Spatial transcriptomic validation of a biomimetic model of fibrosis enables re-evaluation of a therapeutic antibody targeting LOXL2

Abstract

Matrix stiffening by lysyl oxidase-like 2 (LOXL2)-mediated collagen cross-linking is proposed as a core feedforward mechanism that promotes fibrogenesis. Failure in clinical trials of simtuzumab (the humanized version of AB0023, a monoclonal antibody against human LOXL2) suggested that targeting LOXL2 may not have disease relevance; however, target engagement was not directly evaluated. We compare the spatial transcriptome of active human lung fibrogenesis sites with different human cell culture models to identify a disease-relevant model. Within the selected model, we then evaluate AB0023, identifying that it does not inhibit collagen cross-linking or reduce tissue stiffness, nor does it inhibit LOXL2 catalytic activity. In contrast, it does potently inhibit angiogenesis consistent with an alternative, non-enzymatic mechanism of action. Thus, AB0023 is anti-angiogenic but does not inhibit LOXL2 catalytic activity, collagen cross-linking, or tissue stiffening. These findings have implications for the interpretation of the lack of efficacy of simtuzumab in clinical trials of fibrotic diseases.

Keywords: LOXL2; disease-relevant biomimetic models; fibrosis; spatial transcriptomics; target engagement.

Graphical abstract

Figure 1

The spatial transcriptome of fibroblast foci is enriched for ECM development and ossification gene expression signatures (A) Principal component analysis (PCA) plot showing variance of the transcriptome (GSE169500) of laser-capture microdissection (LCMD) samples for control alveolar septae, IPF alveolar septae, and IPF fibroblast foci (n = 10 control and IPF donors). (B–F) Violin plots showing expression of ACTA2 (B), TNC (C), COL1A1 (D), COL5A1 (E), and SFTPC (F). Relative expression levels are calculated as counts per million reads (CPM). Fibroblastic foci gene expression values are compared with control alveolar septae. (G) Gene set variation analysis (GSVA) of LCMD data, showing top 20 significantly enriched Gene Ontology (GO) terms in IPF fibroblastic foci compared to control alveolar septae. GO terms are ranked by −log₁₀(adjusted p value). (H, I) Violin plots showing GSVA scores for GO terms collagen fibril organization (H) and bone morphogenesis (I). p values calculated using the EdgeR R package (B–F) and the Limma R package (H, I) comparing fibroblastic foci and IPF alveolar septae to control septae using Benjamini-Hochberg multiple test correction ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

Figure 2

The 3D spheroid model most closely reflects the transcriptome of fibroblast foci (A) Schematic representation of the experimental design. Primary parenchymal lung fibroblasts were cultured in standard 2D culture, the Scar-in-a-jar model, or the 3D spheroid model in the absence or presence of TGF-β at early and late time points (n = 4 per condition) and each transcriptome compared through bioinformatic analysis with the transcriptome of fibroblast foci. (B) Representative phase-contrast micrographs of standard 2D culture, scar-in-a-jar, and 3D spheroid model. The 3D morphology of spheroids is visualized by confocal microscopy utilizing fluorescence of carmine red staining at bandwidths of 500–550 and 690–740 nm (green and red false colors, respectively). Scale bars are all 200 μm. (C) Heatmap representing unsupervised hierarchical clustering of highly variable GSVA scores in models and fibroblastic foci (Kruskal-Wallis test, p < 0.01). (D–H) GSVA plots of highly variable GO terms (Kruskal-Wallis W test, p < 0.01) in models compared to fibroblastic foci.

Figure 3

Clustering of RNA sequencing data from different fibroblast culture models reveals enrichment for ECM formation within the 3D spheroid model (A) Heatmap showing hierarchical clustering of RNA sequencing (RNA-seq) data of the different fibroblast culture models. Genes shown are all differentially expressed between different models. Box shows upregulated genes within the 3D spheroid model used to identify GO terms in (B). (B) G:profiler GO enrichment of genes associated with the 3D spheroid model (+TGF-β) (box in A). Box denoted GO terms discussed within the manuscript text. (C–G) Violin plots of genes from (A) associated with regulation of the ECM microenvironment, (C) COL3A1, (D) P4HA3, (E) PLOD2, (F) HAS1, and (G) TWIST1.

Figure 4

The 3D spheroid model recapitulates features of the complex ECM structure identified within fibroblast foci (A) Venn diagram showing the number of unique and common ECM proteins identified between the fibroblast focus spatial proteome (Herrera et al.⁵⁹) and the 3D spheroid model. (B) G:Profiler analysis of 88 ECM proteins common between the fibroblastic focus spatial proteome and the 3D spheroid model proteome. Top 20 most significant biological process GO terms shown. (C) A comparison of proteomic data derived from the ECM of IPF tissue (Booth et al.⁶²), the fibroblast focus spatial proteome (Herrera et al.), and the 3D spheroid model, showing the top ECM proteins by abundance in IPF ECM and their presence or absence in the other datasets. X indicates proteins that are present and those that are absent. (D) STRING map showing proteins (Data S2C) identified within the 3D spheroid model that have a hazard ratio >1 in Oldham et al. for increased risk for death or lung transplantation. Each protein has up to 10 medium confidence interaction partners. (E) G:Profiler analysis showing top 10 pathways from proteins identified in (D). (F and G) Immunohistochemical staining of IPF lung tissue and 3D spheroids. Top left is a Masson’s trichrome stain of IPF tissue with a fibroblast focus identified by ∗, with a serial section bottom left with immunohistochemical staining for tenascin C (F) or procollagen type 1 (G); on the right-hand side is the corresponding immunohistochemical staining for the 3D spheroid model with a higher magnification inset. Inset scale bars are 50 μm. (H and I) Transmission electron microscopy images of the 3D spheroid model identifying (H) a complex intercalated ECM and (I) the D-banding characteristic of fibrillar collagen. Scale bars are 1 μm (H) and 100 nm (I).

Figure 5

The anti-LOXL2 antibody AB0023 does not inhibit collagen-cross linking, tissue stiffness, or catalytic amine oxidase activity in a disease-relevant model of fibrosis (A) LOXL2 protein levels were assayed in the serum of patients with IPF (n = 12) and control subjects (n = 13). (B) Expression of LOXL2 within spatially resolved regions of control and IPF lung tissue. Data from Eyres et al. (C) Representative images of mRNA expression of LOXL2 (green chromagen) and PLOD2 (red chromagen) within a fibroblast focus (∗) of IPF lung tissue and the 3D spheroid model using RNAscope RNA in situ hybridization. Scale bars are 20 μm. Arrows indicating cells with co-expression of LOXL2 and PLOD2. (D–I) Lung fibroblasts from patients with IPF (n = 3 donors across 2 independent experiments) were used in the 3D spheroid model in the presence of AB0023 or an isotype control antibody at the same concentrations, as well as with PXS-S2A or its vehicle control (0.1% DMSO). (D) Total mature trivalent (PYD + DPD) collagen cross-links determined by ELISA (n = 3). (E) Tissue stiffness measured from parallel-plate compression testing determined by Young’s modulus and represented as a proportion of control (n = 6). (F) Total collagen content determined by hydroxyproline assay. (G) LOXL2 catalytic amine oxidase activity within the conditioned media was assessed using an activity-based probe (n = 3). (H) VEGFA within cell-conditioned media determined by ELISA. (I) Fibronectin within cell-conditioned media determined by ELISA. Data are mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. Unpaired two-tailed t test (A) (t = 2.651, degrees of freedom = 23), Wilcoxon test with Benjamini-Hochberg multiple test correction (B), and ANOVA with Šídák’s multiple comparisons test (D–H) were used to evaluate statistical significance (F values: (D) 7.508, (E) 2.862, (F) 0.8966, (G) 33.06, (H) 34.1, and (I) 45.03. Degrees of freedom: (D) 12, (E) 30, (F) 12, (G) 12, (H) 12, and (I) 12). Error bars are standard deviation.

Figure 6

The anti-LOXL2 antibody AB0023 is a potent inhibitor of angiogenesis Endothelial tube formation by human umbilical vein endothelial cells was assessed in the presence of AB0023 or an isotype control antibody at the same concentration (n = 10 replicates per condition across 3 independent experiments). (A) Representative images of endothelial tube formation under each condition visualized by calcein staining. Scale bars are 500 μm. (B–D) Quantification of topological parameters of capillary structure by computer-aided image analysis for (B) number of loops, (C) number of branching points, and (D) total tube length. Data are mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. ANOVA with Šídák’s multiple comparisons test was used to determine statistical significance (F statistics: B 120.3, C 36.31, and D 74.56. Degrees of freedom: 36). Error bars are standard deviation.