Abstract

   Myocardial infarction (MI) continues to be a leading cause of death
   worldwide. Even though it is well established that the complex
   interplay between different cell types determines the overall healing
   response after MI, the precise changes in the tissue architecture are
   still poorly understood. In this study, we generated an integrative
   cellular map of the acute phase after murine MI using a combination of
   imaging-based transcriptomics (Molecular Cartography) and
   antibody-based highly multiplexed imaging (Sequential
   Immunofluorescence). This enabled us to evaluate cell type compositions
   and changes at subcellular resolution over time. We observed the
   recruitment of leukocytes to the infarcted heart through the
   endocardium and performed unbiased spatial proteomic analysis using
   Deep Visual Proteomics (DVP) to investigate the underlying mechanisms.
   DVP identified von Willebrand factor (vWF) as an upregulated mediator
   of inflammation 24 hours after MI, and functional blocking of vWF
   reduced the infiltration of C-C chemokine receptor 2 (Ccr2)-positive
   monocytes and worsened cardiac function after MI.

   Subject terms: Computational biology and bioinformatics, Systems
   biology, Imaging, Cardiovascular diseases
     __________________________________________________________________

   Wünnemann et al. generate a subcellular resolution spatial map of the
   murine heart after myocardial infarction, revealing that immune cells
   can infiltrate the organ through the endocardium.

Main

   Myocardial infarction (MI) is an acute disease characterized by large
   shifts in cellular composition and tissue architecture due to cell
   death of cardiac muscle tissue caused by local hypoxia^[58]1. MI
   remains one of the leading causes of death worldwide, despite
   considerable improvements in the prevention and treatment of the
   disease^[59]2,[60]3. Although advancements in restoring blood flow to
   the heart muscle (termed reperfusion) and pharmacological treatment
   strategies have largely reduced short-term deaths, long-term mortality
   after MI continues to be high^[61]4–[62]6. One of the reasons why
   mortality remains high is that we do not understand how molecular cues
   and tissue microenvironment alterations during acute disease affect
   healing and remodeling in the long run. During an acute infarct,
   necrotic cells in the heart release stress signals, including
   pro-inflammatory cytokines and chemokines, leading to the invasion of
   the infarct zone by immune cells—specifically neutrophils,
   monocytes/macrophages (Mo/Mɸ) and later T cells, B cells and natural
   killer cells^[63]1,[64]7–[65]10. Modulation of the types and amount of
   immune cells infiltrating the infarct have been postulated as potential
   treatment targets to improve healing and outcome after
   MI^[66]11,[67]12. A better understanding of the immune cell
   infiltration routes and pathways, detailed tissue microenvironment and
   cellular interactions in the heart during the early phases of MI thus
   holds the promise to deliver potential novel treatment strategies.

   Extensive research has been performed on the temporal dynamics of
   immune cell infiltration during the course of MI in humans and
   mice^[68]1,[69]8,[70]13,[71]14. Cell count estimations of the immune
   infiltrate have been generated using fluorescence-activated cell
   sorting (FACS) of left ventricular cardiac tissue^[72]14–[73]17.
   Single-cell RNA sequencing (scRNA-seq) provided further insight into
   the diverse subtypes of immune cells that infiltrate the necrotic
   myocardium and their different pathway activities and physiological
   roles^[74]18–[75]21. Two recent studies applied single-nucleus RNA
   sequencing (snRNA-seq) alongside untargeted spatial transcriptomics to
   investigate the border infarct zone during MI in mice^[76]22,[77]23.
   Combining the spatial readout with transcriptome-wide measurements,
   both studies identified important transcriptional signatures of the
   infarct border zone. Another large study generating a spatial multiomic
   map of human MI also used untargeted transcriptomics to investigate the
   different cell neighborhoods and tissue architectures during MI
   progression in humans^[78]24. Although these studies were able to
   connect critical molecular patterns and processes during MI to their
   spatial context, they could not accurately quantify the tissue
   microenvironment and cellular neighborhoods, due to the limited spatial
   resolution (approximately 55 µm) of the technology used, which
   effectively captures a mix of 10–20 cells per measurement. Novel
   targeted spatial omics technologies with subcellular resolution are
   transforming our understanding of tissue architecture and the
   corresponding cell type interactions in health and
   disease^[79]25,[80]26. Highly multiplexed approaches to measure tens to
   thousands of transcripts, antibodies or both combined enable a detailed
   description of the changing cellular phenotypes and neighborhoods
   during homeostasis and disease^[81]27–[82]30. Furthermore, developments
   in computer vision now enable cell identification using automated cell
   segmentation and classification algorithms^[83]31–[84]34.

   In the present study, we used combinatorial imaging technologies based
   on RNA detection via fluorescence in situ hybridization (FISH)
   barcoding (Molecular Cartography) and antibody-based Sequential
   Immunofluorescence (SeqIF, Lunaphore COMET), respectively, to
   characterize the changing tissue microenvironment during acute MI in
   mice. Subsequently, we developed scalable computational pipelines
   (nf-core/molkart) tailored for cardiac tissues, using state-of-the-art
   methodology to process these complex, highly multiplexed imaging
   datasets. Using Molecular Cartography and SeqIF across a timecourse of
   the acutely infarcted heart (control prior to infarct, 4 hours,
   24 hours, 2 days and 4 days after MI), in a minimally invasive MI
   model, we were able to characterize the acute MI microenvironment at
   single-cell resolution^[85]35. From this spatiotemporal map of MI, we
   discovered that myeloid cells (specifically Mo/Mɸ) enter the infarct
   region not only via the border and epicardial infarct zone but also via
   the endocardial infarct zone, which was previously unknown. Laser
   microdissection of endocardial regions 24 hours after infarction
   followed by ultrasensitive proteomics was performed to investigate this
   infiltration route, which revealed local signatures of inflammation and
   coagulation factors and highlighted vWF as an immune mediating agent in
   endocardial cells^[86]36. To further investigate the role of vWF,
   antibody-based functional blocking of vWF during the first day after MI
   showed a significant reduction of Mo/Mɸ infiltration via the
   endocardial route and reduced left ventricular function. Therefore, our
   study highlights, to our knowledge for the first time, a critical role
   of the endocardium for infiltration of immune cells into the infarct
   via local upregulation of adhesion factors such as vWF. These results
   highlight previously unknown routes of immune cell infiltration and
   provide novel potential targets for pharmacological intervention.

Results

Single-cell resolved spatial transcriptomic analysis of acute MI

   To characterize the cellular environment during homeostasis and acute
   MI in the mouse heart, we used a combinatorial single-molecule FISH
   (smFISH)-based technology called Molecular Cartography by Resolve
   Biosciences (Fig. [87]1a and Extended Data Fig. [88]1a–c). Molecular
   Cartography allows for the detection of RNA transcripts for up to 100
   candidate genes at single-molecule resolution with high sensitivity in
   selected regions of interest (ROIs). Based on marker gene expression
   from existing publicly available scRNA-seq datasets and expert
   knowledge, we subsequently designed a 100-gene transcript panel
   specifically for this study, to capture major cell types as well as
   inflammatory signals during acute MI (Supplementary Table
   [89]1)^[90]37–[91]41. Using this panel, we selected one ROI per heart
   (2.09 mm^2 to 2.97 mm^2) cross-section to measure endocardial,
   myocardial and epicardial regions at four different timepoints during
   acute MI (control prior to infarct, 4 hours, 2 days and 4 days)
   (Extended Data Fig. [92]1a–c). Identification and abundance of RNA
   transcript spots were highly reproducible across technical replicate
   slides of Molecular Cartography, as highlighted by a strong correlation
   of pseudobulk transcript counts across sections from the same
   biological sample (Supplementary Fig. [93]1a). On average across all
   timepoints, we detected 980,000 RNA transcript spots per mm^2 of heart
   tissue with lower transcript density in infarcted tissues due to fewer
   viable cells within the infarct region (Supplementary Fig. [94]1b).
   Principal component analysis (PCA) of pseudobulk transcriptional
   profiles showed separation of samples by time relative to the induced
   infarct, confirming strong transcriptional shifts in response to acute
   infarction (Supplementary Fig. [95]1c). For each timepoint (control,
   4 hours, 2 days and 4 days), we assayed two biological replicates using
   Molecular Cartography and processed the data with an
   in-house-developed, open-source computational pipeline that we call
   nf-core/molkart (Fig. [96]1b). Despite advances in deep-learning-based
   methods, segmentation of cardiac cells from microscopy images remains
   extremely challenging due to their differences in cell size, shape,
   orientation, multinucleation and RNA content. Comparing four different
   cell segmentation approaches (DeepCell Mesmer’s ‘nuclear’ and
   ‘whole-cell’ models, the Cellpose cytoplasmic model (‘cyto’) and a
   custom Cellpose 2 model), we found that Cellpose with a
   human-in-the-loop trained model performed best on our murine cardiac
   tissue samples when evaluating against independently annotated ground
   truth segmentations; it also showed an overall higher percentage of
   assigned RNA spots to cells (Extended Data Fig.
   [97]2)^[98]33,[99]34,[100]42. Cell typing of cell transcript profiles
   across all images (69,028 cells in total, average number of cells per
   sample = 8,629) identified all major cardiac cell types during acute MI
   (Fig. [101]1c). We found healthy as well as stressed cardiomyocytes
   distinguished by their expression of atrial natriuretic peptide (Nppa)
   and brain natriuretic peptide (Nppb), cells of the vasculature
   (pericytes, smooth muscle cells and endothelial cells), cardiac
   fibroblasts as well as infiltrating myeloid and lymphoid cells (Fig.
   [102]1d). The hypoxic environment in the left ventricle during acute MI
   leads to massive cell death and changes in tissue composition and
   architecture within the left ventricle. In line with these expectations
   and known cell dynamics from literature, we observed a strong shift in
   the cell type composition from healthy to infarcted left ventricular
   tissue during the first 4 days after MI (Extended Data Fig. [103]3a,b).
   Healthy left ventricular tissue composed of cardiomyocytes, endothelial
   cells and fibroblasts showed compositional changes toward an
   environment of dying and stressed cardiomyocytes (Nppa^+) surrounded by
   extracellular matrix (ECM)-producing cardiac fibroblasts and invading
   myeloid cells after MI. Because our selected ROI contained large areas
   of infarct tissue, healthy cardiomyocytes showed the highest decrease
   in cell number from an average of 57.9% of identified cells in controls
   to only about 10% at day 4 in the processed regions. In line with
   increased ECM production and early scar formation, cardiac fibroblasts
   increased from 6.6% to 30.8% of cells within the ROIs. Notably, we
   observed a strong increase of myeloid cells (neutrophils, Mo/Mɸ and
   dendritic cells) from 4% in controls to almost 28% at day 4 in the
   imaged ROIs, indicating ubiquitous infiltration of myeloid cells into
   the infarct (Extended Data Fig. [104]3b). A deeper analysis of immune
   cell types allowed classification of neutrophils, different types of
   Mo/Mɸ as well as lymphoid cells and dendritic cells (Extended Data Fig.
   [105]3c,d). Overall, our analysis highlights the cellular landscape of
   acute MI, recapitulating many known cellular dynamics in a spatial
   context.

Fig. 1. Molecular Cartography of acute MI enables spatial cell typing of the
left ventricular infarct tissue.

   [106]Fig. 1
   [107]Open in a new tab

   a, Schematic overview of the study design. Two biological replicates
   chosen across two technical replicate slides were used for Molecular
   Cartography. Two or three biological replicates were used for SeqIF. b,
   Schematic for spatial transcriptomics data generation (Molecular
   Cartography) and processing (nf-core/molkart). c, UMAP showing the
   joint embedding of 69,028 cells from eight samples (two biological
   replicates per timepoint) over four timepoints across acute MI. d,
   Representative spatial cell type distributions for a sample at 2 days
   after MI. A composite image with all cell types is shown on the left,
   and each cell type’s individual distribution is shown on the right.
   Schematics in a and b were created with [108]BioRender.com. d, days; h,
   hours; LV, left ventricular.

   [109]Source data

Extended Data Fig. 1. Region of interest (ROI) selection and spot
distribution of mouse heart sections from Molecular Cartography.

   [110]Extended Data Fig. 1
   [111]Open in a new tab

   a) Brightfield images of transverse sections of mouse hearts at
   different time points during acute MI (Control = prior to infarct).
   Black rectangles highlight regions selected for Molecular Cartography.
   b) Molecular Cartography RNA spots (100-plex) for corresponding regions
   highlighted in a). Regions with low spot density within the tissue at
   the 4 h, 2 d and 4 d post-MI timepoints demarcate infarct regions with
   cell death, apoptosis and RNA degradation. Note that images in b) show
   scatterplots of RNA spot centroid positions after spot calling by
   Resolve Bioscience and not the raw FISH signals. c) Exemplar regions of
   RNA expression maps of indicated markers over all 4 timepoints.

Extended Data Fig. 2. Comparison of segmentation methods for Molecular
Cartography and SeqIF heart sections.

   [112]Extended Data Fig. 2
   [113]Open in a new tab

   a) Example evaluation on a single image of 4 segmentation models across
   8 panels: ground truth (blue) and prediction (yellow) panels show all
   ground truth (GT) and predicted cells, respectively. True positives
   show GT regions in blue, prediction regions in yellow, and their
   overlaps in green. False negative panels show unmatched GT cells, and
   false positive panels show unmatched predicted cells. Merges, splits,
   and catastrophe panels show GT cells (blue), predictions (yellow), and
   their overlaps (green). b) Heatmap showing counts of true positives,
   false positives, false negatives, merges, splits, and catastrophes
   across the 4 segmentation models evaluated on independently annotated
   GT from 16 image crops (4 per time point) in the SeqIF dataset (total:
   1208 cells). c) Segmentation evaluation metrics based on the GT
   annotations calculated on means of image-specific metrics, with error
   bars showing the standard deviation across images. d) Percentage of
   transcripts assigned to cells (Molecular Cartography) and percentage of
   segmented tissue area (SeqIF). Data show mean ± s.d.

   [114]Source data

Extended Data Fig. 3. Spatial cell type distribution and composition changes
during acute MI as quantified by Molecular Cartography.

   [115]Extended Data Fig. 3
   [116]Open in a new tab

   a) Spatial distribution of cell-types in Molecular Cartography samples
   at four time points each with two biological replicates. b) Cell-type
   composition across acute myocardial infarction as quantified by
   Molecular Cartography. Barplots show mean percentage, points represent
   individual replicate measurements. c) Dotplot showing marker expression
   for identified immune cell subtypes. Immune subtype names are followed
   by their distinguishing marker with an underscore. d) Spatial
   distribution of these immune cell subtypes in two biological replicates
   for 4 h and 2 days.

   [117]Source data

Cellular neighborhood analysis of spatial transcriptomic data highlights the
dynamic spatiotemporal changes during acute MI

   To get a global understanding of the tissue architecture and
   intercellular relationships across acute MI tissues, we applied the
   Multiview Intercellular SpaTial (MISTy) modeling framework to our
   dataset^[118]43. MISTy captures the cell type relationship patterns
   across entire slides and datasets in an unbiased manner ([119]Methods).
   In homeostatic cardiac control tissue, the majority of cell types were
   distributed mostly homogeneously, with cardiomyocytes interspersed with
   cardiac fibroblasts, vascular endothelial cells and pericytes. The
   location of the majority of cell types was, therefore, not informative
   as a predictor of the localization of other cell types. MISTy did,
   however, identify a spatial signature for Nppa^+ cardiomyocytes, whose
   spatial localization was best predicted by endocardial cells (Fig.
   [120]2a). Myeloid cells in control tissue were rare, with a relatively
   homogeneous distribution across the tissue during homeostasis and
   4 hours after MI, indicating that these are likely resident leukocytes.
   To further explore the potential co-localization and interaction
   between cell types, we performed local, bivariate analysis between
   pairs of cell types using LIANA+ (ref. ^[121]44). This local analysis
   between endocardial cells and Nppa^+ cardiomyocytes in controls
   pinpointed the interaction between these cell types to the
   subendocardial region close to the left ventricular lumen (Fig.
   [122]2b). This aligns well with known localization of Nppa in
   trabecular ventricle regions during development, which is maintained in
   homeostatic adult hearts^[123]45–[124]47. By contrast, no spatial
   interaction was identified between myeloid cells and endocardial cells
   in controls (Fig. [125]2c). Visualization of the RNA signal within that
   region confirmed strong Nppa expression close to markers for
   endocardial/endothelial cells (Pecam1) (Fig. [126]2d). Two days after
   MI, MISTy analysis revealed interactions that were not identified in
   control conditions. Besides the remaining strong relationship between
   endocardial cells and Nppa^+ cardiomyocytes, MISTy also identified a
   spatial context between endocardial cells and myeloid cells (Fig.
   [127]2e). Both of these spatial interactions showed clear demarcation
   of a (sub)endocardial infarct zone around the infarct core (Fig.
   [128]2f,g). In line with this finding, we observed a strong RNA signal
   of myeloid markers within the endocardium and the subendocardial layers
   (demarcated by Nppa) (Fig. [129]2h). The local relationship between
   endocardial cells and myeloid cells was also identified at 4 days after
   MI alongside a signal between cardiac fibroblast and myeloid cells
   (Fig. [130]2i). Myeloid cell locations at 4 days after MI were
   additionally predicted by Nppa^+ cardiomyocytes, cardiac fibroblasts
   and cardiomyocytes (Fig. [131]2j). Interestingly, the interaction
   between myeloid cells and fibroblasts was enriched in the border zone
   and epicardial infarct zone (Fig. [132]2k). The highly increased
   abundance of cardiac fibroblasts and their spatial co-localization with
   myeloid cells were further highlighted by the drastic increase of RNA
   molecules encoding ECM components such as Col1a1 (Fig. [133]2l). As
   multiple spatial analyses highlighted unexpected but potentially
   important interactions between endocardial cells and myeloid cells, we
   aimed to validate and quantify this increased local relationship using
   simple measures. Therefore, we calculated the Euclidean distance in
   two-dimensional tissue space between each endocardial cell and its
   nearest neighbor myeloid cell to quantify myeloid cell proximity to
   endocardial cells during acute MI (Fig. [134]2m). Average distances
   between endocardial cells and myeloid cells showed significant
   differences over the MI timecourse (Fig. [135]2n). At day 2 and day 4
   after MI, endocardial cells showed significantly shorter distances to
   myeloid cells compared to the control, whereas the average distance to
   myeloid cells did not change significantly between day 2 and day 4
   (Fig. [136]2n). We performed the same distance analysis to see whether
   myeloid cells show a similar relationship with cardiac fibroblasts at
   the interface to the infarct core and found increased proximity between
   both cell types after MI (Fig. [137]2o). Taken together, our cellular
   neighborhood analysis identified and highlighted an unexpected spatial
   relationship between endocardial cells and myeloid cells, suggesting
   that immune infiltration to the infarct might be mediated via the
   endocardium and the subendocardial Nppa^+ infarct zone (collectively
   referred to as endocardial infarct zone).

Fig. 2. Spatial analysis of Molecular Cartography cell composition during MI
highlights myeloid interactions with the endocardial layer.

   [138]Fig. 2
   [139]Open in a new tab

   a, Spatial cell type relationships in cardiac control tissue as
   calculated by MISTy. Importance indicates spatial interactions across
   the slide between the two cell types highlighted. For all MISTy
   analyses, only interactions with an importance >0.4 and only cell types
   with a gain in R^2 > 5% are shown. R^2 represents the change of
   variance when including the spatial context (paraview,
   radius = 125 µm). b,c, Local bivariate analysis between endocardial
   cells and cardiomyocytes Nppa^+ (b) and endocardial cells and myeloid
   cells (c), respectively. Color indicates the local product as
   calculated by LIANA+. d, RNA spot localization in an endocardial region
   of control tissue, highlighting spatial co-localization of marker genes
   for endothelial/endocardial cells (Pecam1), cardiomyocytes (Pln and
   Nppa), fibroblasts (Pdgfra and Col1a1) and myeloid cells (Cd74, Lyz2
   and C1qa). e, MISTy analysis for left ventricular tissue 2 days after
   MI shows an interaction between endocardial cells and myeloid cells.
   f,g, Local interaction analysis shows the interaction of endocardial
   cells with cardiomyocytes Nppa^+ (f) and myeloid cells (g) in the
   endocardial infarct zone. h, RNA marker expression confirming localized
   expression of myeloid markers in the endocardial infarct zone. i, MISTy
   analysis for 4 days after MI highlights the spatial relationship
   between cardiac fibroblast and myeloid cells around the infarct core.
   j,k, Local interaction analysis shows the interaction of myeloid cells
   with cardiomyocytes Nppa^+ (j) and cardiac fibroblasts (k). l, RNA spot
   localization within the infarct tissue at 4 days after MI. m, Euclidean
   distances between all pairs of cell types were calculated. Euclidean
   distances between all pairs of cell types were calculated and the
   distance to the closest myeloid cell was used for endocardial cells and
   cardiac fibroblasts. n, Euclidean distances between endocardial cells
   to myeloid cells were significantly different across the first 4 days
   after MI (n = 2 biological replicates for all groups, type II ANOVA
   P = 0.0095). Post hoc analysis showed significant differences at 2 days
   (post hoc t-test with Bonferroni correction, P = 0.022 after MI
   relative to control but no difference between 2 days and 4 days
   (P = 0.084)). o, Euclidean distances between cardiac fibroblasts and
   myeloid cells were significantly reduced at 4 days after MI (n = 2,
   same biological samples as in n, post hoc t-test with Bonferroni
   correction, P = 0.038). Bars represent mean distance in micrometers,
   and points represent individual measurements. d, days.

   [140]Source data

Highly multiplexed antibody-based imaging confirms immune cell infiltration
via the endocardial layer in acute MI

   To further investigate the regional distribution of myeloid cells after
   MI and capture spatial temporal patterns across entire heart sections,
   we performed SeqIF on samples from the acute phase after MI,
   independent of Molecular Cartography samples (control in biological
   triplicate; 4 hours, 24 hours and 2 days in biological duplicate) (Fig.
   [141]3a)^[142]48. We optimized an antibody panel to identify healthy
   and stressed cardiomyocytes (Tnnt2 and Ankrd1), endothelial cells
   (CD31), smooth muscle cells (αSMA), cardiac fibroblasts (Pdgfra),
   myeloid cells (CD45, CD68, CCR2, Trem2 and Mpo) as well as DAPI and
   wheat germ agglutinin (WGA) to capture nuclei and the cell membrane,
   respectively (Fig. [143]3b and Supplementary Table [144]2). We
   performed image processing of SeqIF data using MCMICRO, a
   Nextflow-based pipeline that performs subtraction of autofluorescence
   signal from each antibody channel, cell segmentation and fluorescence
   intensity quantification (Fig. [145]3a)^[146]49. Based on our
   experience with segmentation for Molecular Cartography data, we applied
   the same strategy of using WGA and DAPI to train a custom Cellpose 2
   model to segment cells in the SeqIF dataset. To assign phenotypes to
   cells, we used the Pixie workflow, which performs pixel clustering
   using self-organizing maps (SOMs) to generate pixel maps of tissues
   (see [147]Methods for more details) (Extended Data Fig.
   [148]4a)^[149]50. In these pixel maps, groups of pixels with similar
   marker intensity profiles across the SeqIF dataset are clustered
   together, allowing for classification of different cell types and
   tissue regions across the entire timecourse of acute MI (Extended Data
   Fig. [150]4b)^[151]50. In line with our results from Molecular
   Cartography, we found a continuous decrease in pixel phenotype clusters
   for cardiomyocyte marker Tnnt2 and an increase in pixel phenotype
   clusters for myeloid cells (Mo/Mɸ and neutrophils) in the first 2 days
   after MI (Extended Data Fig. [152]4b,c). We also found stressed and
   dying cardiomyocytes positive for an Ankrd1 pixel cluster, clearly
   demarcating the infarct core starting already at 4 hours after MI
   (Extended Data Fig. [153]4a). We used the pixel maps to perform cell
   phenotyping using Cellpose cell masks in a second clustering step (Fig.
   [154]3c,d and Extended Data Fig. [155]4d). This pixel-level phenotyping
   workflow enabled us to profile the spatial localization of cardiac
   cells within the infarcted heart based on our highly multiplexed
   imaging data, at an unprecedented scale for entire heart
   cross-sections. To further investigate and independently validate the
   potential relationship between endocardial and myeloid cells that we
   found in Molecular Cartography data, we repeated the distance analysis
   between these two groups of cells. In line with our findings with
   Molecular Cartography, the distance between myeloid cells and
   endocardial cells decreased during the first 2 days after MI (Fig.
   [156]3e). Interestingly, myeloid cells were closest to endocardial
   cells only 24 hours after an infarct (median distance = 24 µm),
   suggesting that attachment and infiltration of these immune cells via
   the endocardial layer might be a rapid process. To quantify the extent
   of different infiltration routes of myeloid cells into the infarct, we
   partitioned the infarcted heart images into regional compartments
   (endocardial infarct zone, epicardium, infarct core and border zone)
   guided by expression patterns of Tnnt2, Ankrd1, WGA and CD31
   ([157]Methods) with subsequent cell quantification (Fig. [158]3f,
   Supplementary Figs. [159]2 and [160]3 and Extended Data Fig. [161]5a).
   Using SeqIF and conventional immunofluorescence staining, we found a
   strong increase specifically for Mo/Mɸ expressing CCR2 (CCR2^+ Mo/Mɸ)
   in the endocardial infarct zone, peaking at 24 hours after MI (Fig.
   [162]3g and Extended Data Fig. [163]5). At 2 days after MI, the
   relative number of myeloid cells remained high within the endocardial
   infarct zone. However, we additionally found an increased density of
   CCR2^+ Mo/Mɸ within the epicardial infarct layer. Quantification of
   absolute CCR2^+ Mo/Mɸ numbers over time identified the border zone as
   the predominant invasion route after 2 days. Closer inspection of
   CCR2^+ Mo/Mɸ distribution in the endocardial layer showed CCR2^+ Mo/Mɸ
   already being attached to the endocardium at 4 hours after MI, with
   occasional infiltration events (Fig. [164]4a–c and Extended Data Fig.
   [165]6). Of note, we did not observe a high abundance of
   non-endocardial endothelial cells in regions with high CCR2^+ Mo/Mɸ
   density within the endocardial infarct zone during the first 24 hours
   (Extended Data Fig. [166]6).

Fig. 3. Highly multiplexed imaging using SeqIF and conventional
immunofluorescence during the first 2 days of acute MI confirms infiltration
of myeloid cells via the endocardial layer.

   [167]Fig. 3
   [168]Open in a new tab

   a, Schema of experimental design for SeqIF data generation and
   processing. For SeqIF, three biological replicates were sampled for
   controls and two biological replicates for the remaining timepoints.
   All SeqIF replicates were different mice than those used for Molecular
   Cartography experiments. b, Representative SeqIF of mouse heart
   cross-sections using 10 antibodies before (left) and at 24 hours after
   (right) MI. Magnifications on the right highlight endocardial niches in
   the infarct, characterized by the presence of stressed cardiomyocytes
   (Ankrd1^+, orange) and attachment of immune cells to endocardial cells
   (CCR2^+ in green, Mpo^+ in violet and CD31 in yellow). ROIs, 50 µm. c,
   Cell phenotyping heatmap from Pixie highlighting marker expression in
   different cell masks across the entire dataset. Legend represents the
   scaled marker expression, which was capped at 3. d, Pixie cell
   phenotyping for one representative sample per timepoint. Each pixel is
   colored based on its cell pixel cluster as calculated by Pixie. Cell
   type colors correspond to heatmap grouping colors in c. e, Distances in
   micrometers from endocardial cells to the closest myeloid cell
   quantified at four different timepoints from SeqIF images. Distances
   show a significant change across the measured timepoint (n = 2
   biological replicates for all timepoints, type II ANOVA P = 0.0279).
   Bars show mean distances across biological replicates, and points
   represent mean distances per biological replicate. f, Schema of
   anatomical annotations used to calculate myeloid cell infiltration. g,
   Quantifications of Ccr2^+CD68^+ Mo/Mɸ in different spatial regions in
   cross-sections from SeqIF. Relative cell type abundance is visualized
   as cells per mm^2. Bars show mean abundance, and points represent
   individual measurements from biological replicates (n = 2 for each
   timepoint). Schematics in a and f were created with [169]BioRender.com.
   CM, cardiomyocyte; d, days; EC, endothelial cell; FB, fibroblast; h,
   hours; Leuko, leukocyte; SMC, smooth muscle cell.

   [170]Source data

Extended Data Fig. 4. SeqIF pixel clusters and cell phenotypes across acute
MI time series.

   [171]Extended Data Fig. 4
   [172]Open in a new tab

   a) Pixel phenotype map for mouse heart images produced with SeqIF
   during a time course of acute myocardial infarction. Pixels were
   clustered using self-organizing maps leveraging Pixie and colored
   according to their assigned pixel cluster. A total of 9 different pixel
   clusters were classified. b) Heatmap of quantified marker expression in
   the corresponding pixel phenotype clusters. Colors for pixel clusters
   correspond to visualization in a. c) Quantification of pixel phenotypes
   across acute MI reveals strong reduction in Tnnt2+ pixels, increase in
   Ankrd1+ pixels and an increase in pixel clusters for myeloid cells
   (CD45+, Mpo+, Ccr2+, Trem2+, CD68+) during the first four days post MI.
   Colors correspond to pixel phenotypes visualization in a. Bars
   represent mean values from two biological replicates and points
   represent individual measurements. d) Zoom-ins for endocardial infarct
   zone regions from SeqIF images with corresponding cell typing
   highlighted. Top row shows SeqIF images with stainings for DNA (Hoechst
   = cyan), Cd31 (orange), Ccr2 (green) and Mpo (pink) at 3 different time
   points (4 h, 24 h and 48 h). Bottom row shows corresponding cell
   segmentations and cell types with endothelial cells (orange), Ccr2+
   monocytes / macrophages (green) and neutrophils (pink). All other cell
   types are marked in grey for visualisation purposes.

   [173]Source data

Extended Data Fig. 5. Quantification of immune cell infiltration based on
conventional IF imaging.

   [174]Extended Data Fig. 5
   [175]Open in a new tab

   a) Schematic highlighting different regions for quantification of
   immune cell infiltration. Absolute (b) and relative (c) numbers of
   CCR2 + /CD68+ Mo/Mɸ in different regions of the heart as depicted in a,
   using conventional immunofluorescence staining for CCR2, CD68, CD31,
   WGA and DAPI. Bars show mean abundance and points represent individual
   measurements. P values were determined by 2-way ANOVA followed by
   Tukey’s multiple-comparison test. Only significant comparisons between
   timepoints within each region are displayed. *P < 0.05 vs. pre,
   #P < 0.05 vs. 4 h, §P < 0.05 vs. 24 h.

   [176]Source data

Fig. 4. Mo/Mɸ infiltrate the infarcted heart via the endocardium.

   [177]Fig. 4
   [178]Open in a new tab

   a–c, Representative SeqIF stainings showing selected markers, including
   CD31 (yellow), CCR2 (magenta), CD68 (green) and DAPI (blue). Staining
   of sections 4 hours (a), 24 hours (b) and 2 days (c) after MI indicates
   increased attachment (asterisk) and transmigration (arrow), resulting
   in high density of CCR2^+CD68^+ cells in the (sub)endocardial infarct
   zone and movement of CCR2^+ cells toward the infarct core. Mid and
   right panels represent magnifications of the marked box in the left
   overview panel. d, days; h, hours; LV, left ventricular.

Extended Data Fig. 6. SeqIF staining of infiltrating Mo/Mɸ in the endocardial
layer after MI and quantification of Mo/Mɸ and CD31+ cells of the vasculature
from SeqIF data using a binning strategy.

   [179]Extended Data Fig. 6
   [180]Open in a new tab

   a–c) Representative SeqIF images showing selected markers including
   CD31 (yellow), CCR2 (magenta), CD68 (green) and DAPI (blue) with
   attachment (asterisk) and transmigration (arrow) of CCR2 + CD68+
   monocytes/macrophages. d) The endocardial infarct zone in SeqIF images
   was split into bins from lumen towards the infarct core to quantify
   cells across the bins over time. Cyan: DAPI, yellow: Cd31, magenta:
   Ccr2, cyan square: endocardial cells, yellow circle: other endothelial
   cells, magenta triangle: Ccr2+ Mo/Mɸ. e) Relative cell abundance in mm2
   across different endocardial infarct zone bins from SeqIF shows an
   increase of Ccr2+ Mo/Mɸ at the endocardial layer around 24 h, while
   abundance Cd31+ cells of the vasculature remains constant over time.

   [181]Source data

   In the epicardial layer, CCR2^+ Mo/Mɸ were either directly attached to
   the epicardium or in close proximity to epicardial vessels
   (Supplementary Fig. [182]4a–c). Taken together, our results—across
   multiple technologies and quantification methods—clearly indicate a
   progressive infiltration of myeloid cells via different cellular layers
   during the acute phase after MI in a non-reperfusion context and
   highlight a previously undescribed route via the endocardial infarct
   zone that immune cells can take to infiltrate the left ventricle to
   reach the infarct region.

Spatial proteomics of the endocardial layer highlights vWF involvement in
immune cell infiltration

   Following the discovery of early immune cell infiltration into the
   endocardial infarct zone, we aimed to identify potential factors
   mediating the recruitment, adhesion and infiltration of myeloid cells
   via the endocardium using ultrasensitive mass spectrometry-based
   proteomics^[183]36. Therefore, we used laser capture microdissection to
   excise the endocardial region from healthy mouse hearts (control) and
   hearts 24 hours after MI. For hearts with MI, we split endocardial
   cells into two groups: those that were within the infarct zone (MI IZ)
   and those that were remote to the infarct (MI remote) (Fig. [184]5a).
   Proteomes of the different endocardial tissue samples encompassed an
   average number of 3,274 proteins after quality control filtering
   (Supplementary Fig. [185]5a). Despite the very low amount of input
   tissue material, the proteomic data were of high quality with a low
   amount of missing protein values (4–16% across samples) (Supplementary
   Fig. [186]5b). PCA of proteomic samples showed clear separation of
   control endocardial samples compared to endocardial cells from
   infarcted samples, indicating a reproducible perturbation of the
   endocardial layer protein signature 24 hours after MI (Fig. [187]5b and
   Supplementary Fig. [188]5c). Interestingly, the remote endocardial
   layer signature from infarcted hearts (MI remote) was sufficiently
   different from control endocardial cells to distinguish them in the
   PCA, but there were very few significantly differentially expressed
   proteins (DEPs) between these two conditions (Supplementary Fig.
   [189]5d). By contrast, significant differential protein expression of
   many proteins was observed in the endocardial region in the infarct
   zone (MI IZ) relative to the remote endocardial region (MI remote)
   (Fig. [190]5c). Pathway analysis using hallmark gene sets of DEPs
   between MI IZ and MI remote revealed upregulated pathways related to
   immune cell activation (complement system, inflammatory response and
   IFNγ response) as well as downregulated pathways related to energy
   metabolism (oxidative phosphorylation and fatty acid metabolism) (Fig.
   [191]5d). Interestingly, we identified genes for coagulation pathways
   strongly upregulated in MI IZ samples relative to MI remote regions
   (Fig. [192]5d). We investigated the cell specificity of these pathway
   results using a published snRNA-seq dataset and found vWF as the most
   specific endocardial protein that was significantly upregulated in MI
   IZ compared to MI remote, similar to known endothelial adhesion
   molecules such as Vcam1 (Fig. [193]5e,f)^[194]22. vWF is a multimeric
   protein that plays a central role in vascular homeostasis and is
   involved in inflammatory processes^[195]51. Interestingly, vWF was not
   significantly upregulated in the remote endocardial regions of
   infarcted hearts (MI remote) compared to endocardial regions of control
   hearts (Fig. [196]5f). To confirm increased localization of vWF
   proteins in the endocardial infarct zone, we performed conventional
   immunofluorescence staining of vWF in infarcted hearts and found a
   significantly stronger signal in the endocardial infarct area, with an
   almost absent signal in the remote region (Fig. [197]5g and Extended
   Data Fig. [198]7a,b). The distribution of vWF^+ staining,
   interestingly, was not uniform across the infarct adjacent endocardial
   layer but stronger at endocardial sites where the ventricular tissues
   formed pockets, compared to smooth regions. Immunofluorescence
   stainings of murine hearts 24 hours after ischemia/reperfusion injury
   also showed similar staining patterns with increased vWF expression
   (Extended Data Fig. [199]7c,d). To investigate whether vWF also plays a
   role in human MI, we reprocessed a cellular indexing of transcriptomes
   and epitomes by sequencing (CITE–seq) dataset of explanted human hearts
   from donors and patients with acute MI by Amrute et al. ^[200]52 (Fig.
   [201]5h). This dataset consists of healthy donors, patients with acute
   MI and patients with chronic ischemic and non-ischemic cardiomyopathy,
   from which we focused on healthy donors (n = 6) and patients with acute
   MI (n = 4). Differential gene expression analysis demonstrated a
   significant increase of endocardial vWF expression in patients after
   acute MI compared to healthy donors (Fig. [202]5i). Collectively, our
   DVP analysis during acute MI revealed local spatial differences between
   endocardial regions within the same heart and highlighted upregulation
   of vWF as a specific response of the endocardium to the local
   inflammatory signals from the infarct zone.

Fig. 5. Laser capture microdissection coupled to ultrasensitive proteomics at
1 day after MI reveals local vWF upregulation in endocardial cells.

   [203]Fig. 5
   [204]Open in a new tab

   a, Schema for experimental design comparing endocardium of control
   (green), infarct zone (IZ, purple) and remote regions (orange). b, PCA
   of the three indicated experimental groups (n = 3–4 biological
   replicates). c, Volcano plot of the proteomic differential comparison
   between infarct endocardium and remote endocardium. Significantly
   differentially expressed proteins are displayed in purple (upregulated
   in infarct endocardium) and orange (downregulated in infarct
   endocardium). Differential expression was assessed by an empirical
   Bayes moderated t-test (limma-voom). d, Pathway enrichment analysis
   results using hallmark gene sets of DEPs between MI IZ and MI remote.
   e, Endocardial cell specificity analysis of DEPs between MI IZ and MI
   remote. The x axis shows specificity of gene expression (as
   approximated by differential marker gene expression) based on snRNA-seq
   data from Calcagno et al.^[205]22, whereas the y axis shows log[2]FC
   from differential protein expression analysis shown in c. f, Expression
   plots of three proteins of interest based on ultrasensitive proteomics.
   Cdh11 is a marker for endocardial cells and not differentially
   expressed, whereas Vcam1 and vWF show significant differential
   expression during acute MI (n = 3 for control and n = 4 for MI remote
   and MI IZ groups; P values shown are from differential expression
   assessed by an empirical Bayes moderated t-test (limma-voom)). Line
   represents mean expression, and points represent individual
   measurements. g, Representative conventional immunofluorescence
   staining of vWF (magenta) alongside CD31 (yellow) 24 hours after murine
   MI. h, UMAP of human cardiac cell types identified in snRNA-seq data
   from Amrute et al.^[206]52. Endocardial cell cluster used for
   differential gene expression analysis is highlighted with a dotted
   circle. i, Violin plot of normalized RNA expression from snRNA-seq data
   aggregated to pseudobulk for all donor samples (n = 6) and acute MI
   samples (n = 4) from Amrute et al.^[207]52, with significant
   upregulation of vWF in human endocardial cells of acute MI samples
   (DESeq2 analysis on pseudobulk expression values; P = 1.4 × 10^−5).
   Schematic in a was created with [208]BioRender.com. AMI, acute MI; FC,
   fold change; NK, natural killer; PC, principal component; Pval, P
   value.

   [209]Source data

Extended Data Fig. 7. Immunofluorescent staining of CCR2+ Mo/Mɸ and
endocardial vWF after MI and ischemia/reperfusion injury.

   [210]Extended Data Fig. 7
   [211]Open in a new tab

   a, b) Quantification of vWF in different regions 24 h after MI based on
   immunofluorescence in female and male mice. c) Immunofluorescence
   stainings of CCR2 and vWF in the infarct zone 24 h after
   ischemia/reperfusion injury. d) Quantification of endocardial vWF after
   ischemia/reperfusion injury. e) Quantification of infiltrating Mo/Mɸ at
   the endocardial region in both control hearts and hearts after
   ischemia/reperfusion injury.

   [212]Source data

Functional blocking of vWF modifies immune cell infiltration and infarct
recovery

   Immunofluorescence co-staining of CCR2 and vWF highlighted a strong
   correlation between the presence of vWF within the endocardial infarct
   region and locally attached or already infiltrated CCR2^+ Mo/Mɸ after
   MI and myocardial ischemia/reperfusion (Fig. [213]6a,b, Extended Data
   Fig. [214]7d,e and Supplementary Fig. [215]6a,b). Of note,
   immunofluorescent co-staining of platelets revealed their presence in
   some, but not all, of these infiltration areas with expression of vWF
   (Supplementary Fig. [216]6c). We conclude that vWF-dependent immune
   recruitment in this context is, at least partially, platelet
   independent. To investigate the functional role of vWF in the
   recruitment and infiltration of myeloid cells via the endocardium
   during acute MI, a well-characterized polyclonal antibody (against
   human vWF, which is also highly cross-reactive with murine vWF) was
   used to block vWF function in murine MI (Fig. [217]6c)^[218]53–[219]55.
   Under baseline conditions, anti-vWF blockade led to no cardiac-specific
   phenotype in healthy mice (Supplementary Fig. [220]7a–d). Control IgG
   or anti-vWF antibodies were injected intravenously at 0 hours and
   24 hours after MI. Blockade of vWF resulted in a significant decrease
   in recruited Mo/Mɸ in the infarcted myocardium as quantified by flow
   cytometry 2 days after MI, whereas blood levels of Mo/Mɸ were unaltered
   (Supplementary Fig. [221]7e,f). Immunofluorescence staining
   demonstrated that this effect was mainly explained by dramatically
   reduced Mo/Mɸ within the endocardial infarct zone (Fig. [222]6d).
   Interestingly, blockade of Mo/Mɸ recruitment mediated by vWF blocking
   at the endocardium led to impaired healing and deteriorated long-term
   outcome 2 weeks after MI induction as shown by echocardiography (Fig.
   [223]6f–j). Moreover, histopathological evaluation revealed more
   pronounced infarct thinning in anti-vWF-treated mice compared to
   control mice (Fig. [224]6k–m). Our findings using DVP and functional
   experiments have, therefore, uncovered a potentially critical role of
   the endocardium in facilitating infiltration of myeloid cells that is
   likely mediated by vWF.

Fig. 6. Blockade of vWF results in decreased Mo/Mɸ recruitment and impaired
infarct healing.

   [225]Fig. 6
   [226]Open in a new tab

   a, Representative immunofluorescence staining of vWF, CCR2, CD31 and
   DAPI 24 hours after MI. b, Correlation of CCR2^+ cells and mean
   fluorescence intensity (MFI) of vWF within the endocardial infarct
   area. A total of 164 annotations of similar size were added across the
   endocardium to assess vWF MFI and the presence of CCR2^+ cells within
   each annotation (n = 3 samples; dashed lines indicate 95% confidence
   intervals). R represents the correlation coefficient. c, Schema of the
   experimental setup for functional vWF blocking during acute MI. d,
   Quantitative analysis of cardiac Mo/Mɸ 2 days after MI based on
   conventional immunofluorescence in different regions (BZ, border zone;
   core, infarct core; endo, endocardial IZ; epi, epicardial IZ). Boxes
   represent mean ± s.d. (n = 3 samples per group). P values were
   determined by two-way ANOVA followed by Sidakʼs multiple comparison. e,
   Representative immunofluorescence stainings of CCR2^+ cells (green)
   after both IgG and anti-vWF treatment 2 days after MI in the
   endocardial IZ. f, Representative B-mode images 14 days after MI
   induction. Green lines depict left ventricular endocardial
   displacement. g–j, Global longitudinal strain (g), left ventricular
   ejection fraction (h) and end-systolic (i) and end-diastolic volume (j)
   determined by echocardiography 14 days after MI induction. k, Heart
   weight (HW) to body weight (BW) ratios after organ removal 14 days
   after MI. l, Quantification of infarct thickness based on
   histopathological evaluation 14 days after MI. Data show mean ± s.d.;
   points display individual measurements (n = 8 samples per group). P
   values were calculated using two-tailed Studentʼs t-test. m,
   Representative images of Masson trichrome stainings of both IgG-treated
   and anti-vWF-treated mice 14 days after infarction. h, hours; IF,
   immunofluorescence; LV, left ventricular.

   [227]Source data

Discussion

   This study reveals a previously undescribed route for immune cell
   infiltration of monocytes during the acute phase of MI in mice. Using a
   combination of state-of-the-art spatial omics approaches such as
   Molecular Cartography, SeqIF and DVP, we identified the endocardium as
   an important region of immune cell attachment and infiltration,
   mediated by the local upregulation of vWF.

   A wide range of clinical trials are currently testing the potential of
   immunomodulation to prevent adverse remodeling after MI. A better
   mechanistic understanding of leukocyte accumulation and the
   differentiation of beneficial and harmful players in this context might
   lead to novel and improved strategies to improve MI healing.
   High-resolution spatial assays allowed us to observe and identify the
   attachment and infiltration of myeloid cells through the endocardial
   layer of the left ventricle during the critical early timeframe of
   acute MI. Over time, the healing heart appears to develop different
   hubs of local inflammation: early on, a large number of monocytes
   accumulate via the endocardial layer and, at later stages, with a
   dominant and well-described focus in the border zone. Depending on the
   timepoint, it should also be considered that additional infiltration
   might occur via the subendocardial microvasculature. Although most
   experiments were carried out after permanent coronary occlusion, we
   also observed this effect in the context of cardiac
   ischemia/reperfusion.

   The central mediator for the attachment of monocytes in the lumen of
   the left ventricle was identified to be vWF, released from endocardial
   cells in the infarct zone. vWF is known to be involved in platelet
   adhesion and aggregation, and, although we did observe some vWF cell
   clusters with platelets present, many myeloid cell–endocardial cell
   interactions occurred in the absence of platelets, suggesting that the
   vWF effect on immune cell infiltration is at least partially platelet
   independent^[228]56. The in vivo blockade of vWF resulted in a
   significant reduction in subendocardial monocyte accumulation. Recent
   studies suggest that vWF can have direct effects on immune cells,
   independent of platelets^[229]57. vWF might not only impact migratory
   capabilities but might also modify their pro-inflammatory activation or
   shift macrophage metabolism toward glycolysis in a p38-dependent
   manner. Although we did not observe any hemorrhage or other signs of
   bleeding in the treated animals, blockade of vWF might also affect
   primary hemostasis and, thereby, impact monocyte accumulation
   subsequently. The observed effect that a reduced accumulation of CCR2^+
   monocytes (by vWF blockade) results in impaired left ventricular
   function was surprising and could indicate that this particular subset
   plays a beneficial role in the context of infarct healing and might be
   distinct from monocyte subsets present in the border zone of the
   infarcted heart. Whether the route of infiltration or the local
   micro-milieu instructs the flavor of the myeloid cell remains to be
   explored.

   We observed a localized upregulation of vWF only in endocardial cells
   within the infarct zone, suggesting potential signaling effects from
   dying cardiomyocytes and other cells in the spatial neighborhoods of
   the infarct. It appears that vWF is primarily released from endocardial
   cells within hypokinetic or akinetic areas of the left ventricle, which
   could also indicate an impact of altered blood flow on the secretome of
   this cell layer. Although we do not have any data to confirm these
   hypotheses, future studies might investigate the signaling molecules
   reaching endocardial cells and try to isolate which factors are causing
   the upregulation of vWF.

   Although our findings are based on multiple types of experimental
   approaches, and we consistently observed immune cell infiltration via
   the endocardial infarct zone regardless of the technology used, our
   study does have several limitations. First, the transcript panel that
   we selected for Molecular Cartography is limited to 100 different
   transcripts and was manually curated and is, therefore, lacking many
   potentially interesting and important markers for detailed description
   of immune cell types. Furthermore, some transcripts are not reliably
   detected with probes that we selected in our target tissue and,
   therefore, might observe dropout in cells that do express these RNA,
   leading to dropout effects. Another limitation we face is the lack of
   high-quality reference data for cardiac cell segmentation. Although we
   trained a custom cell segmentation model using Cellpose 2, with as
   little bias as possible, this segmentation does not represent a real
   ground truth but solely an approximation. Hopefully, as more
   high-resolution spatial omics datasets become available, public
   datasets with representative cardiac cell segmentation masks will
   become available for future studies.

   We attempted additional validations using publicly available data
   sources to find a signature of endocardial immune cell infiltration
   during acute MI. Unfortunately, most published studies either do not
   match the temporal window that is critical to observe this effect or
   employ spatial methods with insufficient resolution^[230]22,[231]23.
   Although we employed cell deconvolution methods on some of these
   datasets, it was not possible to identify endocardial regions with
   sufficient sensitivity to validate our findings using such datasets.

   In conclusion, our findings of immune cell infiltration via the
   endocardial infarct zone in acute MI open new and exciting avenues to
   modulate the immune response after an infarct and provide novel
   opportunities for therapeutic avenues and drug delivery.

Methods

Mouse experiments

   C57BL/6NRj female mice were obtained from Janvier Labs and were studied
   at 10–12 weeks of age. Mice were housed under standard laboratory
   conditions with a 12-h light/dark cycle and access to water and food ad
   libitum. All animal procedures were approved by the institutional
   review board of the University of Heidelberg, Germany, and the
   responsible government authority of Baden-Württemberg, Germany (project
   numbers G-106/19 and G-94/21).

Minimally invasive induction of MI

   MI was induced in a minimally invasive manner under echocardiographic
   guidance^[232]35. In brief, mice were anesthetized with inhalation of
   2% isoflurane and placed on a Vevo imaging station connected to a Vevo
   2100 system (VisualSonics). After a brief evaluation of cardiac
   function, the left coronary artery was visualized. After attaching a
   neutral electrode, a monopolar needle controlled by a micromanipulator
   was inserted into the chest and placed on the coronary artery. The
   vessel was coagulated with high-frequency electricity using an
   electrosurgical station that was connected to both electrodes. After
   removal of the needle, successful MI was confirmed by persisting
   absence of a Doppler signal and akinesia in the affected part of the
   left ventricular wall. For induction of myocardial
   ischemia/reperfusion, the left coronary artery was occluded
   (60 minutes) via two micromanipulator-controlled needles under
   echocardiography guidance as previously described^[233]58. Healthy
   untouched mice showed similar Mo/Mɸ levels in the myocardium compared
   to sham-treated mice and were, therefore, used as biological controls
   (Supplementary Fig. [234]8).

Organ removal and preparation

   Peripheral blood was collected by facial vein puncture in heparinized
   tubes. Hearts were excised after cervical dislocation and rinsed
   extensively in ice-cold PBS to remove remaining blood within the left
   ventricular lumen and vasculature. After transverse sectioning using a
   scalpel, freshly dissected mouse cardiac samples were embedded in
   Tissue-Tek Optimal Cutting Temperature (OCT) compound (Sakura) in a
   plastic cassette and were immediately placed in an isopentane bath on
   dry ice for further processing.

Molecular Cartography (highly multiplexed smFISH) of murine MI samples

   Next, 10-µm-thick cryosections were placed within the capture areas of
   cold Resolve Biosciences slides. Samples were sent to Resolve
   Biosciences on dry ice for analysis. Upon arrival, mouse tissue
   sections were thawed and fixed with 4% v/v formaldehyde (Sigma-Aldrich,
   F8775) in 1× PBS for 30 minutes at 4 °C. After fixation, sections were
   washed three times in 1× PBS for 1 minute, followed by a 1-minute wash
   in 70% ethanol at room temperature. Fixed samples were used for
   Molecular Cartography (100-plex combinatorial smFISH) according to the
   manufacturer’s instructions and as previously described^[235]59. The
   probes for 100 genes were designed using Resolve Biosciences’
   proprietary design algorithm. Supplementary Table [236]1 highlights the
   gene names and catalog numbers for the specific probes designed by
   Resolve Biosciences.

Image processing of Molecular Cartography data

   Slides used for combinatorial smFISH imaging with Molecular Cartography
   for 100 candidate transcripts were subsequently stained for nuclei
   (DAPI) and WGA. DAPI and WGA images as well as RNA spot tables from
   Molecular Cartography were processed using an in-house-developed
   Nextflow pipeline: nf-core/molkart (10.5281/zenodo.10650748, revision:
   81eafe9f9993d4daf16371ba3804ce9ae08053ad). The pipeline is part of the
   nf-core collection of workflows adhering to strict guidelines for best
   practices in Nextflow workflow development. Many core components of the
   pipeline were made available as nf-core DSL2 modules to facilitate easy
   enhancement of the pipeline by other users and to enable reuse of
   pipeline components by Nextflow imaging pipelines in the
   future^[237]60. First, consecutive Gaussian blurring was used to fill
   in black grid lines from Molecular Cartography imaging using the Python
   tool MindaGap^[238]61. Next, image stacks of DAPI and WGA were created,
   and contrast-limited adaptive histogram equalization (CLAHE) was
   applied to improve contrast across stainings for automated
   segmentation^[239]62. Training images for Cellpose (TIFF format) were
   created with the use of the ‘create training subset’ functionality of
   the nf-core/molkart pipeline. Cell segmentation was performed using
   several different segmentation algorithms to compare them on cardiac
   images (Supplementary Fig. [240]3). The segmentation algorithms used
   here were DeepCell Mesmer, which performs nuclear and whole-cell
   segmentation, and Cellpose^[241]33,[242]34,[243]42. For Cellpose
   segmentation, we applied the ‘cyto’ model, but we additionally trained
   a custom Cellpose 2 model, by selecting small crops (1,000 × 1,000
   pixels) of DAPI and WGA image stacks across the entire dataset, using a
   human-in-the-loop approach, as described by Pachitariu and
   Stringer^[244]33,[245]34. We used the baseline model CPx with
   flow_threshold = 0.6 and cellprob_threshold = 0 to segment an initial
   image crop, and we corrected wrong segmentations and retrained the
   model. This process was consecutively applied to retrain the model on
   new image crops until segmentation captured most cells correctly in the
   presented image. After segmentation, the resulting masks were size
   filtered to remove extremely small and extremely large objects that do
   not represent real cells (below 200 and above 200,000 pixels). To
   assign spots to cells, spots are first filtered for potential
   duplicates using the MindaGap duplicatefinder function, which filters
   potential duplicate RNA spot calls along black grid lines. Deduplicated
   RNA spots are then assigned to segmentation masks using spot2cell.
   Finally, quality control metrics of all relevant steps are collected
   and compiled for inspection via MultiQC. Images and spots were
   visualized using the napari toolkit, which enables fast and interactive
   visualization of large imaging data ^[246]63.

Single-cell analysis of Molecular Cartography data

   Cell-by-feature matrices from nf-core/molkart were imported into R and
   processed using the Seurat package (version
   Seurat_5.0.1)^[247]64–[248]67. We filtered out cells with fewer than 20
   and more than 4,000 RNA counts. We also filtered outlier cells based on
   their segmentation mask shape with extent <0.25 and solidity <0.75 (as
   estimated by the regionprops_table function from the scikit-image
   package in Python) to ensure that only high-quality segmented cells are
   included in the analysis. Cell transcript profile counts were
   normalized using SCtransform in Seurat; principal components were
   calculated; and the first 30 principal components were used for
   integration of samples across time using the IntegrateLayers function
   in Seurat with the method set to
   ‘HarmonyIntegration’^[249]65,[250]67–[251]71. Harmony embeddings were
   then used for uniform manifold approximation and projection (UMAP) and
   cluster identification using shared nearest neighbor analysis using the
   first 30 harmony dimensions. We used cell type label transfer to
   annotate the Molecular Cartography cells using a reprocessed snRNA-seq
   reference from Calcagno et al.^[252]22. To reprocess the snRNA-seq data
   from Calcagno et al., we performed Seurat analysis on the raw data as
   described by the authors in the original publication. The transferred
   labels from Calcagno et al. were used to guide manual annotation of
   cells into cell types and states and produce final labels for all cell
   clusters. To improve the annotation of endocardial cells, we
   additionally manually labeled the endocardial region in all DAPI+WGA
   images using QuPath^[253]72. We exported endocardial masks as GeoJSON
   files in QuPath and processed them using the sf package in R to overlay
   centroid positions of cell masks. Cells that overlapped the endocardial
   region and expressed Pecam1 (normalized count >0) or were clustered
   into an Npr3^+ cell cluster were considered as endocardial cells.
   Although Npr3 is primarily expressed in the endocardial cells in our
   dataset, the absence of detected Npr3 transcripts in a subset of
   endocardial cells may reflect transcriptional heterogeneity and/or
   technical limitations of Molecular Cartography in capturing
   low-abundance transcripts (Supplementary Fig. [254]9).

   We characterized structural patterns in immediate cellular
   neighborhoods by extracting cell type to cell type relationships with
   MISTy (R package mistyR, version 1.99.9)^[255]43. MISTy is a multiview
   framework for analysis of spatial omics data by identification of
   robust relationships within the data coming from different spatial
   contexts. Based on Molecular Cartography-assigned cell types, we
   represent each cell type intrinsically as a one-hot encoded vector. To
   capture the structure of the spatial neighborhood of each cell, we
   added a paraview with a radius of 125 µm. The paraview captures the
   neighborhood composition by distance-weighted sum of one-hot encoded
   representation of the cell types in the surrounding of each cell. The
   weights are calculated by a radial basis function with parameter
   equation to the chosen radius. Subsequently, a MISTy model was trained
   using the same view composition for each sample independently. The
   MISTy models are trained on the task of predicting each intrinsic cell
   type by using all variables from the paraview. The MISTy output
   consists of the gain of variance explained per target and importances
   of each predictor–target interaction. The importance of each
   interaction was standardized to 0 mean and unit s.d. across all
   predictors for a given target. The performance and interaction results
   were aggregated per timepoint and filtered to exclude all targets with
   gain of variance explained less than 5% and relationships with
   importances lower than 0.4. MISTy captures robust relationships on a
   global scale—that is, consistently across the whole slide.
   Additionally, MISTy can learn not only simple linear relationships but
   also complex nonlinear relationships. To linearly approximate the sign
   of the remaining relationships and estimate their consistency across
   each slide, we calculated the correlation between the predictor
   variables from the paraview and the target variables from the
   intraview. Although strong correlations are indicative of linear and
   consistent relationships, correlations close to 0 point toward
   nonlinearity or heterogeneity of the form of the interaction across the
   slide, warranting a more targeted local bivariate spatial analysis. We
   used LIANA+ (version 1.0.4) to calculate spatially informed local
   bivariate metrics between pairs of cell types of interest. Similar to
   MISTy, we used one-hot encoded cell type vectors and calculated the
   cell neighborhoods using a Gaussian kernel with a cutoff of 0.1 and a
   bandwidth of 125 µm and l1 standardization of terms. We then use the
   lr_bivar function in LIANA+ to calculate the normalized weighted
   product between two cell type vectors as input. Interactions were
   visualized by plotting local scores on tissue coordinates. To calculate
   Euclidean distances between pairs of cells, we used the Scipy spatial
   packages cdist function^[256]73. Statistical testing on distances was
   performed with ANOVA with a linear ordinary least squares (OLS) model
   and post hoc t-tests with Bonferroni correction using the statsmodel
   application programming interface in Python^[257]74.

SeqIF imaging using Lunaphore COMET platform

   To establish an antibody panel for SeqIF, we sourced high-quality
   antibodies from trusted vendors and performed test stainings at the
   vendor-recommended concentrations on the COMET priority access
   platform. If staining was too strong or weak, dilution curves were
   performed, and staining specificity was manually evaluated.
   Fluorescence signal acquisition in the Cy5 channel demonstrated higher
   signal-to-noise ratios and overall cleaner results, whereas the TRITC
   channel exhibited increased autofluorescence. Of the 16 antibodies
   initially selected, 12 were imaged in the Cy5 channel due to the
   predominance of functional antibodies raised in rabbit (Supplementary
   Table [258]2). The panel was finalized after 58 optimization runs,
   supported by the Lunaphore COMET priority access platform for
   evaluating intensity, sensitivity, elution efficacy, incubation time,
   antibody dilution, exposure time and cycle position. For the SeqIF
   stainings, samples were sectioned on a cryotome (8 μm) and collected on
   adhesion slides (Epredia SuperFrost Ultra Plus GOLD; Thermo Fisher
   Scientific) and dried on a 37 °C heat plate for 15 minutes. After
   storage at −80 °C, sections were brought to room temperature and were
   incubated in 4% formaldehyde for 40 minutes at room temperature.
   Samples were washed for 5 minutes at room temperature in Multistaining
   Buffer (Lunaphore Technologies, BU06), followed by incubation with
   Multistaining Buffer supplemented with 0.2% Triton for 20 minutes at
   room temperature. Subsequently, slides were stored in Multistaining
   Buffer until use. Slides were dried off and placed into COMET stainers
   with microfluidic chips positioned on top of the tissue section as
   described by Rivest et al.^[259]48. Antibody mixes were prepared by
   diluting the stock antibody solutions in Intercept-T20 to help with
   blocking of non-specific binding (Supplementary Table [260]2).
   Automated SeqIF staining and imaging were performed on the COMET
   platform (Lunaphore Technologies). Slides underwent 13 cycles of
   iterative staining and imaging, followed by elution of the primary and
   secondary antibodies^[261]75. The 16-plex protocol template was
   generated using COMET Control Software, and reagents were loaded onto
   the device to perform the SeqIF protocol. A list of primary antibodies
   with corresponding incubation times can be found in Supplementary Table
   [262]2. Secondary antibodies were used as a mix of two
   species-complementary antibodies: Alexa Fluor Plus 647 goat anti-rabbit
   (Thermo Fisher Scientific, cat no. A32733, 1:250 dilution) and Alexa
   Fluor Plus 555 goat anti-rat (Thermo Fisher Scientific, cat no. A48263,
   1:250 dilution). Nuclear signal was detected using DAPI (Thermo Fisher
   Scientific, cat no. D1306, 1:1,000 dilution) by dynamic incubation of
   2 minutes. All reagents, if not otherwise stated, were diluted in
   Multistaining Buffer (Lunaphore Technologies, BU06). The elution step
   lasted 2 minutes for each cycle and was performed with Elution Buffer
   (Lunaphore Technologies, BU07-L) at 37 °C. The quenching step lasted
   for 30 seconds and was performed with Quenching Buffer (Lunaphore
   Technologies, BU08-L). The imaging step was performed with Imaging
   Buffer (Lunaphore Technologies, BU09). The output from the COMET
   platform is a stitched and registered, multistack OME-TIFF file that
   was directly used for further processing with MCMICRO and downstream
   applications^[263]49.

Image processing and analysis of SeqIF (Lunaphore COMET data)

   Post-acquisition registration of full-slide images was needed for two
   images due to interrupted runs and was performed with Palom
   ([264]https://github.com/labsyspharm/palom). Processing of multichannel
   OME-TIFF files was performed using a modified version of
   MCMICRO^[265]49. Raw marker intensities were corrected for
   autofluorescence signal with Backsub
   ([266]https://github.com/schapirolabor/background_subtraction) by
   subtracting the respective autofluorescence image intensity scaled to
   each marker’s exposure time. The corrected intensity was computed using
   equation (1):
   Marker[corrected] = Marker[raw] − Background × Exposure[Marker] / Expos
   ure[Background]. Preprocessing of the images to improve segmentation
   was performed using CLAHE on the DAPI and membrane (WGA) channels. Cell
   segmentation was performed similarly to Molecular Cartography data
   using CLAHE-adjusted images to train a custom Cellpose 2 model via the
   human-in-the-loop approach. Feature quantification was performed on the
   autofluorescence-subtracted images based on the labeled segmentation
   masks. To assign cell phenotypes, we used a pixel-level clustering
   workflow using SOMs implemented via the Pixie pipeline^[267]50. For
   that purpose, we manually annotated the image region containing the
   heart using QuPath (0.4.3, MacOS version) and set all background pixels
   to 0 across all channels to reduce the number of pixels to be processed
   by Pixie^[268]72. We then performed pixel clustering with 10 functional
   markers: Ankrd1, αSMA, CCR2, CD31, CD45, CD68, Mpo, Pdgfra, Tnnt2 and
   Trem2 across all nine COMET images using 5% of pixel subsets to train
   the SOM. Pixel meta-clusters were visualized using the Pixie Jupyter
   widget, and 100 SOM clusters were manually merged and visualized as
   pixel phenotype maps for validation. Pixel clusters were subsequently
   used in a second clustering step with Cellpose masks to assign cell
   phenotypes via SOM clustering across all images. Similar to pixel
   clustering, 100 SOM clusters were manually merged into cell phenotypes,
   and each cell segmentation mask was assigned a phenotype. For
   phenotyping accuracy evaluation, we compared assigned cell phenotypes
   to an independently annotated subset of ground truth data and
   quantified performance using confusion matrices (raw and normalized) as
   well as per-class precision, recall and F1 scores (Supplementary Fig.
   [269]10). To quantify cell type abundances in different regions of the
   heart (endocardial infarct zone, epicardial infarct zone, infarct core
   and border zones), we manually annotated regions in QuPath and used
   them to subset cells based on their presence or absence in the
   annotation. For the quantification of myeloid cells in the endocardial
   infarct zone, we included cells in the left ventricular lumen, only if
   they were in direct contact with endocardial cells. Endothelial cells
   in the manually annotated endocardial layer were assigned as
   ‘Endocardial cells’. The endocardial infarct zone was annotated as the
   endocardial layer and the adjacent 2–3-cell-thick layer of
   cardiomyocytes expressing Tnnt2 near the infarct core, not including
   papillary muscles. The infarct core was defined based on heart geometry
   in the region where the infarct was transmural. The epicardial infarct
   zone was annotated as the epicardial layer and a 3–4-cell-thick layer
   along the infarct core. The border zone was annotated as the border
   region of the infarct including both stressed and Tnnt2-expressing
   cardiomyocytes. The remote endocardial region was annotated as the
   endocardial layer and a 3–4-cell-thick region on the opposite side of
   the lumen to the infarct core with no nearby stressed cardiomyocytes,
   and the same applies for the control samples. The areas outside the
   tissue samples are denoted as ‘Background’, and the areas within the
   left ventricular lumen were annotated as ‘Lumen’. Matching the size
   thresholds adapted for pixel size from Molecular Cartography data,
   cells with an area below 72 and above 72,000 pixels were excluded.
   Additionally, cells were filtered based on small size, low solidity and
   high eccentricity. Based on the distance of each cell to the lumen,
   cells were binned into strips with 18.4-μm thickness (Extended Data
   Fig. [270]6). A Minerva Story was created using Minerva Author for a
   representative sample 24 hours after MI^[271]76,[272]77.

Segmentation method comparison

   To create an evaluation of used segmentation models DeepCell Mesmer
   ‘nuclear’ and ‘whole-cell’, Cellpose ‘cyto’ and our custom-trained
   Cellpose 2 model, we selected an additional 16 crops (500 × 500 pixels)
   spanning four representative regions per timepoint and independently
   annotated the ground truth cells based on the CLAHE-adjusted nuclear
   and membrane markers using the napari
   toolkit^[273]33,[274]34,[275]42,[276]63. The crops included a total of
   1,208 ground truth annotations ranging from 37 to 112 cells per crop,
   with total counts of cells per timepoint ranging from 253 to 357.

   Segmentation evaluation was performed on a crop-level basis using an
   object-based approach as described in Greenwald et al.^[277]33. We
   constructed the cost matrix as 1 − Intersection over Union (IoU) for
   linear sum assignment. The cost matrix was padded to allow for
   unassigned cells with a penalty of 0.5. If an assigned ground truth and
   prediction pair had an IoU above 0.4, it was counted as a true-positive
   match. Ground truth cells without matches to prediction cells were
   labeled false negatives, and prediction cells without matches to ground
   truth cells were labeled false positives. Unmatched ground truth and
   prediction cells were used as nodes in a graph with edges between them
   if they had an IoU above 0.1 to account for tissue density. According
   to Greenwald et al., merges are defined as events where multiple ground
   truth cells are connected to one prediction cell; splits are defined as
   events where one ground truth cell is connected to multiple prediction
   cells; and catastrophes are defined as events where multiple ground
   truth cells are connected to multiple prediction cells. An example
   image (crop of sample_d1.r1 of the SeqIF dataset) is provided to
   highlight observed segmentation errors (Extended Data Fig. [278]2a).

   Across the annotated 1,208 ground truth cells, DeepCell Mesmer
   ‘nuclear’ had 353 true positives, 400 false positives, 855 false
   negatives, 12 merges, nine splits and two catastrophes. DeepCell Mesmer
   ‘whole-cell’ had 567 true positives, 515 false positives, 641 false
   negatives, 49 merges, 31 splits and 38 catastrophes. Cellpose ‘cyto’
   had 521 true positives, 194 false positives, 687 false negatives, 28
   merges, three splits and five catastrophes, and our custom Cellpose
   model had 1,124 true positives, 124 false positives, 84 false
   negatives, four merges, six splits and two catastrophes (Supplementary
   Fig. [279]3b). IoU and Dice scores were measured for each true-positive
   pair, and the mean value was calculated per image crop. Precision,
   recall and F1 score were also calculated on a per-crop basis and
   highlighted our custom model outperforming the other methods on this
   specific dataset (Extended Data Fig. [280]2c).

   Additionally, we provide the segmented percentage of the tissue in
   SeqIF data calculated as the percentage of recovered tissue not within
   ‘Background’ or ‘Lumen’ annotations. We also provide the percentage of
   assigned transcripts in the Molecular Cartography data. Both metrics
   highlight that our custom model recovers the most relevant tissue
   information (Extended Data Fig. [281]2d). Taken together, using our
   custom Cellpose models (for SeqIF and Molecular Cartography), we were
   able to recover more tissue area than we could with existing models
   without sacrificing accuracy on this specific dataset.

Laser microdissection coupled to ultrasensitive proteomics

   To collect cells with a Leica Laser Microdissection 7 (LMD7)
   microscope, three reference points are required to triangulate the
   shape coordinates into laser cutting coordinates. These reference
   points were etched using the LMD7 on the membrane of the Leica Frame
   slides (order no. 11600294) before the tissues were placed on them.
   These etchings were easily recognizable features to which we could go
   back and designate them as reference points. Five-micrometer-thick
   tissue sections from fresh-frozen heart tissue were prepared similarly
   to the SeqIF samples and were placed onto Leica Frame slides with
   reference points. Slides were stained using DAPI, WGA and CD31 (as
   described in the immunofluorescence subsection) and imaged using a
   Zeiss Axioscan 7. Stitched images of whole hearts were imported and
   annotated in QuPath for downstream laser capture microdissection.
   Endocardial regions to be collected by the LMD7 were annotated using
   QuPath’s brush tool with a brush diameter of approximately 20 µm,
   centered around endocardial cells with 10 µm to each side as buffer for
   the laser cutting. For control hearts, only one endocardial group was
   annotated, whereas, for hearts, 1-day post-infarct endocardial cells in
   the infarct region and endocardial cells in the remote region were
   labeled separately. QuPath annotations were then exported as GeoJSON
   files, which were further processed using the Qupath_to_LMD scripts
   ([282]https://github.com/CosciaLab/Qupath_to_LMD). The script assigns
   annotation classes to wells of the 384-well plate. It uses the py-lmd
   package from Madler et al. ([283]https://github.com/MannLabs/py-lmd) to
   transfer GeoJSON polygons into LMD readable .xml files.

Laser capture microdissection

   We used the Leica LMD7 system and Leica Laser Microdissection version
   8.3.0.08259 software to collect tissue contours. Tissue was cut with a
   ×63 objective (HC PL FLUOTAR L ×63/0.70 CORR XT) in brightfield mode.
   Laser settings were as follows: power 60; aperture 1; speed 25; middle
   pulse count 2; final pulse 5; head current 39–41%; pulse frequency
   2,028; and offset 105. Contours were cut and sorted into a low-binding
   384-well plate (Eppendorf, 0030129547) configured over the ‘universal
   holder’ function.

Sample preparation for liquid chromatography–mass spectrometry analysis

   To collect tissue pieces stuck on the sides of the 384 wells, we washed
   them down with 30 µl of acetonitrile, briefly vortexed and vacuum dried
   (15 minutes at 60 °C). We added 2 µl of Lysis Buffer (0.1% DDM, 5 mM
   TCEP, 20 mM CAA resuspended in 100 mM TEAB (pH 8.5)) to each well,
   closed the plate with a PCR ComfortLid (Hamilton) and heated at 95 °C
   for 60 minutes. We added 1 µl of LysC (2 ng µl^−1 in water) and
   incubated for at least 2 hours at 37 °C. Subsequently, 1 µl of trypsin
   was added (1 ng µl^−1 in water), and the samples were incubated
   overnight at 37 °C. The next day, the samples were vacuum dried before
   peptide cleanup. Peptide cleanup took place with Evotips (Evosep)
   following the manufacturer’s recommendations. In brief, the Evotips
   (EV2013, Evotip Pure) were washed with Buffer B (99.9% acetonitrile,
   0.1% formic acid) and then Buffer A (99.9% water, 0.1% formic acid) and
   then activated with isopropanol. Digested tissue samples were
   resuspended in Buffer A, loaded into the tips, washed with Buffer A
   once and then eluted with Buffer B into a 96-well plate (Thermo Fisher
   Scientific, AB1300) and vacuum dried. Samples were stored at −20 °C
   until liquid chromatography–mass spectrometry (LC–MS) analysis. For
   LC–MS analysis, 4.2 µl of MS loading buffer (3% acetonitrile, 0.1%
   trifluoroacetic acid in water) was added, from which 4.0 µl was finally
   injected into the mass spectrometer.

LC–MS analysis

   LC–MS analysis was performed with an EASYnLC-1200 system (Thermo Fisher
   Scientific) connected to a trapped ion mobility spectrometry quadruple
   time-of-flight mass spectrometer (timsTOF SCP; Bruker Daltonik) with a
   nano-electrospray ion source (CaptiveSpray; Bruker Daltonik). Peptides
   were loaded on a 20-cm home-packed high-performance liquid
   chromatography column (75-µm inner diameter packed with 1.9-µm
   ReproSil-Pur C18-AQ silica beads; Dr. Maisch). Peptides were separated
   using a linear gradient of 21 minutes and analyzed in dia-PASEF mode.

Proteomics data analysis

   We used DIA-NN (1.8.2) for dia-PASEF raw file analysis, and the
   generated libraries were used for mouse proteins (UniProt mouse
   released in 2021) and known contaminants^[284]78. Deep-learning-based
   spectra, retention times and ion mobility predictions were enabled for
   the appropriate mass range of 300–1,200 m/z. N-terminal M excision and
   cysteine carbamidomethylation were enabled as fixed modifications. A
   maximum of two miscleavages were allowed, and the precursor charge was
   set to 2–4. DIA-NN was operated in the default mode with minor
   adjustments. In brief, MS1 and MS2 accuracies were set to 15.0; scan
   windows were set to 0 (assignment by DIA-NN); and isotopologues were
   enabled, as were matched-between-runs, heuristic protein inference and
   no shared spectra. Proteins were inferred from genes; neural network
   classifiers were set to single-pass mode; and quantification strategy
   was set as ‘Robust LC (high precision)’. Cross-run normalization was
   set to ‘RT-dependent’, library generation as ‘smart profiling’ and
   speed and RAM usage as ‘optimal results’. Protein lists were filtered
   for missing values by group, requiring at least two observed values in
   the control group or three observed values in the MI remote or MI IZ
   group. We also filtered out known contaminants based on previously
   described contaminants libraries^[285]78. Differential protein
   expression analysis was performed on data normalized using variance
   stabilizing normalization using empirical Bayes statistics in
   limma^[286]79. Proteins with a false discovery rate lower than 0.05
   were considered as significantly differentially expressed. Overlap in
   the proteins differentially expressed between conditions was visualized
   using UpSet plots with the ComplexUpset package^[287]80. Endocardial
   specificity of DEPs was compared by correlating log fold changes from
   DEP analysis with log fold changes from marker gene estimations
   (Seurat’s FindMarkers function) for endocardial cells from sham
   controls from the reprocessed Calcagno et al. ^[288]22 dataset (see the
   ‘Single-cell analysis of Molecular Cartography data’ subsection).

Analysis of human CITE–seq data

   Processed CITE–seq data from Amrute et al.^[289]52 were received from
   the original authors as a processed Seurat object (version
   4.0.0)^[290]52. Cell type annotations from the original authors were
   used to visualize cell types on the UMAP plot. Differential expression
   of vWF between donor and acute MI samples was performed by calculating
   pseudobulk expression per sample and performing DESeq analysis between
   pseudobulk expression of donor (n = 6) and acute MI (n = 4) samples.
   Normalized RNA expression was visualized per group using violin plots
   from the SCpubr package^[291]81.

Echocardiography

   Echocardiographic analyses were performed in conscious mice using a
   Vevo 2100 ultrasound system (VisualSonics). Left ventricular
   end-diastolic volume, end-systolic volume and ejection fraction were
   measured based on the left parasternal long-axis view and were acquired
   using VevoLab software (VisualSonics). Global longitudinal strain was
   quantified in the longitudinal axis by speckle tracking using
   VevoStrain software (VisualSonics). Investigators were blinded to the
   sample group allocation during the experiments and analyses.

Flow cytometry

   Single-cell suspensions of infarcted hearts were obtained by mincing
   the tissue with fine scissors and digesting it with a solution
   containing 450 U ml^−1 Collagenase I, 125 U ml^−1 Collagenase XI,
   60 U ml^−1 DNase I and 60 U ml^−1 hyaluronidase (MilliporeSigma) for
   1 hour at 37 °C while shaking. For flow cytometry of blood samples,
   erythrocytes were lysed in red blood cell lysis buffer (Miltenyi
   Biotec). The fluorescent antibodies are described in Supplementary
   Table [292]3. Flow cytometry was performed on a FACSVerse (BD
   Biosciences). Data were analyzed using FlowJo software. Mo/Mɸ were
   identified as CD45^+, Lin^−(CD19;CD4;NK1.1;Ly6G;Ter119) and CD11b^+.

Histology

   Histopathological evaluation of left ventricular remodeling was
   performed on day 14 after MI induction. Hearts were excised and rinsed
   in PBS. After transverse sectioning using a scalpel, hearts were then
   embedded in OCT compound and placed in 2-methylbutane (Honeywell) on
   dry ice. Hearts were stored overnight at −80 °C and sectioned using a
   cryostat (9-μm thickness). Tissue sections were stained with a Masson’s
   Trichrome Stain Kit (MilliporeSigma) according to the manufacturer’s
   instructions. Scar thickness was averaged from five measurements in
   short axes in a blinded fashion.

Conventional immunofluorescence stainings

   For conventional immunofluorescence staining, samples were sectioned on
   a cryotome (8 μm), collected on adhesion slides (Epredia SuperFrost
   Ultra Plus GOLD; Thermo Fisher Scientific) and dried on a 37 °C hot
   plate for 15 minutes. After storage at −80 °C, sections were brought to
   room temperature and incubated in 4% formaldehyde at room temperature
   for 40 minutes. Sections were permeabilized for 20 minutes, blocked
   with 5% BSA for 1 hour and stained overnight with primary antibodies
   for CD31, vWF, CCR2, CD68 or CD41 in 1% BSA staining buffer. On the
   following day, sections were washed and stained for 1 hour with the
   corresponding secondary antibodies combined with the labeled antibody
   against WGA. After washing, sections were stained with 300 nM DAPI
   (Thermo Fisher Scientific, D1306) for 10 minutes, washed again and
   covered with a mounting medium. For simultaneous staining of CCR2 and
   vWF, sections were fixed, permeabilized and blocked as described above.
   After incubating primary antibodies for CD31 and CCR2 overnight, vWF
   antibody was labeled using a secondary antibody labeling kit (FlexAble
   CoraLite Plus 750 Antibody Labeling Kit for Rabbit IgG; Proteintech).
   Labeling was carried out in accordance with the manufacturer’s
   instructions. In brief, vWF primary antibody was incubated with
   FlexLinker and FlexBuffer for 5 minutes. FlexQuencher was added and
   incubated for five additional minutes. Then, 1% BSA staining buffer was
   added, and tissue sections were incubated with the labeled primary
   antibody. On the next day, sections were washed and stained with WGA
   conjugated to Alexa Fluor 488 (Thermo Fisher Scientific, [293]W11261)
   for 1 hour. After the WGA incubation time, sections were stained with
   300 nM DAPI (Thermo Fisher Scientific, D1306) for a duration of
   10 minutes, washed again and covered with a mounting medium. Images
   were captured using an Axio Observer (Zeiss) fluorescence microscope
   and analyzed using QuPath (0.4.3, Windows version). In brief, cell
   segmentation was performed by using the cell detection tool based on
   DAPI staining. Mean cell intensity was used to define cells positive
   for Ccr2 (threshold 450) and Cd68 (threshold 400). An overview of
   antibodies used for conventional immunofluorescence stainings can be
   found in Supplementary Table [294]4.

Statistics and reproducibility

   Mice were randomly assigned to different experimental groups. No
   exclusion of specific animals from the experiments was performed. All
   experiments were conducted on independent biological replicates:
   Molecular Cartography (two replicates per timepoint; technical
   replicates were only done to replace some initial samples that failed
   quality control and test correlation of measurements across
   slides/technical replicates); SeqIF (three controls; two replicates at
   4 hours, 24 hours and 2 days after MI); and DVP (3–4 replicates per
   region). Human pseudobulk CITE–seq analyses included six healthy donors
   and four patients with acute MI from publicly available data.
   Functional blocking of vWF employed eight mice per experimental group.
   Sample size calculation was performed in G*Power 3.1 in accordance with
   the approved animal protocol and was based on our experience with
   similar experimental studies to achieve 80% power at a significance
   level of P < 0.05. Echocardiographic and histological analyses were
   performed in a blinded fashion.

   Quantitative data are presented as mean ± s.d. Spatial distance
   measurements between cell types were compared by type II ANOVA on a
   linear OLS model, followed by two-sided post hoc t-tests with
   Bonferroni correction. Differential protein expression in endocardial
   regions was assessed with limma, employing variance stabilizing
   normalization and empirical Bayes statistics, with a false discovery
   rate of less than 0.05 defining significance. Human pseudobulk
   differential gene expression was determined by DESeq2. For functional
   measurements, comparisons between two groups were performed using
   unpaired two-tailed Student’s t-test. Differences between more than two
   groups were analyzed by one-way ANOVA followed by Tukey’s post hoc
   analysis or two-way ANOVA followed by Sidak’s multiple comparison.
   Unless stated otherwise, P < 0.05 was considered statistically
   significant.

Reporting summary

   Further information on research design is available in the [295]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [296]Supplementary Information^ (1.7MB, pdf)

   Supplementary Figs. 1–10
   [297]Reporting Summary^ (2.5MB, pdf)
   [298]Supplementary Tables 1–4^ (20.1KB, xlsx)

   Supplementary Tables 1–4

Source data

   [299]Source Data Fig. 1^ (159.7KB, xlsx)

   Spatial coordinates of Molecular Cartography sample in Fig. 1d.
   [300]Source Data Fig. 2^ (56.8KB, xlsx)

   MISTy spatial cell type relationships and Euclidean distances between
   endocardial cells and myeloid cells and between cardiac fibroblasts and
   myeloid cells.
   [301]Source Data Fig. 3^ (54KB, xlsx)

   Marker expression for Pixie pixel clusters, endocardial to myeloid cell
   distances and myeloid cell densities.
   [302]Source Data Fig. 5^ (428KB, xlsx)

   Statistical data for laser capture microdissection high-sensitivity
   proteomics results.
   [303]Source Data Fig. 6^ (53.1KB, xlsx)

   Functional data from anti-vWF blocking experiments: FACS mean
   fluorescence intensity data, conventional immunofluorescence signal
   quantification and quantitative murine cardiac readouts.
   [304]Source Data Extended Data Fig. 2/Table 2^ (12.5KB, xlsx)

   Cell segmentation measures.
   [305]Source Data Extended Data Fig. 3/Table 3^ (50KB, xlsx)

   Cell type composition changes across MI timeline.
   [306]Source Data Extended Data Fig. 4/Table 4^ (85.4KB, xlsx)

   Pixie pixel cluster heatmap values and Pixel cluster changes across MI
   timeline.
   [307]Source Data Extended Data Fig. 5/Table 5^ (49.5KB, xlsx)

   Relative and absolute cell type number changes based on conventional
   immunofluorescence measurements.
   [308]Source Data Extended Data Fig. 6/Table 6^ (44KB, xlsx)

   Relative cell abundance changes in distance bins to cardiac lumen.
   [309]Source Data Extended Data Fig. 7/Table 7^ (82.8KB, xlsx)

   vWF and Ccr2 quantification after infarction and ischemia/reperfusion
   in both female and male mice.

Acknowledgements