A lipid-associated macrophage lineage rewires the spatial landscape of adipose tissue in early obesity

Adipose tissue macrophage (ATM) infiltration is associated with adipose tissue dysfunction and insulin resistance in mice and humans. Recent single-cell data highlight increased ATM heterogeneity in obesity but do not provide a spatial context for ATM phenotype dynamics. We integrated single-cell RNA-Seq, spatial transcriptomics, and imaging of murine adipose tissue in a time course study of diet-induced obesity. Overall, proinflammatory immune cells were predominant in early obesity, whereas nonresident antiinflammatory ATMs predominated in chronic obesity. A subset of these antiinflammatory ATMs were transcriptomically intermediate between monocytes and mature lipid-associated macrophages (LAMs) and were consistent with a LAM precursor (pre-LAM). Pre-LAMs were spatially associated with early obesity crown-like structures (CLSs), which indicate adipose tissue dysfunction. Spatial data showed colocalization of ligand-receptor transcripts related to lipid signaling among monocytes, pre-LAMs, and LAMs, including Apoe, Lrp1, Lpl, and App. Pre-LAM expression of these ligands in early obesity suggested signaling to LAMs in the CLS microenvironment. Our results refine understanding of ATM diversity and provide insight into the dynamics of the LAM lineage during development of metabolic disease.


Supplementary Algorithms
Algorithm 1: Clustering and Visualization Input: Data matrix X m⇥n = (x 1 , ..., x n ) 2 R m⇥n where m rows are genes and n columns are cells.Output: Cell clusters and a low dimensional projection 1: Compute the sample mean µ n and the centered matrix X c = X µ n 1 > where 1 is a vector of ones where each column in P is a right singular vector of X c .Here r can be chosen using the optimal hard threshold (6)  1.Two state matrices, S x 2 R nx⇥m and S y 2 R ny⇥m where n x , n y rows are the number of cells in states S x , S y respectively and m columns are genes.Note that n x 6 = n y , but m is assumed to be consistent between S x and S y .The states S x and S y should be chosen as hypothetical poles of a continuum of biological interest.2. Data matrix D 2 R n⇥m where the n rows are cells and the m columns are the genes, consistent with m above.Cells in D will be quantified along the continuum defined by states S x and S y .
Output: Cell continuum values along user-defined axis for cells in D 1: Define signatures, t x , t y 2 R m for states S x and S y .For example, a function f aggregating expression of each gene over all cells: 2: Define gene-set of interest.For example, select the top k di↵erentially expressed genes between S x , and S y over m, ranked by their fold change.
3: Compute the similarity between each cell and the state signatures: d x = similarity(D, t x ) and d y = similarity(D, t y ).The choice of similarity measure depends on the data and user preference.
4: Determine the continuum axis with respect to S x .For example, using ordinary least-squares (OLS), structure the following minimization problem: where X n⇥2 = (d x , 1) 2 R n⇥2 and 1 is a column vector of ones.The solution to Equation 10 is: where w is a vector containing the slope w 0 and the intercept w 1 of the line of best fit for the data.
5: Compute the position along the continuum axis for each cell.Let dy be the predicted similarity values obtained from the OLS solution.We obtain a vector of positions along the continuum, dy , using Equation 12: Let the coordinates for each cell along the continuum axis be Compute the distance along the continuum axis for each cell with respect to a reference point, p.For example, the reference point may be defined as the cell with the highest similarity to either pole.Let p 1⇥2 = (x, y) 2 R 2 , then the distances, h, are defined by For convenience, we rescale distances h using: Supplementary Tables  3: Protein Validation of Predicted Cell Types.Results of Wilcoxon rank-sum tests for di↵erential protein expression for each cell type against all other cells at each time point.We adjusted ↵ using Bonferroni's correction and required that the fold change (log2) was greater than 0.5.Using this conservative criteria we show that our cell type annotations are highly aligned with protein expression.

Supplementary Figure 5 :
Immune cell scRNA-seq summary.(A) The total number of cells of each type by diet condition.(B) The total number of macrophages in each subcluster by diet condition.(C) The frequency of each cell type by diet condition.(D) The frequency of each macrophage subcluster by diet condition.(E) The fold change (log2) of cell number over ND for all cell types.(F) The fold change (log2) of cell number over ND for all macrophage subtypes.Supplementary Figure 8: Inflammatory response-related gene expression in ATM subtypes.(A) Distribution of log2 fold changes for 196 inflammation-related genes in the Molecular Signatures Database pathway MM3890 in ATM subtypes.Each point represents a gene.(B) Expression of key inflammationrelated genes in macrophage subtypes.Small points represent cells and subtype means are represented by large points.(C) Top 15 and bottom 15 di↵erentially expressed (Wilcoxon rank-sum test) inflammationrelated genes from MM3890 across ATM subtypes.All genes are statistically significant in at least one ATM subtype using Bonferroni's correction (↵ = 0.05).Genes are sorted by log2 fold change.(D) Top 15 and bottom 15 di↵erentially expressed inflammation-related genes (Wilcoxon rank-sum test) for grouped ATM subtypes.All genes are statistically significant in at least one group using Bonferroni's correction (↵ = 0.05).Genes are sorted by log2 fold change.Supplementary Figure 12: Adipose tissue T cell, monocyte, and dendritic cell subtypes in obesity by scRNA-seq.(A-C) T cell subclusters included regulatory T cells (T1), conventional T cells (T2), and all other T cells (T3).(A) UMAP of T cell subclusters.(B) Di↵erentially expressed T cell subtype markers shown as log2 fold change of each population versus all others.(C) T cell subtype quantity per diet condition, shown as count per hundred (proportion of parent).(D-F) Monocyte subclusters included inflammatory (MO1, MO3) and patrolling (MO2) subtypes.(D) UMAP of monocyte subclusters.(E)Di↵erentially expressed monocyte subtype markers shown as log2 fold change of each population versus all others.(F) Monocyte subtype quantity per diet condition, shown as count per hundred (proportion of parent).(G-I) Dendritic cell subclusters included classical (DC1, DC2) and plasmacytoid (DC3) subtypes.(G) UMAP of dendritic cell subclusters.(H) Di↵erentially expressed dendritic cell subtype markers shown as log2 fold change of each population versus all others.(I) Dendritic cell subtype quantity per diet condition, shown as count per hundred (proportion of parent).DE genes were determined using Wilcoxon rank-sum tests.
on X c 4: Construct a similarity matrix A n⇥n from P by determining the distance between each row.The choice of distance measure depends on the data type and user preference.Examples include Gaussian similarity, Euclidean distance, Manhattan distance (city block distance), Kullbeck-Liebler divergence, and correlation 5: Perform clustering: spectral or modularity clustering on A with k clusters.k can be chosen using domain knowledge or by testing multiple values of k and evaluating the best performance.Note: k may be  r 6: Visualization: t-SNE or UMAP to reduce the dimensions of P and visualize data colored according to clusters Algorithm 2: Continuum Quantification Input:

Table 1 :
Mean adipocyte size with HFD feeding.Adipocyte area distributions measured in images from high-resolution microscopy.

Table 2 :
Number of cells by type at each diet condition.