Cohort summary:
The study cohort is summarized in Table 1. Briefly, the baseline characteristics of the CS cohort differed from the Control cohort in the proportion of female patients (13% for CS group vs. 50% for controls, p = 0.012), and in the average LVEF at tissue sampling (26.5% for CS group vs. 39.7% for controls, p = 0.03). There were no significant differences among baseline characteristics of CS patients contributing earlier-disease-stage biopsy samples to the cohort vs. those contributing advanced-stage disease samples.
The GeoMx workflow was deployed across the entire 48 sample cohort, generating 305 ROIs which yielded a total of 521 AOIs after segmentation of tissue compartments in cardiac parenchyma ROIs. For final analyses, there were n = 39 granuloma ROIs, n = 227 non-granulomatous cardiac parenchyma ROIs (from which myocyte and stroma AOIs were segmented), and n = 39 large vascular bed ROIs (from which vascular AOIs were derived – see Fig. 1). Among parenchymal tissue (non-granuloma) ROIs in the CS cohort, there were n = 62 ROIs from EMB tissues (n = 8 peri-granuloma parenchymal ROIs, n = 54 granuloma-remote parenchymal ROIs), and n = 161 from advanced-stage explant tissues (n = 19 peri-granuloma parenchymal ROIs, and n = 113 granuloma-remote parenchymal ROIs, and n = 29 large vascular bed ROIs). There were n = 35 ‘inflammation(+)’ parenchymal ROIs in the CS cohort which, despite not containing granulomas, had overt histologic immune cell infiltration. There were n = 43 ROIs from Control tissues, including n = 33 parenchymal ROIs and n = 10 large vascular bed ROIs.
Spatial protein expression results demonstrate substantial CS granuloma heterogeneity, both within and between patients:
For most study panel proteins (49/79, 62%), the variance in protein expression between granuloma-containing ROIs was higher than the variance between non-granuloma parenchymal ROIs. This is notable, considering parenchymal ROIs consist of stroma and myocytes from hearts with widely varying LVEFs and which, in some cases, contain overt interstitial inflammatory cell infiltrates. The 10 most variably expressed proteins among granuloma ROIs are predominantly immune cell-type markers for macrophages, T-cells, granulocytes and antigen presenting cells: CD68, HLA-DR, CD11c, CD45, CD3, IDO1, CD44, CD40, CD66b, and BCL6.
There were only modest differences in the granuloma protein expression between ‘advanced-stage disease’ tissues and EMB tissues obtained earlier in the disease course. Immune checkpoint molecule Tim-3 and activated fibroblast marker FAP-alpha have significantly increased expression in advanced-stage CS hearts (p = 0.02 and p = 0.017, respectively). In contrast, immune checkpoint molecule VISTA, nuclear/proliferation marker Histone-H3, and activated MEK1 (part of the RAF/MEK/ERK pathway known to be involved in inflammation and linked to granuloma formation when inhibited)25,26 each had significantly increased expression in earlier/active stage CS (p = 0.007, p = 0.013, and p = 0.017 respectively).
We performed a sub-analysis of the nine CS tissue samples which contributed multiple granuloma ROIs to the dataset. The PCA biplot in Supplemental Figure S.1 suggests substantial intra-sample heterogeneity in CS granulomas, with only modest within-sample groupings. Nearest-neighbor analysis of the PCA plot demonstrates that for cases contributing multiple granuloma ROIs, the closest-clustering granuloma ROI is more likely to be from a different tissue sample than from the same sample (13/23, 56.5%). At the individual protein level, the intra-sample coefficient of variation (COV) exceeded inter-sample COV for an average of 11 protein markers in each multi-granuloma sample, with the most substantial intra-sample variation seen in T-cell and cytotoxic cell markers (CD27, CD8, GZMB), apoptotic markers (CD95/FAS, Cleaved-Caspase-9), and checkpoint molecules (PDL2) (supplemental Table S.3).
Analysis of the Cardiac Parenchyma Highlights the Protein Expression Profiles of Active and ‘Burnt Out’ CS:
Analysis of intrinsic cardiac parenchymal ROIs (e.g. non-granuloma ROIs) demonstrates substantial differences in the expression of immunologic, cell survival, and cell death pathways between EMB tissue samples obtained during the active workup/management phase of CS and advanced-stage disease tissue samples obtained at cardiectomy. These differences are readily apparent via unsupervised data visualization with PCA and t-SNE plots in Fig. 2a. When analyzed as non-compartmentalized ROIs, n = 33 proteins show significant differential expression based on disease stage (Fig. 2b and Supplemental Table S.4). When segmented as compartment-specific AOIs, there are n = 24 significant DEPs in the myocyte compartment, and n = 27 DEPs in the stroma compartment.
While substantial overlap exists between DEPs in the ROI- and AOI-level analyses, AOI analyses add important context. We observed significantly increased pro-apoptotic factors and decreased MAP-kinase and PI3K/AKT pathway activity in the myocyte compartment AOIs of advanced-stage hearts – a finding consistent with prior research on end-stage cardiomyopathy more generally.27,28 In stroma compartment AOIs of advanced-stage hearts, we observed significant increases in markers of activated/differentiated fibroblasts (FAP-alpha and SMA). Again, this is consistent with known advanced-stage cardiomyopathy biology.29 However, we also observed numerous significant shifts in expression of immune-related protein markers in the stroma and myocyte AOIs when comparing tissues acquired earlier vs. later in disease which are not as easily explained.
in advanced-stage CS, there was a significant decrease in several macrophage and effector T-cell lineage markers (CD3, CD4, CD163), which coincided with significant increases in markers of longer-lasting regulatory T-cell (Treg), memory T-cell and B-cell populations (ICOS, FOXP3, CD45RO, CD127, CD20). In addition, the stroma and myocytes of advanced-stage cases manifest a less ‘immune primed’ state, with decreased expression of major histocompatibility (MHC) molecule HLA-DR, checkpoint molecules PD-L1/PD-L2, and interferon-producing STING. Overall, these findings are consistent with the theoretical biology of the ‘late fibrous phase’ stage of CS (sometimes called ‘burnt out’ CS),30 which is thought to involve increased fibrosis along with a decrease in active inflammatory elements.30 Our findings support long-standing theories about this process, highlighting numerous key immune cell types and effectors which change as CS progresses.
Examining the ‘Distance-Gradient’ of Cardiac Parenchymal Biology in CS:
As shown in Fig. 3 and Supplemental Table S.5, spatial analysis of the CS parenchyma reveals a previously unreported ‘distance-gradient’ in protein expression, in which numerous panel proteins were differentially expressed based on a tissue region’s relative distance from granulomas. This distance gradient is apparent with unbiased data visualization via PCA and t-SNE (Fig. 3a), and persists even when accounting for confounders like histologic inflammatory infiltrates during differential expression testing with mixed effects models (Fig. 3b).
We also examined ‘inflammation(+)’ ROIs which do have discrete, non-granulomatous, interstitial inflammatory cell infiltrates, comparing these extreme examples of ‘proximity to inflammation’ to ‘inflammation(-)’ ROIs without any discrete inflammation. Unsurprisingly, there were many significant DEPs between these groups, including increased expression of numerous immune effector cell markers: CD3, CD4, CD8, CD68, CD163, GZMA, CD14, and CD45 (Supplemental Table S.6). Interestingly, at the ROI-level, 70% (12/17) of the significant DEPs which were observed during our inflammation-adjusted comparison of peri-granuloma ROIs to granuloma-remote ROIs are also significant DEPs when comparing overt ‘inflammation(+)’ ROIs to ‘inflammation(-)’ ROIs. However, these overlapping DEPs are largely not classic immune-effector cell markers, and instead suggest that the distance-gradient observed in this study arises from subtler findings of immune activity.
Expression of most specific immune cell-types do not differ between peri-granuloma and granuloma-remote regions. However, peri-granuloma stroma does have a larger population of total immune cells (CD45+). This is primarily due to significant increases in long-lasting immune ‘sentinels’ in the form of CD11c + dendritic cells and CD45RO + memory T-cells, rather than to increases in classic effector cell-types such as those found in overt ‘inflammation(+)’ parenchyma. Beyond specific immune cell-type markers, peri-granuloma stroma exhibits increased expression of class I and II MHC molecules (Beta-2-microglobulin and HLAD-DR) – a phenomenon known to occur in association with myocardial inflammation.31–35 Peri-granuloma stroma also exhibits increased expression of inflammation-associated pro-fibrotic mediators such as arginase 1,36,37, fibronectin,38 and CD4439,40. Interestingly, peri-granuloma stromal cells have decreased expression of immune checkpoint molecule Tim-3, suggesting that that peri-granuloma lymphocytes may be less responsive to immune-checkpoint-mediated inhibition.41 Finally, distance-dependent protein expression also impacts cardiomyocyte biology. Peri-granuloma myocytes manifest a ‘stress-activated’ state,42 with increases in inflammation-associated class I/II MHC molecules, CD40,43 and immune checkpoint PD-L144 along with increased injury-repair, fibrosis, and stiffness-associated CD4445–47 and fibronectin (each of which likely co-localizes with cardiomyocytes rather than being expressed by them).38
The Protein Expression Profile of ‘Granuloma Remote’ CS Parenchymal Tissue Differs from that of Controls:
A fundamental question at the outset of this research was whether tissue biomarkers of CS exist which can be detected even when no histologic evidence of CS is present. As shown in Fig. 4a and Supplemental Table S.7, we identified numerous significant DEPs between granuloma-remote, inflammation(-), CS tissue and tissue from non-CS controls. Compared to control samples, granuloma-remote CS parenchyma is characterized by significantly increased expression of HLA-DR, Treg markers FOXP3, CD25 and GITR, endothelial/stem-cell marker CD34, and global nuclei/proliferation marker Histone-H3. CS parenchyma manifests decreased expression of CD45RO, PDL2, apoptosis marker CD95/FAS, and inactivated (phosphorylated) GSK3β and GSK3α (from which we infer increased activated GSK3 enzyme activity with resultant NF-κB-mediated pro-inflammatory cytokine production).48,49
To place this finding in the context of the ‘distance-gradient’ results described in the previous section, we performed ordinal logistic regression, treating peri-granuloma regions, granuloma-remote regions, and Control regions as ordinal classes representing different degrees of distance from granulomas. Interestingly, we observed that 65.3% of study panel proteins (n = 49) demonstrated a significant change in expression with increasing distance from CS inflammation (Supplemental Table S.8). Taken together, these results demonstrate both the local impact of granuloma proximity on protein expression as well as the more organ-wide impact of CS on tissue protein expression. Figure 4b provides a further, visual, demonstration of this phenomenon, highlighting the change in expression among several key groups of protein markers when moving from regions of inflammation(+) CS parenchyma to inflammation-free CS parenchyma and finally to control cardiac parenchyma.
A limited, pre-specified sub-analysis of ROIs derived from larger vascular beds was performed to assess whether the vasculature in CS differs from controls. Overall, while ROI numbers were limited for this analysis (n = 29 from CS cases, n = 10 from controls), the results suggest an immunologically active environment in CS vascular beds relative to controls, with increased expression of CD3, CD4, CD68, VISTA, CD45, HLA-DR, and CD11c (Supplemental Table S.9).
Spatial Protein Expression Biomarkers Enable Accurate Prediction of Occult CS:
To assess whether the various DEPs between areas of inflammation(-) CS parenchyma and Controls could have diagnostic value, we developed a binary prediction model to classify parenchyma tissue ROIs as originating from CS vs. Control patients. To maximize clinical utility as a tool capable of improving the diagnostic yield of tissue sampling in CS, the model was specifically developed using only data from CS ROIs which were ‘granuloma-remote’ and ‘inflammation(-).’ As shown in Fig. 5, after optimization via leave-one-out cross validation, our final 7-variable logistic regression model was comprised of MHC molecule HLADR, Treg markers FOXP3, CD25, and GITR, immunomodulatory checkpoint molecule VISTA, natural-killer cell marker CD56, and global nuclei/cell proliferation marker Histone-H3. The model achieved excellent performance in the ‘held-out’ validation set, with an accuracy of 90.0%, AUROC of 0.92, sensitivity of 89.7% and specificity of 90.9%. Given that it is also possible to sample peri-granuloma regions during a clinical EMB procedure while still ‘missing’ a granuloma, we also assessed performance of the final model on inflammation(-), peri-granuloma regions. Performance was excellent on these as well, achieving accuracy of 90% (18/20).