increasing expression levels across disease stages (Supplementary
   Fig. [157]10b). This increase is consistent with CEACAM5’s role in
   adhesion and invasion during CRC progression^[158]36. These findings
   underscore the reliability of the FAXP workflow and its capability to
   analyze variability in malignant changes within archived FFPE clinical
   samples.

Fig. 6. FAXP applied to archived colorectal carcinoma samples.

   [159]Fig. 6
   [160]Open in a new tab

   a Heatmap showing the 335 significantly differentially expressed
   proteins (DEPs) across different stages of malignant changes and normal
   regions. Regions are color-coded as N (Normal, blue), L (Low-grade
   dysplasia, yellow), H (High-grade dysplasia, orange), and C (Carcinoma,
   magenta), with patient IDs annotated as P1, P2, and P3. Protein
   abundance is displayed as z-scores. b Uniform manifold approximation
   and projection (UMAP) visualization of all regions by the DEPs shown in
   (a). c Differential protein expression analysis showing up- and
   down-regulated proteins (blue: downregulated, red: upregulated) across
   different stages of malignant changes compared to normal regions, with
   the line chart showing the number of DEPs. Statistical analysis was
   performed using Welch’s t-test (two-sided), with a Benjamini–Hochberg
   adjustment for multiple comparisons. d Biomarker distribution for each
   stage of malignant change, with color-coded bars: C vs N (magenta), H
   vs N (orange), and L vs N (yellow). e Venn diagram comparing
   antibody-drug conjugate (ADC) targets across different stages of
   malignant changes compared to normal regions. Source data are provided
   as a Source Data file.

Discussion

   Compared to our previously published expansion proteomics method called
   ProteomEx^[161]21, this FAXP approach has the following advantages
   (Fig. [162]1a). Firstly, we enhanced sample handling by using
   customized or commercial glass slides for FFPE samples, allowing
   greater flexibility in placement. FAXP accommodates any tissue size
   that fits on a standard glass slide (25 mm × 75 mm) and can process
   thin FFPE sections below 10 microns as well as thicker sections up to
   120 microns via vibratome sectioning. It is also compatible with
   cryopreserved tissues, simplifying the process by removing the need for
   dewaxing and rehydration. We then introduced a 20% SDS denature buffer
   and autoclave-based tissue homogenization to expedite processing.
   Subsequently, we enhanced sample preparation reproducibility by
   conducting reduction/alkylation on the entire tissue sample embedded in
   the hydrogel. We also streamlined staining techniques after tissue
   expansion, ensuring faster and simpler procedures. Compared to
   ProteomEx, FAXP achieves a 2.2-fold improvement in lateral resolution
   (160 µm for ProteomEx vs. 73 µm for FAXP; Fig. [163]4b). Because FAXP
   can process 10-µm sections, while ProteomEx processes 30-µm sections,
   there is a 14.5-fold increase in volumetric resolution (0.61 nL for
   ProteomEx vs. 0.042 nL for FAXP; Fig. [164]4b). This enhancement in
   volumetric resolution demonstrates that FAXP can achieve high spatial
   resolution manually, without requiring high-end instruments like LCM,
   making it more accessible for laboratories. We further enabled LCM to
   analyze subcellular organelles. To surpass the 20 µm working limit of
   LCM^[165]37, our study, for the first time, achieved bona fide
   single-nucleus proteomics under image guidance by integrating FAXP with
   LCM. We boosted sample throughput and robustness by employing
   filter-aided in-gel digestion. Moreover, we ensured comprehensive
   protein identification using DIA-MS with timsTOF Pro and Orbitrap
   Astral. Finally, we integrated robotics into our workflow. FAXP allows
   preparation of microdissected samples in a batch of up to 96, cutting
   total processing time by as much as 53.5% compared to ProteomEx
   (Fig. [166]1b). While 6% of gel samples are lost during peptide
   extraction using ProteomEx^[167]21, FAXP, like other filter-aided
   sample preparation^[168]24–[169]26, not only eliminates the loss of gel
   samples and simplifies sample handling but also notably improves
   proteome coverage and reproducibility. To enhance reproducibility,
   Supplementary Note [170]1 provides detailed troubleshooting and
   critical steps for the FAXP workflow.

   Recent advancements in spatial omics have significantly propelled
   spatial proteomics forward^[171]1,[172]2, with techniques like DVP and
   scDVP that integrate AI-assisted sample region selection, LCM, and
   advanced mass spectrometry for robust analysis of tissue samples down
   to single-cell resolution^[173]6,[174]8. However, the spatial
   resolution of these approaches can be constrained by the limitations of
   LCM, particularly in maintaining the intactness of subcellular
   structures. In contrast, FAXP offers high spatial resolution through
   two distinct approaches. Firstly, it can achieve relatively high
   spatial resolution (down to 0.042 nL, equivalent to eight murine
   hepatocytes^[175]30) manually, without the need for high-end
   instruments, making it more accessible for a broader range of
   laboratories. As this method does not rely on LCM, the punch needle
   size can be further reduced, theoretically enabling even finer
   resolution. Secondly, when combined with LCM, FAXP further enhances
   spatial resolution beyond the capabilities of LCM alone, enabling
   precise analysis of individual subcellular organelles, such as a single
   nucleus, which can be challenging with conventional LCM techniques.
   FAXP also reduces sampling difficulties and improves accuracy compared
   to LCM alone, while minimizing laser-induced damage to tissue samples.
   Importantly, FAXP is not competing with DVP, which uses LCM to dissect
   cell shapes. Instead, FAXP is directly compatible with DVP. Cells or
   tissues could be expanded with FAXP and then followed by the DVP
   workflow, allowing for enhanced spatial resolution and sample
   integrity.

   While the current FAXP workflow is not universally applicable to all
   tissue types due to its relatively complex preparation process, with
   operational challenges, potential contamination risks, and increased
   time costs, it does demonstrate significant adaptability across a
   diverse range of samples. For instance, tissues rich in extracellular
   matrix, such as CRC samples^[176]38,[177]39, require more rigorous
   homogenization, involving higher SDS concentrations, elevated
   temperatures, and extended processing times. These tailored
   optimizations, detailed in Supplementary Fig. [178]11a and
   Supplementary Table [179]1, enhance FAXP’s efficacy across various
   samples, underscoring its versatility for clinical applications.
   Although the FAXP method may be complex, its ability to be adapted for
   different tissue types suggests its potential for broad applicability
   in the future.

   In exploring CRC applications, substantial differences arise among
   different stages of malignant changes. This emphasizes the criticality
   of analyzing specific CRC subtypes to unravel deeper molecular insights
   into the disease^[180]40,[181]41. Particularly noteworthy is FAXP’s
   ability to discern variability in disease progression within the
   slides. This feature is crucial in clinical research, allowing for
   precise sample analysis and enhancing our knowledge of diseases and
   treatment options.

   In summary, FAXP presents a notable advancement in expansion proteomics
   by offering substantially accelerated sample processing, adaptability
   to diverse samples, improved spatial resolution, and enhanced
   sensitivity and reproducibility.

   The FAXP workflow presented here has several constraints. The
   complexity of the protocol introduces the risk of potential
   contamination, particularly with keratin. To successfully execute this
   protocol, it is essential to maintain a clean experimental environment
   and use clean tubes, tips, and pipettes. The workflow supports a linear
   expansion factor of about five times to prevent FFPE tissue cracking,
   which is generally sufficient. However, while FAXP combined with LCM
   shows promise for analyzing larger subcellular structures like the
   nucleus, further optimization of expansion factors is needed for
   smaller organelles such as vacuoles, peroxisomes, and mitochondria.
   Additionally, reliance on Coomassie Brilliant Blue for staining CRC
   FFPE tissue-hydrogel composites limits the robustness and versatility
   required for accurate clinical assessments. As a demonstration of this
   technology, we analyzed samples from several CRC patients. Clinical
   interpretation from the CRC data should be further validated in
   independent cohort samples in the future. Finally, the current robotics
   system requires partial manual assistance, underscoring the need for
   future efforts to achieve full automation of the FAXP protocol.
   Improved automation will not only reduce operational errors but also
   minimize potential contamination.

Methods

   This study complies with all relevant regulations for human study
   participants and was conducted in accordance with the Declaration of
   Helsinki. The collection and use of human tissue were approved by the
   Institutional Ethics Committee of the Second Affiliated Hospital of
   Zhejiang University School of Medicine and the Ethics Committee of
   Westlake University.

Preparation of FFPE mouse organ tissue samples

   All animal procedures adhered to Westlake University Animal Care
   Guidelines and received approval from the Institutional Animal Care and
   Use Committee (IACUC) under protocol #22-083-GTN. Male and female mice
   were used in separate benchmarking experiments, without comparing
   sexes, so results are not disaggregated by sex. Sample size estimation
   for animal studies was not based on statistical methods. As in previous
   work^[182]21, all mice were housed in controlled barrier facilities,
   where the macroenvironmental temperature ranged from 20 to 26 °C and
   humidity from 40 to 70%. Food and water were available ad libitum, and
   the mice were kept on a 12-hour light/dark cycle. The housing
   conditions were carefully monitored and maintained.

   Four male C57BL/6J wild-type mice (Mus musculus, four months old) and
   one female C57BL/6J wild-type mouse (Mus musculus, 2 months old) were
   obtained from the Laboratory Animal Resources Center of Westlake
   University. In the benchmarking studies, all experiments used male mice
   except for the single-nucleus proteomics study, which used the female
   mouse. All mice were deeply anesthetized with 1% sodium pentobarbital.
   Subsequently, transcardial perfusion was conducted using 1×
   phosphate-buffered saline (PBS) and 4% paraformaldehyde (PFA) obtained
   from Electron Microscopy Sciences, USA. The liver and kidney were
   dissected and fixed in 4% PFA for 6 h at 4 °C.

   Fixed tissues underwent dehydration through sequential ethanol
   concentrations (75, 95, and 100%) for 30 min each. Following
   dehydration, the tissues were infiltrated with paraffin wax at 60 °C.
   The embedded tissues were then sliced into either 5-µm or 10-µm
   thickness using a rotary microtome (Leica RM2255, Germany) according to
   standard protocols and mounted onto glass slides.

Collection of FFPE samples from colorectal cancer patients

   Archival FFPE colorectal tissue samples were from three CRC patients
   diagnosed at the Second Affiliated Hospital of Zhejiang University
   School of Medicine (Hangzhou, China). These samples were sliced to an
   8-μm thickness using a rotary microtome (Leica RM2255, Germany). Tumor
   grading was conducted following the tumor, node, and metastasis (TNM)
   classification system for CRC. Tissue regions from both male and female
   patients were randomly selected from FFPE slides, with no analysis of
   sex or gender differences.

   Prior to surgeries, informed consent was obtained from all patients.
   The human tissue collection adhered to the approved institutional
   review board protocol at the hospital. The collection and usage of
   human tissue samples were approved by the Institutional Ethics
   Committee of the Second Affiliated Hospital of Zhejiang University
   School of Medicine (No. 2020-322) and by the ethics committee of
   Westlake University (Permission number: 20220913GTN001).

Update of tissue expansion and staining workflow

   Supplementary Table [183]4 outlines reagents’ compositions and storage
   conditions for tissue expansion, staining, and sample preparation.
   Detailed supplier and catalog information for FAXP-related chemicals,
   reagents, and accessories are provided in Supplementary Table [184]5.

   Deparaffinization of paraffin-embedded sections involved two 10-min
   heptane treatments. Subsequent rehydration steps happen in 100, 90, and
   75% ethanol for 5 min each followed by rehydration in ddH[2]O for
   5 min. For CRC tissue sections, H&E staining were performed following
   the guidelines from the manufacturer (Shanghai Yuanye Biotechnology,
   China).

   The tissue expansion procedure, adapted from ref. ^[185]21, included
   PBS rinsing, anchoring for 1 or 2 h at 25 °C, and subsequent washing of
   the samples three times with the anchoring termination buffer for 5 min
   each. The tissue anchored with NSA underwent infusion with an activated
   monomer solution in a gelation chamber at 4 °C. Subsequently, it
   underwent polymerization in a vacuum oven at 37 °C in a nitrogen gas
   atmosphere for 3 h. For homogenization, sample autoclavation, akin to
   the dExPath protocol^[186]42, was employed. The tissue-hydrogel
   composite was transferred to a dish containing a protein denaturation
   buffer, derived from the dExPath protocol^[187]42. This buffer
   contained 50 mM tris(hydroxymethyl)aminomethane (Tris), 20% SDS, 25 mM
   ethylenediaminetetraacetic acid, disodium salt, dihydrate
   (EDTA-Na[2]·2H[2]O), and 200 mM NaCl. Homogenization occurred at a
   temperature range of 105–121 °C under 1.22–2.03 bars for 60–90 min in
   the autoclave. To enhance the hydrogel’s mechanical stability
   post-autoclaving, the cross-linker concentration was increased by
   17.5-fold from the original concentration (Supplementary Table [188]1).
   Following homogenization, the samples were transferred to 9 cm petri
   dishes and washed thrice with 1× PBS. Reduction and alkylation of
   proteins involved adding 20 mM DTT in 1× PBS for reduction and 55 mM
   IAA in 1× PBS for alkylation, each incubated for 30 min. Subsequent
   washes with 1× PBS occurred thrice for 10 min each. The hydrogel sample
   was then incubated in ddH[2]O, allowing it to expand to its maximum
   size.

   By employing Coomassie Brilliant Blue R-250 staining (Sangon, China),
   the process was accelerated to less than 1 h compared to the original
   10-hour duration. Imaging utilized a Zeiss Fluorescence Stereo Zoom
   Microscope (Axio Zoom.V16) by ZEN v3.1 software.

Improved reduction and alkylation conditions

   To ensure the proper unfolding and stabilization of proteins for
   accurate mass spectrometry analysis, we tested four groups of reduction
   and alkylation conditions on three 10-μm mouse liver hydrogel samples.
   Among the four conditions (10 mM DTT and 55 mM IAA; 20 mM DTT and 55 mM
   IAA; 20 mM TCEP and 55 mM IAA; 45 mM TCEP and 100 mM IAA), the
   combination of 20 mM DTT/55 mM IAA exhibited the highest
   carbamidomethyl modification rate, over an average of 96%
   (Fig. [189]2i). This method not only showed comparable rates of
   identifying cysteine-containing peptides and proteins to the
   established ProteomEx method (marked as “Condition 1” in Fig. [190]2i)
   but also demonstrated superior peptide identification performance.

Tissue-hydrogel sample preparation via conventional in-gel digestion

   Manual microdissection was performed on Coomassie-stained
   tissue-hydrogels following reduction and alkylation. The excised
   samples underwent de-staining in a solution of 50% acetonitrile (ACN)
   and 50% ddH[2]O, followed by dehydration in a solution of 50% ACN and
   100 mM ammonium bicarbonate (ABB). After drying in a SpeedVac, trypsin
   digestion was carried out using 12.5 ng/µL trypsin (Hualishi Tech. Ltd,
   China) in 10% ACN and 90% 50 mM ABB for 4 h at 37 °C, followed by
   overnight incubation with an additional 10 mM ABB.

   The resulting peptide solutions were collected, and subsequent steps
   involved the addition of 25 mM ABB followed by 50% ACN and 2.5% formic
   acid (FA) with vortexing, repeated three times with supernatant
   collection. 100% ACN was added until the hydrogel samples were reduced
   to small white dots. The peptide solutions were concentrated using a
   SpeedVac to 20–30 µL, followed by de-salting using C18 micro spin
   columns (Thermo Fisher Scientific, USA). After purification, the
   samples were dried with a SpeedVac, stored at −80 °C, and prepared for
   subsequent LC-MS/MS analysis.

Tissue-hydrogel sample preparation via filter-aided in-gel digestion

   The C18 disk (Empore, USA) was assembled into either a 10 or 250 µL tip
   to create a spintip device, termed as the FAXP tip. After rinsing and
   activating the spintip with 20 or 100 µL of 80% ACN, microdissected
   tissue-hydrogel samples were added, followed by incubation with 50% ACN
   and 100 mM ABB. Dehydration was achieved by adding 100% ACN. Protein
   digestion utilized 12.5 ng/µL trypsin (Hualishi Tech. Ltd, China) in
   10% ACN and 90% 50 mM ABB. The digestion process happened at 37 °C for
   4 h, followed by overnight incubation with an extra 10 mM ABB.
   Following digestion and overnight incubation, the resulting peptide
   solutions underwent sequential steps: addition of 6 or 12 µL of 2% ACN
   and 0.1% trifluoroacetic acid (TFA), elution with 6 or 20 µL of 70% ACN
   and 0.1% TFA (twice), and a final elution with 6 or 20 µL of 100% ACN.
   These peptide solutions were concentrated using a SpeedVac and stored
   at −80 °C for LC-MS/MS analysis.

Immunostaining and gel-making for single-nucleus proteomics

   A 10 µm FFPE liver section from a 2-month-old female mouse was used for
   this study. The tissue section was incubated for 30 min in citrate
   antigen retrieval solution (Leagene, China) and subsequently in
   blocking buffer (Sangon, China) at room temperature for 1 h. The
   section was then incubated with ASGR1 polyclonal antibody (Proteintech,
   China) at a 1:200 dilution. Following primary antibody incubation, the
   samples were treated with goat anti-rabbit IgG H&L secondary antibody
   (Alexa Fluor® 555) (Abcam, UK) at a 1:1000 dilution. The section was
   washed three times with 1× PBS and counterstained with DAPI (Abcam,
   UK). The section was imaged using a Nikon CSU-W1 SoRa confocal
   microscope with a 20× objective (0.8 NA), controlled by NIS-Elements AR
   v5.42.01 software.

   About 50–100 µL of protein anchoring solution was added to the tissues,
   followed by a 1-h incubation at room temperature. The sample was then
   washed three times with 50–100 µL of anchoring termination buffer.
   After washing, the sample was dried in a fume hood, and an activated
   monomer solution was added to the surface. The assembly was placed in a
   humidified chamber and incubated overnight at 4 °C. The following day,
   the sample was covered with a glass coverslip and placed in a preheated
   vacuum drying oven at 37 °C for 3 h under anaerobic conditions. The
   tissue-hydrogel composite was transferred to a dish and submerged in
   protein denaturation buffer for homogenization, followed by reduction
   and alkylation.

   For imaging, after blocking (Sangon, China), a tissue-hydrogel
   composite was incubated overnight at 4 °C with ASGR1 polyclonal
   antibody (Proteintech, China) diluted to 1:200, then washed in 1× PBS
   three times. Next, the hydrogel was incubated for 2 h at room
   temperature with goat anti-rabbit IgG H&L (Alexa Fluor® 555) (Abcam,
   UK) diluted to 1:200. The hydrogel was washed three times with 1× PBS,
   and counterstained with SYTOX Deep Red Nucleic Acid Stain (Thermo
   Fisher Scientific, USA) for 30 min at room temperature. The expanded
   sample was imaged using a Nikon CSU-W1 SoRa confocal microscope with a
   20× objective (0.8 NA), controlled by NIS-Elements AR v5.42.01
   software. The expanded hydrogel was then placed onto an LCM steel frame
   slide (Leica, Germany) and allowed to dry.

Proteomics data acquisition using Orbitrap Astral

   For LCM using the Leica LMD7 system controlled by Leica LMD v8.4
   software, the Y5 filter cube was used to visualize nuclei, with the
   following cutting parameters: power at 60, aperture set between 17 and
   20, speed ranging from 8 to 12, bridge size set to 0, a final pulse at
   6, head current at 100%, and pulse frequency between 120 and 150. The
   same cutting parameters were applied for hepatocytes, using the RHOD ET
   filter cube.

   Each sample obtained using the Leica LMD7 was transferred into the
   10 µL FAXP tip, and 20 µL of 100% ACN was added. The samples were
   incubated at room temperature for 10 min, followed by centrifugation at
   150 × g for 5 min to remove the supernatant. Subsequently, following
   ref. ^[191]43, we added 0.015% n-dodecyl β-d-maltoside (DDM) to the
   trypsin digestion buffer, achieving a final trypsin concentration of
   5 ng/μL. The samples were then incubated at 37 °C for 4 h. Following
   digestion, the resulting peptide solutions underwent sequential steps:
   addition of 6 µL of 2% ACN/0.1% TFA/0.015% DDM, elution with 6 µL of
   70% ACN/0.1% TFA/0.015% DDM (twice), and a final elution with 6 µL of
   100% ACN. These peptide solutions were then concentrated using a
   SpeedVac and reconstituted for LC-MS/MS analysis.

   Single nuclei (n = 6), single mononuclear cells (n = 8), single
   binuclear cells (n = 7), and blank controls (n = 5) were analysed.
   Peptides were analyzed via an Orbitrap Astral mass spectrometer (Thermo
   Fisher Scientific, USA) controlled by Thermo Scientific Xcalibur
   v4.7.69.37 and Orbitrap Astral Tune Application v1.0.100.40. Separation
   of peptide samples utilized an Aurora Elite TS 15 × 75 C18 UHPLC column
   (IonOpticks, Australia) in a Vanquish™ Neo UHPLC system (Thermo Fisher
   Scientific, USA). The liquid chromatography procedure involved a
   19.5-min gradient running step and a 5.5-min column washing step. The
   mobile phase comprised buffer A (98% MS grade H[2]O, 2% MS grade ACN,
   and 0.1% FA) and buffer B (98% MS grade ACN, 2% MS grade H[2]O, and
   0.1% FA). The injection workflow was direct injection.

   The Orbitrap MS full scan settings included a total carrier gas flow of
   3.8 L/min, an Orbitrap resolution of 240k, a scan range of 400 to
   800 m/z, a FAIMS CV of −42 V, an RF Lens of 45%, a normalized automatic
   gain control (AGC) target of 500%, and a maximum IT of 100 ms. The
   Astral DIA-MS^2 scan parameters covered a precursor mass range of 400
   to 800 m/z and a scan range of 150 to 2000 m/z. The DIA isolation
   window was set at 20, the maximum injection time at 40 ms, with HCD
   normalized collision energy at 25%, RF lens at 45%, maintaining a
   normalized AGC target of 800%, and a loop control time of 0.6 s.

Proteomics data acquisition using timsTOF Pro

   The tissue collection was performed using commercially available biopsy
   punches (Integra Miltex, USA) ranging from 0.35 to 3 mm in diameter. A
   total of four mouse liver FFPE slides (10-μm-thick) were analyzed under
   four different reduction/alkylation conditions (n = 3 per condition,
   each with two technical replicates), with in-gel digestion under
   ProteomEx serving as the control (n = 4), using the Parallel
   Accumulation Serial Fragmentation (PASEF) mode of data-dependent
   acquisition (ddaPASEF). For the comparative study between in-gel
   (n = 6) and in-tip digestion (n = 6) methods, 2000-μm punches from a
   single mouse liver FFPE slide (10-μm-thick) were analyzed using
   ddaPASEF. Peptides from different tissue volumes of two consecutive
   mouse liver FFPE slides (10-μm-thick) were processed by in-gel and
   in-tip digestion and analyzed with three replicates per digestion
   method per volume using both ddaPASEF and data-independent acquisition
   using PASEF (diaPASEF).

   The peptide samples, separated on a custom-packed silica column in a
   nanoElute® system (Bruker Daltonics, Germany), utilized a mobile phase
   composed of MS grade H[2]O + 0.1% FA (buffer A) and MS grade ACN + 0.1%
   FA (buffer B) at a flow rate of 300 nL/min. Employing both 25-min and
   50-min effective linear LC gradients, buffer B (%) increased from 5 to
   27% over 25 and 50 min, respectively. Subsequently, it rose to 40% in 5
   and 10 min, followed by a boost to 80%. The 50-min LC gradient resulted
   in enhanced peptide and protein identification, and a higher mean peak
   full width at half maximum (FWHM) averaging from ~2.5 to 3.5
   (Supplementary Fig. [192]11b).

   For MS analysis, eluted peptides were examined using a trapped ion
   mobility spectrometry (TIMS) combined with a quadrupole time-of-flight
   mass spectrometer (timsTOF Pro, Bruker Daltonics, Germany) through a
   CaptiveSpray nano-electrospray ion source. MS data were acquired using
   the Bruker Otofcontrol v6.2 and HyStar v5.1 mass spectrometer control
   software.

   In the ddaPASEF mode, each cycle included ten PASEF scans, operating
   with a ramp time of 100 ms for a total cycle duration of 1.17 s. Ion
   mobility scanning ranged from 0.7 to 1.3 Vs/cm², and MS1 and MS2
   acquisition covered the m/z range of 100 to 1700 Th. To manage
   precursor intensities, those reaching a target intensity of 20,000
   arbitrary units were dynamically excluded for 0.4 min. Additionally,
   singly charged precursors were excluded based on their position in the
   m/z-ion mobility plane.

   In the diaPASEF mode, most settings were the same as those of the
   ddaPASEF mode, except for a ramp time adjusted to 166 ms. The isolation
   windows were defined as outlined in Supplementary Table [193]6.

Proteomic data analysis

   DDA data analysis was performed utilizing FragPipe (version
   18.0)^[194]44,[195]45, integrating the MSFragger search engine (version
   3.5). The analysis involved querying a FASTA file from SwissProt,
   housing 16,985 mouse protein entries (downloaded in September 2018),
   alongside their respective decoy sequences. Furthermore, a separate
   FASTA file from SwissProt, comprising 20,386 human protein entries
   (downloaded in November 2022), along with corresponding decoy
   sequences, was utilized for human sample analysis. “IM-MS” was chosen
   as the MS data type, employing a fragment mass tolerance of 0.05 Da.
   Trypsin was specified as the digestion enzyme, cleaving after “KR”
   residues, except when followed by “P”. Carbamidomethylation was
   assessed for cysteine modification rates and treated as a variable
   modification in this analysis, differing from other DDA files where it
   was set as a fixed modification.

   In DIA data analysis, parameters were set in DIA-NN (version
   1.8.1)^[196]46,[197]47: a 1% false discovery rate for precursors, fixed
   modifications including “C carbamidomethylation” and variable
   modifications including “N-term M excision” and “Methionine oxidation”.
   The criteria included a peptide length range of 7 to 50, a precursor
   charge range of 2 to 4, and m/z ranges for precursor and fragment ions
   between 100 and 1700. Our approach involved selecting “Unrelated runs”,
   “Use isotopologues”, and “MBR” options. For specific studies, heuristic
   protein inference was applied for mouse liver-based analysis. Protein
   inference was set as “off”, and the quantification strategy was set as
   “Robust LC (high precision)”. Other settings remained as default
   values. In-house libraries were utilized for mouse liver-based and CRC
   studies, respectively.

   For subcellular studies, a library-free mode was employed, using a
   FASTA file from SwissProt containing 17,217 mouse protein entries
   (downloaded in August 2024). Modifications were made to the precursor
   (400–800 m/z) and fragment ion (150–2000 m/z) ranges, while the
   “Unrelated runs” option was not selected. The “--relaxed-prot-inf”
   command was included under Additional Options. Protein inference was
   set to “Genes,” and the quantification strategy was configured as
   “Robust LC (high precision).” All other settings remained at their
   default values.

   To enhance clarity, the terminology for peptides and proteins was made
   consistent. Peptides generally referred to amino acid sequences without
   accounting for PTMs (post-translational modifications) or charge
   states, while proteins referred to razor proteins (in FragPipe reports)
   or protein groups (in DIA-NN reports). Specifically, in Figs. [198]2i,
   [199]3b, [200]f,  [201]g, razor proteins and indistinguishable proteins
   (in FragPipe reports) were considered. In contrast, in Supplementary
   Fig. [202]9, PTMs and charge states of precursor peptides were taken
   into consideration.

Protein matrix preprocessing

   Unless otherwise specified, protein matrices from DIA-NN were converted
   to a logarithmic scale for all subsequent analyses, except for mean
   intensity calculations. Prior to the differential expression analysis,
   proteins were filtered to control the missing data ratio. For CRC
   studies, the missing ratio was less than 20% in at least three out of
   the four groups.

Proteomic quality control analysis

   The CRC application study involved six batches (batch design available
   in the source data for Fig. [203]5e), resulting in a total of 131 files
   due to an MS bug in one occurrence. On average, 28,763 peptides and
   5480 proteins were identified across the pooled samples using DIA-MS
   (Supplementary Fig. [204]9a). Besides, robust reproducibility was
   observed in global peptides and proteins, as indicated by the
   coefficient of variation and Pearson’s correlation analysis in pooled
   samples (Supplementary Fig. [205]9b, c). Furthermore, throughout the
   analysis, a steady and low average missed cleavage rate of 18.27% was
   consistently maintained, as evaluated by DDA-MS (Supplementary
   Fig. [206]9d).

Statistical analysis

   For all numerical comparisons, Welch’s t-test (two-sided) was executed,
   while statistical visualizations were made using PRISM (version 10.0.3)
   and R (version 4.4.0). MATLAB (version R2021a) was utilized for
   isotropic analysis, following a previously established
   protocol^[207]21.

   The comparison of DEPs was conducted using the limma package in R,
   which applies linear models and incorporates the duplicateCorrelation
   function to account for within-patient correlations. “Region” was
   included as the primary fixed effect of interest without an intercept,
   and “Patient” was set as the blocking factor. We defined a set of
   planned contrasts to conduct hypothesis testing between specific pairs
   of regions. An empirical Bayes moderated t-test was applied with FDR
   adjustment to account for multiple testing. For any pairwise
   comparison, proteins with an adjusted p value <0.05 and |logFC|> log2
   were considered statistically significant, while ensuring that the
   overall linear model had an adjusted p value <0.05.

   The Pearson’s correlation coefficient was calculated after the
   exclusion of missing values. Hierarchical cluster analysis was
   performed using Ward’s method and Euclidean distance. The uniform
   manifold approximation and projection (UMAP) dimension reduction
   technique was employed to identify the fuzzy topological representation
   of high-dimensional data.

   The pvca R package was utilized to conduct principal variance
   components analysis (PVCA). The top-N principal components were
   selected based on a 90% threshold for explained variance. Patient,
   slide, and region were included as random effects to account for their
   inherent variability. The magnitude of each effect as a proportion of
   the total variance was estimated by weighted average.

   Gene ID conversion was conducted using the clusterProfiler (version
   4.11.0). The Gene Ontology (GO) term annotations were performed using
   org.Mm.eg.db (version 3.18.0), and redundant GO terms were simplified
   manually. The fast gene set enrichment analysis (GSEA) was performed
   using clusterProfiler to identify Cellular Component Ontology terms
   ranging in size from 10 to 300. The Reactome pathway-based analysis was
   conducted using the ReactomePA (version 1.47.0) package to identify
   gene sets ranging in size from 5 to 300. Gene sets with an adjusted p
   value of less than 0.05, as reported by GSEA, were considered
   statistically significant.

Reporting summary

   Further information on research design is available in the [208]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [209]Supplementary information^ (2.8MB, docx)
   [210]Reporting Summary^ (409.2KB, pdf)

Source data

   [211]Source data^ (655.6KB, pdf)
   [212]Transparent Peer Review file^ (43.3MB, xlsx)

Acknowledgements