Discovery of HCC PDX tumors that are highly sensitive to JAG1 inhibition

A) In vivo efficacy of aJ1.b70 treatment in LIV78 HCC PDX model. Data is shown for multiple doses. B) Analysis of tumor growth inhibition in an unbiased panel of liver tumor PDX models following aJ1.b70 treatment. Dark green panel, complete response; light green panel, partial response, white panel, no response; red panel, increased tumor growth. C) In vivo efficacy of aJ1.b70 treatment in models selected based on shared features with LIV78. Extent of growth response is highlighted as in B) within each subpanel. D) In vivo efficacy of aJ1.b70 treatment HCC PDX model Li-1035. Data is shown for multiple doses.

Treatment sensitive tumors depend on an intrinsic JAG1 to NOTCH2 signal A) Expression of Notch ligands and receptors in tumor cells (human reads) and stroma (mouse reads) in models LIV78 (blue, n=5) and Li-1035 (red, n=4). Data was generated from tumors 3 days after single dose treatment with control antibody. B) Efficacy of aNRR2 treatment in LIV78 model. C) Efficacy of aNRR1 or aNRR2 treatment in Li-1035 model. D) Immunohistochemical detection of NICD1, NRR2 and JAG1 in model Li1035 treated with control antibodies or antibodies blocking JAG1 or the indicated receptor. Scale bar: 100µm E) Efficacy of treatment with aJ1.b70 or ahJAG1 (=aJ1.b6A9, human specific, in LIV78 model. F) Efficacy of aJ1.b70 or ahJAG1 treatment in Li-1035 model. Extent of growth response in panels B,C,E, and F is highlighted as in Fig.1B.

JAG1/NRR2 Inhibition-sensitive HCCs are progenitor-like

A) Heatmap of top 1,000 most variable expressed genes in all tested HCC PDX models at baseline (48-72hrs post single dose control antibody treatment). K-means clustering highlights sensitive models as subset of one larger cluster. B) PCA of HCC PDX models colored by k-means clustering. C) Expression of HCC subtype specific signatures in sensitive and non-sensitive models stratified by k-means cluster (Goossens et al., 2015b). D) Expression of AFP in sensitive and insensitive HCC models in k-means cluster 3. E) Co-expression of GAS7 and FGF9 separates sensitive and non-sensitive models. F) RNAscope detection of GAS7 or FGF9 mRNA expression on tissue sections from treatment responsive or non-responsive tumors. Scale bar: 50µm.

JAG1/NRR2 Inhibition induces progenitor-to-hepatocyte conversion

A) Summary of significant gene expression changes following JAG1 inhibition in regressing and non-regressing models. For LIV78 expression changes following NRR1 or NRR2 inhibition are shown. Significantly changed transcripts [log2 fold change ≥|1|; FDR<0.05] post single dose treatment are plotted. B) Differential expression of cell type or tumor-type specific signatures (Goossens et al., 2015b; Prior et al., 2019;et al., 2019). Changes are obtained from bulk RNA sequencing data for sensitive and insensitive models at 3d post single dose treatment with the indicated blocking antibody. For LIV#078 data is shown for multiple consecutive timepoints in case of treatment with aJ1.b70 vs control antibody. C) Four-way comparison of differential gene expression highlights significant overlap in upregulated genes in LIV78 and Li1035. D) Expression of NOTCH target genes 72 hrs after JAG1 inhibition in LIV78 and Li1035. N/A: not computed because of low expression values across all groups.E) GSEA of differential expression between aJ1.b70 and control antibody treated (72hrs) LIV78 tumors using MSigDB hallmark gene sets. NES–normalized enrichment score, FDR–false discovery ratio. F) Hematoxylin and eosin stained LIV78 tumors (17d). Scale bar: 100µm.

Single cell level analysis identifies HNF4A and CEBPA as Notch controlled transcriptional regulators responsible for progenitor to hepatocyte-like switch underlying treatment efficacy

A) UMAP plot of LIV78 tumor cell populations with cluster associations indicated. B) UMAP plot of LIV78 tumor cell populations faceted by treatment groups. C) Cluster distribution comparison of tumors with Notch signaling inhibition (aJAG1/NOTCH2) and control (agD) tumors. Cl, cluster. D) Identification of topic (Topic3) associated with Notch inhibition. Top 20 genes associated with topic3 (left). UMAP color-coded by Topic3 score (top right), topic3 score comparison between treatment groups (bottom right). E) Top 10 transcription factor activities that correlate positively or negatively with Topic3 program. F) Correlation of Topic3 score and HNF4A activity (top) or CEBPA activity (bottom) with treatment groups indicated. G) Correlation of transcription factor expression vs activity for HNF4A (left) and CEBPA (right) with treatment groups indicated. Boxplots provide summary of expression (x-axis) or activity (y-axis) changes with treatment. H) Volcano plot of differentially accessible regions in tumors treated with aNRR2 or control antibody (3d). Values with logFC>|1| and adjusted P-value <0.05 are highlighted in red and blue, respectively. I) Motif analysis in differentially accessible regions following NRR2 inhibition. The top 3 transcription factors with significantly differential activity values are shown.

Mechanism of Notch signaling inhibition-induced tumor cell differentiation.

In tumor cells with active NOTCH signaling, HES1 is expressed and represses transcription of CEPBA. In absence of CEBPA, HNF4A does not engage with chromatin regions associated with genes involved in hepatocyte differentiation and the program is repressed, locking the tumors cells in a proliferative, progenitor-like state (left panel). Conversely, following inhibition of the relevant JAG1 ligand or NOTCH2 receptor with specific blocking antibodies, HES1 is no longer available to repress CEBPA expression. CEBPA becomes available and enables HNF4A’s interaction with chromatin regions of genes functioning during hepatocyte differentiation (right panel). Initiation of hepatocyte differentiation is incompatible with tumor maintenance. Created with BioRender.com

A) Percentages of human and mouse reads in PDX tumors used for bulk RNA sequencing. Data is shown for control antibody treated Li1035 (n=4) or LIV78 tumors (n=5). B) Immunohistochemical detection of NICD1, NRR2 and JAG1 in PDX model LIV78 treated with control antibodies or antibodies blocking JAG1 or the indicated receptor. Scale bar: 100µm C) Immunohistochemical detection of NRR2 in treatment-insensitive PDX models treated with control antibodies or antibodies blocking JAG1. Scale bar: 100µm D) Reporter assay showing selective binding of human but not mouse JAG1-induced signaling using aJ1.b6A9.

A) Expression of cell type specific signatures (Segal et al., 2019) in sensitive and non-sensitive models summarized by k-means cluster. B) Histological evaluation of JAG1-inhibition sensitive (top row) or non-sensitive (bottom row) tumors from hematoxylin and eosin stained sections of PDX tumors. All tumors are moderately-poorly differentiated HCCs with variable nuclear grade. Scale bar: 100µm. C) Two-dimensional representation of TCGA liver cancer/normal transcriptomes using UMAP projection. Each sample is represented by one data point and colored by sample type (i), Hoshida HCC subtype (ii - iv), and GAS7 (v) or FGF9 (vi) expression.

A) Scatterplots of differential gene expression in anti-J1.b70 treated tumors compared to control antibody treatment. Transcriptional responses 72 hours following single dose treatment are shown for regressing (top two panels) and non-regressing (bottom three panels) HCC PDX models. B) Summary of significant gene expression changes in LIV78 over time following treatment with anti-J1.b70. For the 72 hours post treatment time-point changes following NRR1 or NRR2 inhibition are shown for comparison. In (A) and (B) significantly changed transcripts [log2 fold change ≥|1|; adj.p-value >0.05] post single dose treatment are plotted. C) Heatmap showing differential gene expression over time for significantly up-/down-regulated genes. Transcription factors are highlighted on the left hand side and a color key provides the pattern of change across time points, which is summarized in the panel on the right. D-F) Expression of NOTCH target genes 72 hours after JAG1 inhibition in models LIV78 and Li1035 determined by bulk RNA sequencing. G) GSEA of differential expression between aJ1.b70 and control antibody treated LIV78 tumors at 72 hours and 7 days post treatment. H) GSEA of differential expression between aJ1.b70 and control antibody treated Li1035 tumors 72 hours after treatment. MSigDB hallmark gene sets were used for analyzes shown in (A) and (B). ES: enrichment score, NE: normalized enrichment score, FDR–false discovery rate. I-L) Expression of markers specific to cholangiocytes and bi-potent liver progenitors. M) Cell cycle analysis of LIV78 tumors 5 days following aJ1.b70 or control antibody treatment using propidium iodine staining and flow cytometry. Cycle histograms are shown for a representative sample for each group. In addition to confirming a treatment-induced change in cell cycle progression, this approach also allowed establishing baseline ploidy, which differs across HCC tumors and may be used as prognostic marker (Bou-Nader et al., 2020), LIV78 tumor cells were homogeneously diploid at baseline, and JAG1 inhibition did not result in a ploidy change 5 days post treatment. N) Quantification of Ki67 immunofluorescence staining in tissue sections of Li1035 tumors 3 days following treatment with aJ1b70, aNRR1, aNRR2, or control antibodies. Average percentage of Ki67+ cells from 3 frames (20x) per tumor are shown (n=3 per group). Scale bar: 100µm. O) Quantification of BrdU staining in tissue sections of LIV78 tumors 7 days following treatment with aJ1b70, or control antibody. BrdU was administered 1.5h prior to tumor harvest. Representative images are shown for each group (panels on left; scale bar: 100µm). Average percentage of BrdU+ cells from 3 frames (20x) per tumor are shown (panel on right; n=4 for agD,(control), n=5 for aJ1.b70).P-R) Expression of genes involved in fatty acid metabolism. S) Expression of the hepatocyte marker ALB. Data shown in I-L, and P-S was obtained by bulk RNA sequencing 72 hours after JAG1 inhibition in models LIV78 and Li1035.

A) Expression of JAG1 and NOTCH2 in single cells arranged by UMAP cluster contribution and pathway inhibition status. aJ1.b70 and aNRR2-treated samples are summarized in one treatment group (aJAG1/aNRR2). B) UMAP plots of LIV78 tumor cell populations with JAG1 expressing cells (red) or NOTCH2 expressing cells (green) or cells expressing both JAG1 and NOTCH2 (purple) indicated. C) UMAP plots of LIV78 tumor cell populations with expression of common Notch pathway target genes indicated (blue). D) Neighborhood graph showing Milo differential abundance analysis between NOTCH inhibited and control samples. Each dot represents a neighborhood and is colored by its log fold change between conditions; circle sizes indicate the size of individual neighborhoods; graph edges represent the number of cells shared between neighborhoods. E) Beeswarm plot of the distribution of log fold change across clusters from figure 5A. Each dot shows one neighborhood. “Mixed” classification includes those neighborhoods belonging to multiple clusters. Colors represent the log fold change in abundance between NOTCH inhibited and control samples; neighborhoods with false discovery rate (FDR) > 0.1 are colored in gray. F) Proportion of cells contributing to individual UMAP clusters displayed for all individual samples G) Single cell expression of bulk RNA sequencing-derived signatures of significantly up-regulated (i,iii) or down-regulated (ii,iv) genes 3 days past JAG1(i,ii) or NRR2 (iii,iv) inhibition in LIV78 tumors. Single cell expression of each signature is shown in UMAP plots (left panels) and summarized for each cluster in violin plots (right panels). H) Single cell expression of cell type-specific signatures from Segal et al (Segal et al., 2019) in LIV78 tumors following Notch inhibition or control antibody treatment. UMAP plots and expression summaries for each single cell cluster as violin plots are shown for expression of a human fetal progenitor (left) and human fetal hepatocyte signature (right). I) AIC and BIC scores across a range [2-15] of topics. K=9 topics were determined to be the most informative. J) Topic modeling of LIV78 single cell data. For each topic a UMAP plot color-coded by the topic score (top left), a comparison of the topic score between treatment groups (bottom left) and the top 20 genes associated with the topic (right) are shown. K) Immunofluorescence detection of HNF4A expression on tissue sections from aJAG1.b70 or control antibody-treated Li1035 tumors (3days). Scale bar: 50µm. L-M) Single cell expression of HNF4A (L) or CEBPA (M) is shown in UMAP plot (left panels) and summarized for each cluster in violinplots (right panels). N) Centrimo analysis showing the location-specific preferential enrichment of HNF4A and CEBPA motifs. HNF4A binding shows centrally located enrichment, while CEBPA binding motifs show lateral enrichment in chromatin regions that are opening following NRR2 inhibition. O) Genome tracks of ATAC-sequencing signals in anti-NRR2-treated and control antibody treated LIV78 tumors across HNF4A- and CEBPA-bound promotors of genes associated with hepatocyte function (left panel) and their corresponding RNA-seq expression levels (right panel). P) Genome tracks of ATAC-sequencing signals in anti-NRR2-treated and control antibody treated LIV78 tumors in the SOX-transcription factor-bound promoter region of the progenitor marker NES (left panel) and corresponding RNA-seq-expression levels (right panel).