Abstract
Tissue inhibitor of metalloproteinases (TIMPs/Timps) are an endogenous family of widely expressed matrisome-associated proteins that were initially identified as inhibitors of matrix metalloproteinase activity (Metzincin family proteases). Consequently, TIMPs are often considered simply as protease inhibitors by many investigators. However, an evolving list of new metalloproteinase-independent functions for TIMP family members suggests that this concept is outdated. These novel TIMP functions include direct agonism/antagonism of multiple transmembrane receptors, as well as interactions with matrisome targets. While the family was fully identified over 2 decades ago, there has yet to be an in-depth study describing the expression of TIMPs in normal tissues of adult mammals. An understanding of the tissues and cell-types that express TIMPs 1 through 4, in both normal and disease states are important to contextualize the growing functional capabilities of TIMP proteins, which are often dismissed as non-canonical. Using publicly available single cell RNA sequencing data from the Tabula Muris Consortium, we analyzed approximately 100,000 murine cells across nineteen tissues from non-diseased organs, representing seventy-three annotated cell types, to define the diversity in Timp gene expression across healthy tissues. We describe that Timp genes display unique expression profiles across tissues and organ-specific cell types. Within annotated cell-types, we identify clear and discrete cluster-specific patterns of Timp expression, particularly in cells of stromal and endothelial origins. Differential expression and gene set pathway analysis provide evidence of the biological significance of Timp expression in these identified cell sub-types, which are consistent with novel roles in normal tissue homeostasis and changing roles in disease progression. This understanding of the tissues, specific cell types and conditions of the microenvironment in which Timp genes are expressed adds important physiological context to the growing array of novel TIMP protein functions.
Introduction
Tissue inhibitors of metalloproteinases (TIMPs/Timps) are named so due to their principally ascribed function as inhibitors of extracellular matrix proteolytic activity, namely Metzincin proteases of the Matrix Metalloproteinase (MMP), A Disintegrin and Metalloproteinase (ADAM) and ADAM with thrombospondin motifs (ADAMS-TS) families. The Metzincins are major regulators of extracellular matrix (ECM) structure and composition [1]. As a result, TIMPs are routinely designated solely as tissue inhibitors of metalloproteinases. Although their metalloproteinase inhibitory functions in many disease states are well recognized, TIMPs are also widely expressed in non-diseased tissues with little detectable Metzincin protein expression and/or activity [2-4]. TIMPs display a broad targeting of MMPs, excluding TIMP1 which shows poor affinity for the six membrane-associated MMPs [4]. Inhibition of ADAM and ADAM-TS proteinases is more selective between the TIMP family members, with TIMP3 displaying the broadest inhibitory capabilities in this regard [5, 6]. Since their original discovery additional functions have been identified in TIMPs that are independent of their metalloproteinase-inhibitory activities, including direct agonism/antagonism of cell-surface receptors and interactions with matrisome partners [7, 8]. Although it has been over two decades since all four members of the TIMP family were identified, there remains to be an in-depth analysis of TIMP gene expression in organs, tissues and cells of normal adult mammals. An understanding of the tissues and cell types that express TIMPs is important to add context to the ever broadening and increasingly complex functions of TIMPs that are all too often dismissed as non-canonical.
The advent of single cell RNA (scRNA) sequencing has greatly expanded researcher’s abilities to profile gene expression at the cellular level in various tissues. This was perfectly illustrated by the Tabula Muris Consortium’s breakout manuscript presenting whole organism scRNA sequencing data in mice [9]. We have utilized this rich source of data, analyzing approximately 100,000 murine cells, isolated by fluorescence-activated cell sorting (FACS) or microfluidic partitioning (droplet-based), for the expression of the Timp family of genes. We describe the diversity in Timp gene expression across nineteen murine tissues and seventy-three annotated cell types, providing new insights into the proportion of expressing cells and the degree of expression in specific cell populations. We identify several discrete cell types, particularly of mesenchymal and endothelial origins, with unique gene expression profiles that correlate with distinct expression of specific Timp family genes. Our detailed analysis of the tissues, cell-types and microenvironmental conditions in which Timps are expressed, adds significant physiological context and enhances our understanding of both protease inhibitor activity and metalloproteinase-independent functions of TIMP proteins in their resident tissues.
Results
Tissue expression of Timp family genes in murine organs
We first set out to describe Timp gene expression across the 17 organs analyzed in the Tabula Muris dataset in a simple format. Normalized expression of the 97,420 analyzed cells were subject to an arbitrary low count cut-off expression value of 0.5, selected to ensure that at least 80% of the cells with detectable expression of Timps1-4 were included. Cells above the cut-off expression levels were marked as positive. Additionally, of these positive cells, the mean expression levels were calculated. These data were incorporated into a circle packing graph describing Timp family expression profiles in whole organs (Figure 1A). Our analysis of all four Timp family members demonstrates Timp2 expression is predominant, followed closely by Timp3. Timp1 and Timp4 display restricted expression across the analyzed organs, present above threshold in only 12% and 5% of total analyzed cells, respectively (Figure 1A). Mean Timp expression values in organs, calculated by excluding cells below threshold expression, generally correlate with the proportion of Timp expressing cells. The complete tabulated data sets describing Figure 1A can be found in Table S1.
It is important to appreciate that gene expression often poorly correlates with protein levels, which is a consequence of numerous levels of regulation that follow transcription (e.g. post transcriptional regulation by miRNA, post-translational modifications). Despite a lack of correlation between RNA and protein levels, a recent study from the Genotype-Tissue Expression (GTEx) consortium identified that TIMPs 1-3 displayed good correlation, with Spearman’s correlation coefficients of between 0.68-0.7, all within the upper-quartile of over 12,000 genes/proteins analyzed (Table S2) [10]. Although single cell proteomics currently remains out-of-reach to most researchers, we corroborate the observed organ expression profiles by immunoblotting of normal organs obtained from C57BL6 mice for TIMPs1-3. Antibodies were pre-validated to confirm lack of cross-reactivity between the TIMP family members using recombinant proteins produced in-house. Unfortunately, despite great effort, we could not identify a suitable, commercial TIMP4 antibody that would reliably detect TIMP4 in murine tissues without excessive cross-reactivity with other TIMPs and/or unknown targets. Immunoblotting of murine tissues supports the findings that TIMP2 is abundant across all tested samples. TIMP1 expression is largely restricted in healthy organs, only detected at high amounts in cardiac tissue for which the contribution from circulating TIMP1 levels is uncertain. Indeed, The Blood Atlas (from The Human Protein Atlas; https://www.proteinatlas.org/) quantified human TIMP1 levels in plasma at a concentration of 750ug/L (by mass spectrometry), which is almost an order of magnitude higher than that of TIMP2 in the circulation (83ug/L) [11]. It is unclear whether the same is true for circulating TIMP1 concentrations in murine blood/plasma. Although mRNA levels of Timp3 are high in many tissues, protein detection by immunoblotting is difficult. This is attributed to the observation that, in most tissues, TIMP3 is anchored to the extracellular matrix through interactions with glycosaminoglycans [12], leading to loss of TIMP3 during homogenization and lysate clearing in sample preparation for SDS-PAGE and immunoblot analysis. However, TIMP3 was readily detected in lung tissue and in most other analyzed tissues albeit at low levels.
Cell-type expression of Timp genes
After identifying these organ-specific patterns of Timp expression, we then proceeded to characterize the specific cell-types expressing each Timp family member within tissues of specific organs. For this, we retained the Tabula Muris Consortium’s cell type designations (cell_ontology_class). In addition we created our own broader classifications from cell_ontology_class, which we called CellTypeB. CellTypeB was then grouped into one of seven general cell classifications, CellTypeA to simplify our data presentation (epithelial, hematopoietic, stromal, neuronal, myocytes, adipocytes and unknown), Table S3. As a result, CellTypeB classifications are more organ specific, but broad in terms of ontogeny.
Timp1, classically viewed as an “inducible” member of the Timp family [3], displays a restricted cell-type expression profile in normal organs that is consistent with this designation, Figure 1. Only chondroblasts identified in the skeletal muscle data maintain high levels of Timp1 expression, Figures 2 & 3. Analysis of Timp4 expression, although low in most organs, identifies discrete cluster-specific expression in endothelial cells across several tissues, Figure 2. In addition, Timp4 transcripts can be found at appreciable levels in aorta-associated adipocytes and astrocytes of the brain, particularly Bergmann glial cells (an astrocyte sub-type found in the cerebellum), Figures 2 & 6. Timp2 has the widest expression profile of the Timp family members (Figure 2). This is also evident in the cell-type analysis, with broad expression of Timp2 identified across all tissue types in cells of stromal lineages, Figure 4. On the contrary, Timp3 displays highly enriched expression in stromal, endothelial and other vascular-associated cell types including aorta-associated adipocytes, brain pericytes and cardiac fibroblasts (Figure S1 & S2). In general, Timp expression is not associated with cells of epithelial origin yet the highest expressing cell types for Timp2 and Timp3 are mesothelial cells (of the plural cavity, identified from lung tissue samples) and kidney tubule cells, respectively. However, it is important to note that only 24 mesothelial cells were identified in the Tabula Muris dataset, which may not reflect sufficient sampling of this cell population for definitive conclusions regarding the Timp expression levels in these cells. A detailed breakdown of the tabulation of cell type annotations and their Timp expression can be found in Tables S4 and S5.
Timp1 expressing cell types
The distribution violin plots describing Timp1 expression in several tissues reiterates that Timp1 expressing cells are generally a minority in most tissue types (Figure 3A). To determine whether the Timp1-positive (Timp1+) cells are a result of discrete cell-subtypes, we performed clustering by UMAP of the individual cell-types (Figures 3B & S2). These analyses clearly identify that Timp1 expression can be found in discrete subtypes of cells (Figure 3B). Lung stromal cells show a distinct population of that are Timp1+ and are also notably high in Timp2 and Timp3 (Figure S2). Differential expression (DE) and pathway analysis indicates that these cells are responsible for high levels of ECM remodeling and local inflammatory responses.
We also note a unique mammary stromal cell population that expresses modest-high Timp1 transcripts levels but display a significant reduction in Timp2 and Timp3 expression. Interestingly, DE and pathway analysis suggests that these cells are associated with a reduction in local inflammatory activities (Figure 3B). Tracheal stromal cells show less distinct clustering although two general Timp1+ populations can be identified. Of these Timp1+ populations, one represents a much larger portion of the analyzed cells compared to the other population, and was analyzed further, Figure 3B. Gene set analysis in this Timp1-expressing tracheal stromal cell population indicates that these cells may be directly involved in local fibrotic responses. In normal tissues this response may represent repair and maintenance of important tissue architecture, rather than a pathologic process. Muscle-associated chondroblasts are the only cell-type that displays widespread Timp1 expression. DE of the Timp1 high versus the Timp1 low cluster highlights a small number of genes (only 32 genes displaying a greater than 2-fold difference in expression). Pathway analysis of these DE genes highlights changes in Ap1/Jun signaling pathway, that are indicative of alterations in mechanotransduction in fine tuning chondroblast differentiation [13].
Timp2 expressing cell types
As the member of the Timp family with the broadest tissue expression, unsurprisingly Timp2 also displays expression in the widest range of cell types. As anticipated, the dominant source of Timp2 transcripts across all tissues are cells of stromal lineages, particularly those that are characterized as fibroblasts (see Table S3 for cell type annotations). In these cell types, Timp2 expression is uniformly consistent across the multiple tissue types, including mammary, lung, cardiac, adipose and bladder tissues (Figures 4A & S2). In contrast, other cell types display heterogeneous patterns of expression that are readily appreciated by assessment of the violin plots describing the distribution of Timp2 expression across numerous tissues and cell types in Figure 4B, clearly exemplified by the Timp2 expressing cells of the brain, skeletal muscle and heart. Assessment of UMAP clusters identifies several tissues that display cluster specific DE of Timp2. The Timp1+ cluster of lung stromal cells described above (Figure 3B), also display a >3-fold increase in expression of Timp2, and a similar increase in Timp3 (Table S6). Lung and mammary stromal cells demonstrate distinct clusters of Timp low (Timp1/2/3) cells (Figure 4, 5B & S2). In mammary stroma this small cluster of Timp low cells demonstrates a significant decrease in genes involved in ECM deposition and remodeling, reflected in the uniform downregulation of core matrisome components and a significant decrease in ECM-associated (MA) gene expression, as shown in Figure 4B. Pathway analysis of DE genes in this small cluster indicates that these cells are metabolically active with elevated oxidative phosphorylation, but low levels of ECM turnover. Further characterization of this cell population may identify novel functions of stromal cells in mammary gland physiology.
Timp2 expression in endothelial cells is minimal across most tissues, with the notable exception of adipose, trachea and muscle where cluster analysis identifies discrete populations that are Timp2+. In skeletal muscle endothelia there are two Timp2+ unique populations (Figure 4B). In adipose, trachea and one of the two muscle Timp2+ clusters these endothelial cells display an increase in pro-fibrotic gene expression as determined by DE and pathway analysis (Figure 4B, Tables S11, S12 & S13). As noted earlier, this pattern of profibrotic gene expression is likely not associated with a pathologic process in these normal tissues. Instead, these populations may reflect specific sub-populations of endothelial cells that are reactive to normal physiologic stresses, such as local inflammation, metabolic changes and or exercise-associated (hypoxic stress) changes. These physiologic factors are reflected in the similarity of gene expression for the Timp2+ clusters in all endothelial cells. There are 11 genes that display similar patterns of DE that include Nrp1, collagen IV (Col4a1 and Col4a1) and Igfbp4 (Table S30). In addition, adipose and tracheal derived endothelial cells also share a similar cluster of 113 elevated differentially expressed genes (Table S31), that are suggestive of specific differentiation pathway for endothelial cells evoking enhanced ECM deposition in Timp2 expressing cells. This finding, indicative of a quiescent endothelial phenotype, is consistent with prior reports suggesting that TIMP2 suppresses endothelial cell activation and angiogenic responses [14].
Hematopoietic cells are generally characterized by low levels of Timp expression, and this conjecture is supported by our data analyzing the bone marrow and spleen (Figure S1). Our analysis reveals Timp expression levels are detectable in select hematopoietic cell types of the bone marrow and in non-hematopoietic tissues, such as adipose and skeletal muscle tissue. Clustering of adipose granulocytes presents as two clear clusters of either Timp2 low and Timp2 high cells, with over 2000 DE genes when comparing these extreme opposite clusters, as described in Figure 4B and S2. These differences are strongly indicative of two distinct granulocyte subtypes that are Timp2 high and Timp2 low. In comparison, Timp2+ granulocytes of the bone marrow do not form clearly segregated clusters, but represent a more homogenous population, consistent with a pre-differentiation status or reliance on tissue-induced maturation. Comparison of the overlap between DE genes from Timp2+ adipose and bone marrow adipocytes reveal negligible cross-over between the two gene sets. On the contrary, comparison between the DE genes in Timp2 high adipose granulocytes and muscle macrophages highlights significant overlap of 96 genes between the gene sets of these myeloid cousins (Table S29). Pathway analysis indicates that the DE gene sets are associated with immune activation (antigen presentation and complement activation) consistent with tissue-induced differentiation when compared to bone marrow resident precursors.
Timp3 expressing cell types
Like Timp2, Timp3 is widely expressed across cells of mesenchymal lineage (stromal cells/fibroblasts), particularly within lung, cardiac and adipose tissue (Figures 5, S1 & S2). Other stromal cells display a heterogenous pattern of Timp3 expression that is associated with defined cell clusters. DE genes from Timp3 high trachea stromal cells shows a large proportion of ECM genes, many of which are components of the core matrisome, leading to altered ECM composition that is likely associated with an altered or activated fibroblast status in these tissues (Figure 5). Again, these changes are most likely associated with normal maintenance and ECM turnover, requisite for homeostasis. The major clusters of mammary stromal cells can be split into Timp1-/Timp2++/Timp3++ (Timp3++) stroma or Timp1+/Timp2+/Timp3-(Timp3-) stroma. The Timp3++ cluster is associated with a DE gene set indicative of MMP inhibition and inflammation, that in these normal tissue most likely reflects tissue remodeling associated with mammary gland involution following lactation [15] (Figure 5B). Bladder mesenchymal cells are uniformly high in Timp2 transcript levels (Figure 4A) yet display one large cluster that is high in Timp3 expression (Figure 5B). This Timp3+ mesenchymal cell cluster displays altered expression of several matrisome genes and reduced cell adhesion, and may reflect cell populations associated with bladder function and micturition. Across the four stromal cell types that display unique cluster specific patterns of high Timp3 expression (bladder mesenchymal, mammary stromal, trachea stromal and the Timp3++ lung stromal cells; Figures 3B & 5B. Full data shown in Figure S2) there were 11 correspondingly upregulated genes that may represent a set of stromal cell markers that are associated with enhanced Timp2/3 expression (Table S30).
Adipose mesenchymal stem cells are prolific suppliers of Timps, dominated by Timp2 and Timp3. However, unlike other tissues, it is difficult to discern any cluster specific expression patterns in these cells (Figure S2). In addition to ample expression of Timp3 from mesenchymal stem cells, adipose Timp3 levels are also supplemented further by high expression in leukocytes and endothelial cells, Figure 5A. Likewise, leukocytes are also a prolific source of additional Timp3 in bladder tissue (Figure 5A). The presence of tissue specific Timp3+ leukocyte populations is likely reflective of niche specific cues supporting the propagation of unique immune subtypes.
Endothelial cells represent the dominant cell type with regards to Timp3 expression across all tissues analyzed (Figure 5). The strong link between vasculature and Timp3 expression is further supported by strong expression in cardiac tissue and pericytes, the latter of which are only accounted for in brain tissue of the Tabula Muris dataset (Figure 5A). The brain has an interesting profile of Timp expression across its different cell types. As discussed earlier, Timp2 is the dominant expressed Timp across the cell types analyzed, found at appreciable levels in microglia, oligodendrocytes, pericytes and neuronal cells. Timp3 expression can be found in oligodendrocyte precursors and astrocytes, Figure 5A, as well as the pericyte population mentioned above. The kidney represents a major source of Timp3 expression, found in high levels in kidney tubule cells in addition to endothelia/fenestrated cells (Figure 5). Clustering of kidney tubule cells reveals 3 distinct groups of clusters, the major one of which is represented by high Timp3 expression and is associated with an altered amino acid metabolism and Nrf2 stress response (Figure 5B).
Timp4 expressing cells
Timp4 displays the most restricted expression of the Timp family members as it is only found in astrocytes and endothelial cells (Figure 6). Astrocytes represent the only stromal cell-type that maintain consistent expression of Timp4, particularly in the Bergmann Glial cells of the cerebellum (Figure S2). Timp4 is found in distinct clusters of endothelial cells from the heart, skeletal muscle, adipose and trachea (Figure 6B). In the trachea, Timp4+ endothelia represent a minority, whereas in the heart they represent a significantly larger proportion of the cell population. Interestingly, in both tissues DE identifies very few significant gene changes above the two-fold difference threshold (38 for trachea endothelia, 10 for heart, Tables S23 & S25). Timp4+ endothelia of skeletal muscle and adipose tissue represent a large proportion of the cells in these tissues that, unlike cardiac endothelia, form distinct cell clusters. DE identified 135 gene changes in Timp4+ skeletal muscle endothelia and 366 in Timp4+ adipose endothelia, of which there was an overlap of 73 genes with similar changes (Tables S24, S26 & S32). Gene set analysis shows that Pparγ and/or a loss of Hif1a signaling are correlated with an increase in endothelial Timp4 expression (Tables S23, S24, S25 & S26) that may be associated with oxygenation status in these endothelial populations.
Discussion
Despite their key functional attributes and expanding biological roles, Timps are an understudied family of endogenous proteins. Timp1 and 3 have been referred to as inducible Timps [3].In our study we show that Timp1 expression, although present, is low in many normal tissues, supporting the idea that Timp1 is generally inducible in response to tissue damage or inflammation. In fact, there are numerous reports describing an increase in Timp1 expression in response to local tissue pathologies such as fibrosis and cancer [16-20]. Comparison of the differential expression (DE) profiles of Timp1+ cells reveals very little crossover between the 4 cell types analyzed in detail (Figure 3B, Tables S6, S7, S8 & S9), suggesting that the stimuli for Timp1 expression may be broad and/or tissue specific. Indeed, increases in Timp1 levels have been described in numerous and varied experimental systems that include models of chemotherapy [21], β-adrenergic stimulation [22], inflammation [23-25], immune function [26], infection [27-31], cancer [16, 32], and pain [33]. For example, Timp1 expression in lung and mammary stromal cells may be related to either its canonical (MMP inhibition) or non-canonical functions, that may be further influenced by hormonal status (menstrual/lactational state). Concurrent increases observed in Timp1-3 expression in lung stromal cells may be directly associated with a stromal cell type that functions to actively remodel ECM composition in concert with Mmps/Adams. In mammary stromal cells, Timp1 displays an inverse correlation with Timp3 expression and is also associated with a reduction in Timp2. Interestingly these cells also display a concurrent increase in Cd63 expression (Table S7) a known Mmp-independent Timp1 receptor [18]. This correlation between Timp1 and Cd63 expression is also noted in chondroblasts (Table S9), a relationship that may function to mediate cell survival/proliferation under conditions where there is reduced Mmp expression described elsewhere [16]. In contradistinction to the restricted, inducible expression of Timp1, we find that Timp3 is highly expressed in many normal tissues, particularly within the vascular and renal systems [34, 35] (Figures 5, S1 & S2). Vascular and tubular cells are exposed to fluid shear stress from the flow of fluid [36, 37]. Whether Timp3 expression in these cells is lineage dependent or a result of outside-in signaling due to luminal flow is an important question of interest, with shear stress having been shown to enhance Timp3 expression in pericytes [38]. A potential advantage for expression of Timp3 in these situations, over that of other Timp family members, is the fact that Timp3 is matrix-anchored due to specific ECM interactions thereby maintaining a presence of MMP-inhibitory capabilities that may not be maintained with readily diffusible Timps [5]. This would contribute to ECM stability in the presence of active flux due to the barrier and exchange functions of endothelia and kidney tubular epithelia. Timp3 expression is found in the stromal compartment of all organs. It is also uniquely expressed in tissue-resident leukocytes of both the bladder and adipose tissue, which warrants further investigation with regards to Timp3’s potential immunological role in these cell that has attracted previous interest [39].
Cellular arrangement in organs and tissues is astoundingly complex, and this is particularly evident in the cellular organization within the CNS. The brain is also resident to the most unique expression pattern of the Timp family. In our analysis of the Tabula Muris dataset we find little/no expression of Timp1 in the cellular subsets of the brain. This observation is supported by perusal of the Human Protein Atlas (https://www.proteinatlas.org/ENSG00000102265-TIMP1). Despite this, Timp1 has been reported in the central nervous system (CNS) on a limited number of occasions [40, 41]. In our analysis, Timp2 is the dominant expressed member in the CNS. Its highest levels are found in microglia, pericytes and oligodendrocytes, whilst maintaining a somewhat mixed expression in neuronal cells. Timp3 is found in pericytes, endothelial cells, astrocytes and oligodendrocyte precursors. This latter observation is curious considering the switch to Timp2 expression as these cells mature to oligodendrocytes. The overall findings regarding Timp expression in the CNS are consistent with the predominant pattern of stromal cell expression in other organs.
The only stromal cell type that retains a consistent expression of Timp4 are astrocytes, expression of which is particularly enriched in Bergmann glial cells of the cerebellum. The preferential expression of Timp4 in these certain cell-types constitutes an interesting focus for future research. In many tissues, Timp expression can be associated with specific or discrete cell clusters. The same is not true in the CNS, where we find little/no cell cluster specific Timp expression patterns, which can be attributed to the complexity of resident cell phenotypes. One potential means to tease out unique differences would be to perform cluster analysis with filtered gene lists entailing matrisome and matrisome-related genes only. Brain vascular function is crucial for the maintenance of normal function, and this vasculature becomes dysfunctional with age and disease [42]. Our observation of the consistent expression of Timp3 across the vascular system, including the CNS, is consistent with the multiple roles that all Timps plays in the maintenance of vascular function [43-46]. In addition, it is reasonable to conclude that pathways involving Timps are involved in the pathogenesis of CNS dysfunction associated with aging and degenerative neurologic conditions, such as Parkinson’s and Alzheimer’s disease.
The stroma is a broad term that is sometimes used to encapsulate all non-epithelial cells in tissues. In more specific-terms, stroma (or stromal cells) is used to represent cells of mesenchymal lineages. In the case of the Tabula Muris dataset, this includes fibroblasts, mesenchymal cells, mesenchymal stem cells, smooth muscle cells, chondroblasts, pericytes, oligodendrocytes (and precursors) and astrocytes. One of the richest sources of Timps can be identified in adipose mesenchymal stem cells (aMSCs) (Figure S2). These multipotent stromal cells cluster into numerous visibly distinct populations/sub-types. We observed that aMSCs are universally high in Timp2 expression, followed by somewhat lower but diffuse expression of Timp3 and, to a much lesser extent, Timp1 (Figure S2). Like the CNS, this cell-type is another unusual case where there are no clearly defined cluster-specific patterns of Timp expression, an observation that also extends to skeletal muscle mesenchymal stem cells (Figure S1). Timp high stromal cells display gene expression profiles that clearly align with active remodeling of ECM structure and composition. In contrast, the few clusters that display low levels of Timp expression in stromal cells (in mammary and lung stroma) correlate with a major downregulation of both matrisome and matrisome-associated genes. This finding strongly implies that these Timp low stromal cells have a limited role in the regulation of ECM composition and function. DE identified 618 and 449 genes for the Timp low clusters of mammary stromal and lung stromal cells, respectively (Tables S10 & S21). Assessment of the shared downregulated genes identifies 112 transcripts that are similarly altered in Timp low stromal cells (107 of which are downregulated) (Table S31). These co-downregulated genes included several key fibroblast subtype markers that have been identified in various studies, such as Pi16, Scara5 and Col14a1 [47-49]. Of the stromal cell types included in this study, those identified in the trachea, lung, mammary and bladder tissues form specific cell clusters that display unique patterns of Timp expression, an example of which was discussed earlier regarding Timp1/3 expression in breast and lung stromal cells. Timp2 is almost universally expressed in stromal cells in most cases, the expression of which is often correlated with that of Timp3. Timp2/3 high stromal cells display more specific patterns of expression, with shared upregulation of various defining markers of fibroblast sub-populations such as Cd34, Pi16, Col14a1 and Scara5 (Table S30) [47-50]. These findings suggest there is a direct, positive correlation in stromal cell population between Timp expression and a subgroup of marker genes (Pi16, Col4a1 and Scara5). It remains to be determined if there is a direct functional link between this pattern of gene expression and regulation of ECM turnover.
Despite Timp2’s strong association with stromal cell types, the highest expression of Timp2 was identified in mesothelial cells identified from lung tissue. Mesothelial cells are descriptively referred to as a simple squamous epithelial lining of the pleural and abdominal cavities. However, they are derived from the embryonic mesoderm and have been ascribed a wide variety of functions, including fluid transport, ECM modulation, as well as cytokine and growth factor secretion. There have been limited reports of Timp2 expression in the mesothelium [51], and it is not known whether the observations in the Tabula Muris dataset will correlate with enhanced expression at the protein level. Additionally, the fact that only 24 mesothelial cells were identified in the lung tissue warrants tempered enthusiasm about the biological significance of this observation and explains why this cell-type was not investigated further. However, given their mesodermal origin it is not surprising that mesothelial cells show significant similarities with stromal/mesenchymal cells in regards to their Timp expression levels.
Adipose tissue in the Tabula Muris dataset is somewhat skewed by large numbers of hematopoietic cells that were retained for analysis. Ignoring these cells shows that adipose tissue is particularly enriched for Timp expression. Although not generally associated with hematopoietic cells, we observed that select tissue resident myeloid cells are associated with cluster specific Timp2 expression, specifically adipose-associated granulocytes and muscle macrophages. Considering there is significant cross-over in the DE genes identified in the Timp2 high populations of these cells, even though they represent different cell types, suggests that shared stimuli/pathways are responsible for mediating Timp2-expression in these tissue specific, myeloid cell sub-types (Tables S16, S17 & S29).
Timp3 expression is strongly associated with the vasculature and is widely expressed across all structural components of the cardiovascular system. Timp2 and Timp4 are detected in endothelial cells of some specific tissues at low-to-medium expression levels. Cluster analysis further identifies specific sub-types of endothelial cells enriched for the expression of these two Timps and deserves further investigation. Timp2 and Timp4 display varying levels of sub-type specific expression. For example, adipose and trachea endothelia Timp2 expressing clusters represent a distinct minority, this proves true also for Timp4 in the trachea, Figure 4B and 6B, respectively. Timp2 and Timp4 display inverse correlation of expression in trachea, adipose and skeletal muscle endothelial cells, with the latter showing the most prominent inverse correlation that consists of two opposing Timp2+ populations that are separated by a Timp4+ population. Gene set pathway analysis strongly suggests that Pparγ signaling is responsible for Timp4 expression in these cells and, despite being poorly characterized, Timp4 has been shown to be induced through Pparγ signaling (Tables S23, S24, S25 & S26) [52]. Timp2 high endothelial cells of the trachea and adipose tissue share a significant number of similar DE genes (113 in number), supporting the idea of a common stimulus/pathway as a regulating factor. Comparing all Timp2 high endothelial cell clusters (two of which are identified in skeletal muscle), identifies 11 common DE genes that includes a downregulation of Col4a1/Col4a2. Interestingly, there is a clear inverse link between Timp2 and Col4a1/Col4a2 expression in mammary stromal cells, indicating there may be a more definitive connection between these two genes of the matrisome than previously appreciated.
Why do distinct cell type clusters express different Timps, and is differential expression of Timp family members of real physiological consequence? In some cases, cluster specific Timp expression is associated with a host of transcript changes, including both matrisome and non-matrisome associated genes. In these instances, it is possible that there is collateral transcript expression due to overlap of broad and divergent signaling pathways. So, does Timp transcription translate to equivalent protein levels? These outstanding questions require further investigation. In other cases, cluster definition and significant transcript changes are minimal, as observed with Timp1+ chondroblasts. The utility of a more targeted increase in Timp expression is yet to be determined, as enhanced TIMP protein levels can have multiple outcomes that include MMP-dependent and -independent biological activities. However, this study does support the concept that in normal tissues Timp function can directly modulate cell behavior. This can be evidenced by the demonstration that Timp1+ chondroblasts correlate with an increase in Cd63 expression, a known Timp1 mitogenic receptor, suggesting that the phenotypic outcome may be related to an autocrine signaling loop [18].
Despite the findings of our study, analysis of the Tabula Muris dataset has its limitations. The techniques that are used to isolate and extract single cell RNA displays internal biases towards complicit cell types, leading to a misrepresentation of the cell type proportions, exampled by the lack of skeletal muscle myocytes, no stromal cell representation in several tissues (colon, liver, skin, bone marrow) and the disproportionately low number of neurons. Ideally, data from parallel single cell RNA sequencing and bulk RNA sequencing experiments would be used to address these issues with cell type biases. Regardless, cell type proportions can vary between regions within individual organs, as exampled by the variable ratios between neuronal and glial cells in the CNS suggesting that there may be certain microenvironments within that possess variable levels and/or combinations of Timp expression [53]. The GTEx consortium recently reported that, though transcript-protein level correlation is generally poor, human TIMPs 1-3 correlate well with transcript levels [10]. Although murine and human Timps/TIMPs display very high sequence identity, it is unknown whether this corresponds with a similar tissue/cell-type expression profile.
We describe the tissues and cell-types that express members of the Timp family in murine tissues. Tissue expression of Timps is altered in disease [3, 54-56], and in situations where Timp expression is unaffected many matrisome/matrisome-associated genes that can modulate Timp activity are affected [57]. By understanding how these cell clusters and their matrisome gene expression patterns change in disease may reveal mechanistic insights that can be targeted for novel therapeutic interventions. Our study identifies unique Timp expressing cells of stromal, epithelial and hematopoietic origin across a broad range of tissues. Considering the therapeutic utilities of Timp family genes, we propose that targeted expansion of these tissue-resident cell subtypes will potentially represent suitable treatment options in a range of pathologic conditions.
Experimental Procedures
Tissue collection and protein extraction
Organs for immunoblotting were harvested from wild type female C57/BL6 mice (4-month-old) immediately following euthanasia by CO2 narcosis with aortic transection. Whole bladder, pancreas, gastrocnemius, left inguinal mammary fat pad and anterior subcutaneous white adipose tissue were collected. A ∼5mm2 section of the left & right cardiac ventricle, lower left lung and sagittal section of the brain were collected and rinsed in ice cold phosphate buffered saline, before being manually diced using two scalpels in a scissor-action. Diced organs were then placed into 1mL of ice-cold lysis buffer (1X RIPA buffer, 1% Phosphatase Inhibitor, 1% Protease Inhibitor, 1% Anti-foam Y-30) in M-tubes (gentleMACS™ Tubes, Miltenyi Biotec). Organs were then mechanically homogenized using a gentleMACS Dissociator (Miltenyi Biotec) using program Protein_01. Tissues needing further homogenization were placed on ice for 5 minutes before running the program again, to avoid overheating. The M-tubes were then spun at 200G for 2 minutes at 4°C and incubated on ice for 5 minutes. The homogenized tissue was then sonicated at 60% power for 1 minute in 10 second intervals (10s run, 10s rest, 110s total). The debris was pelleted via centrifugation at 10,000G for 15 minutes at 4°C. The supernatant was then placed in QIAshredder tubes (Qiagen) and run at full speed for 2 minutes at 4°C. The flowthrough was aliquoted and placed in -80°C for storage.
Immunoblotting
Protein concentration was quantified using Pierce BCA Protein Assay (Thermo Scientific, USA) and normalized to 40 µg of total protein for each tissue sample. After heat denaturation, reduced samples were resolved by SDS-PAGE on 4-20% gels and transferred on to nitrocellulose membranes. Blots were blocked in 2.5% milk in TBS with 0.1% Tween-20 (TBST) for 30 min at room temperature and then incubated with primary antibody overnight at 4°C. Blots were washed with TBST, incubated with HRP-conjugated secondary antibody for 1 hour at room temperature, and then visualized using Bio-Rad ChemiDoc™ Imaging System after detection with chemiluminescent substrate (SuperSignal™ West Pico, Thermo Scientific; or Radiance Plus, Azure Biosystems). Primary antibodies were extensively optimized. The following primary antibodies were used for detection: TIMP1 (Cell Signaling Technology, USA) at 1:500 dilution, TIMP2 (R&D Systems; AF971) 1:500, and TIMP3 (Cell Signaling Technology, USA) 1:750.
Secondary anti-rabbit HRP-linked antibody (Cell Signaling Technology, USA) or anti-goat HRP-linked antibody (Invitrogen, USA) was used at 1:5000 dilution. Protein loading was determined by Ponceau Stain (Sigma Aldrich, USA). Immunoblot images were processed using Image Lab Software Version 6.1 (Bio-Rad). In-house produced recombinant human TIMP1/2/3 were included in the immunoblots at 4ng (TIMP1/2) or 10ng (TIMP3) per lane. Human TIMPs display high sequence identity with murine Timps (74.3%, 97.3%, 96.2% for Timp1/2/3, respectively).
Data harvest and processing
Single-cell transcriptomic data generated through FACS (fluorescence-activated cell sorting) or microfluidic-droplet methods, as well as individual cell annotations, were downloaded from Tabula Muris Figshare (https://figshare.com/articles/dataset/Single-cell_RNA-seq_data_from_Smart-seq2_sequencing_of_FACS_sorted_cells_v2_/5829687)(https://figshare.com/articles/dataset/Single-cell_RNA-seq_data_from_microfluidic_emulsion_v2_/5968960). Brain, colon, fat, pancreas and skin tissues only have FACS data. Bladder, heart, aorta, kidney, liver, lung, mammary, marrow, muscle, spleen, thymus, tongue and trachea organs have data from both methods. In our data analyses the Timp expression data from aorta was separated from heart, and limb muscle (which we refer to as skeletal muscle) was separated from the diaphragm. Additionally, brain myeloid and non-myeloid organs were combined into one organ. As a result, there are 17 organs analyzed (see Table S1) compared to the original 20 organs in the Tabula Muris dataset. Seurat (V4) R package (https://satijalab.org/seurat/) was used for data quality control (QC), normalization, integration, clustering, visualization as well as differential expression (DE) analyses.
For organs that only have FACS data, SCTransform [58] was applied for normalization before clustering. For tissues with datasets from both methods, due to the observed technical batch effect between the two methods, Seurat integration pipeline [59] was applied along with SCTransform normalization (https://satijalab.org/seurat/articles/integration_introduction.html).
When studying specific cell types, based on annotations, gene raw counts data were retrieved for those cell types followed by the same normalization or integration workflow. For tissue-specific analyses of cell-types, cells obtained from the aorta and diaphragm tissue were separated from heart and skeletal muscle datasets, respectively. Expression visualizations were generated manually using Adobe Photoshop and the circlepackeR R package.
Clustering
Principal components (PCs), explaining most variance, were selected based on their rankings displayed by Seurat’s Elbowplot() function. Selected PCs for the different datasets ranged between 30-40. FindNeighbors() function was run to generate a nearest neighbor graph with the identified PCs as dimensions of reductions with the default settings. Clusters of cells were identified by a shared nearest neighbor (SNN) modularity optimization-based algorithm by utilizing the FindClusters() function at the default 0.8 resolution. Clusters are visualized with Uniform Manifold Approximation and Projection (UMAP) dimensional reduction technique with the selected PCs.
Differential expression (DE) and pathway analysis
Based on the UMAP clusters, different clusters and/or groups of clusters were selected for differential expression analysis. Differentially expressed genes were identified for groups and/or clusters with the FindMarkers(?) function. For samples that were integrated to remove the batch effect, a logistic regression framework was used with batch information as a latent variable. For samples without multiple batches, default Wilcoxon Rank Sum test was used with SCT assay data. Identified markers were filtered for Bonferroni corrected adjusted p-values <0.05. Pathway analysis was performed using Ingenuity Pathway Analysis (Qiagen) and MetaCore (Clarivate).
Contributions
D.P. and W.G.S.S. conceived the study. D.P., Y.F. and S.R. performed data harvest and analysis. S.G. and C.L. performed tissue harvest and immunoblots. All authors assisted in interpretation of data. The manuscript was written by D.P. and W.G.S.S. All authors reviewed the manuscript.
Ethics Declarations
The authors declare no competing interests.
Data Availability
The authors confirm that the data supporting the findings of this study are available within the article [and/or] its supplementary materials. Example code for tissue-specific and cell-type specific data collection can be found in supplementary Text S1 & S2.
Acknowledgements
This research was supported by the Intramural Research Program of the NIH (W.G.S.-S. Project ID ZIA SC 009179)
Abbreviations
- (TIMPs/Timps)
- Tissue inhibitors of metalloproteinases
- (MMP)
- Metzincin proteases of the Matrix Metalloproteinase
- (ADAM)
- A Disintegrin and Metalloproteinase
- (ADAMS-TS)
- A Disintegrin and Metalloproteinase with thrombospondin motifs
- (scRNA sequencing)
- single cell RNA sequencing