NRCAM variant defined by microexon skipping is a targetable cell surface proteoform in high-grade gliomas

Abstract

To overcome the paucity of known tumor-specific surface antigens in pediatric high-grade glioma (pHGG), we contrasted splicing patterns in pHGGs and normal brain samples. Among alternative splicing events affecting extracellular protein domains, the most pervasive alteration was the skipping of ≤30-nt-long exons. Several of these skipped microexons mapped to L1-immunoglobulin cell adhesion molecule (IgCAM) family members, such as neuronal CAM (NRCAM). Bulk and single-nuclei short- and long-read RNA-seq revealed uniform skipping of NRCAM microexons 5 and 19 in virtually every pHGG sample. Importantly, the Δex5Δex19 (but not the full-length) NRCAM proteoform was essential for pHGG cell migration and invasion in vitro and tumor growth in vivo. We developed a monoclonal antibody selective for Δex5Δex19 NRCAM and demonstrated that "painting" pHGG cells with this antibody enables killing by T cells armed with an FcRI-based universal immune receptor. Thus, pHGG-specific NRCAM and possibly other L1-IgCAM proteoforms are promising and highly selective targets for adoptive immunotherapies.

Keywords: CP: Cancer; CP: Immunology; alternative splicing; antibodies; cell adhesion molecules; glioblastoma; glioma; immunotherapy; mRNA processing; microexons.

Figure 1.. Aberrantly spliced surface protein-encoding transcripts in pHGG

(A) The pipeline used to identify pHGG-specific splicing events. (B) Difference in exon inclusion level (ΔPSI) between pHGG and normal brain samples (y axis). Splicing events were binned based on exon lengths (x axis). Small box indicates the only bin with >50% difference in absolute PSI values. (C) Bar graphs showing the expression of select splicing factors in normal brain tissues vs. pHGG. Each bar represents mean ± SEM. (D) Scatterplot showing ΔPSI values, event prevalence, and transcripts-per-million (TPM) metrics for select microexon-containing pHGG mRNAs. (E) Top aberrantly spliced pHGG microexons. (F) Graphical output of MAJIQlopedia showing average PSIs of NRCAM exons 5 and 19 across tumor types represented in TCGA and CBTTC repositories. (G) Visualization of the ENST00000351718.8 transcript and its expression level in normal brain (purple) vs. Kids First Brainstem glioma (teal) using the Xena portal.

Figure 2.. HGG-specific expression of Δex5Δex19 NRCAM isoform

(A) Boxplots showing average PSI values for exon 5 and exon 19 in NRCAM across normal brain cell types. Horizontal lines correspond to median values. (B) (Left) Boxplots showing expression of mRNA encoding the established B7-H3 antigen in indicated GTEx tissues (neural and non-neural) and pHGG samples. Read counts corresponding to a constitutive exon-exon junction were used to estimate transcript levels and were plotted as junctions per million on the y axis. (Right) Expression of the Δex5 isoform of NRCAM mRNA (the exon 4-exon 6 junction) in the same samples. Horizontal lines correspond to median values. (C) Validation of NRCAM exons 5 and 19 skipping in PDX7316–1769 by direct cDNA long-read RNA-seq. Reads extending from the 3′ to the 5′ ends of the NRCAM transcript were visualized in IGV and depicted as sashimi plots. (D and E) Validation of NRCAM exons 5 and 19 skipping and inclusion by targeted long-read RNA-seq in KNS42 and SMS-SAN cells, respectively. (F) Stacked plot showing estimated abundances of NRCAM Δex5 and Δex19 transcripts as measured in counts per million (y axis) across pHGG cell lines (KNS42) and PDXs, GBM cell lines (TM31 and U251), and a neuroblastoma cell line (SMS-SAN). ΔexΔ5ex19* represents transcripts with additional alternative splicing events besides the skipping of exons 5 and 19.

Figure 3.. Single-nuclei analysis of PDX 7316–3058

(A) Short-read RNA-seq analysis of >3,700 nuclei sequenced using the 10× Genomics platform. Uniform manifold approximation and projection (UMAP) plots were generated using Cell Ranger and colored by cell-type annotation (i). Additional UMAP plots of the same cells were colored by expression of individual genes using Loupe Browser (ii–iv). (B) Long-read RNA-seq analysis of full-length transcripts. ScisorWiz plots show reads spanning exons of interest in NRCAM and CHL1 genes (dotted rectangles) and grouped by clusters. Each horizontal line indicates one transcript; thick blocks denote exons, and thin lines denote introns (not drawn to scale to aid visualization). The bottom images show annotated GENCODE transcripts.

Figure 4.. Aberrantly spliced NRCAM proteoforms in pHGG

(A) Immunoblotting showing NRCAM levels in KNS42 pHGG cells, wild-type (Ctrl) or transfected with NRCAM-targeting gRNAs (KO). GAPDH served as loading control. (B) Immunoblotting showing levels of full-length (FL) and Δex5Δex19 NRCAM isoforms in cell-surface and flow-through fractions and total cell lysates of KNS42 Ctrl and NRCAM KO cells. EGFR, tubulin and actin were used as internal controls. (C) Immunoblotting showing NRCAM levels in KNS42 NRCAM KO cells reconstituted with the FL or the Δex5Δex19 NRCAM isoform. (D) Proliferation rates of NRCAM KO cells expressing the FL or the Δex5Δex19 NRCAM isoform. (E) Migration (top) and invasion (bottom) potential of NRCAM KO cells re-expressing FL and Δex5Δex19 isoforms of NRCAM. The Transwell assays were performed without or with Matrigel layers, respectively. (F) Quantitation of migrated and invaded cells from (G). Each bar represents mean ± SEM. Significance (asterisks) was determined using an unpaired Student’s t test. (G) Optical imaging of the same cells additionally engineered to expressed firefly luciferase and orthotopically injected into the cortices of NSG mice. 3 mice out of 8 were imaged on day 28. (H) Raw photon counts corresponding to tumors in (G). Each bar represents mean ± SEM. (I) Kaplan-Meier curves reflecting survival of mice depicted in (G), with p value determined by log rank (Mantel-Cox) test.

Figure 5.. Detection of NRCAM proteoforms by mAb 3F8

(A) AlphaFold models of the ectodomains of the canonical (left) and the Δex5Δex19 (right) NRCAM isoforms, with the signal peptide (amino acids 1–24) removed. Ectodomain and signal peptide were identified using UniProt annotations. Ig-like domains are shown in green, fibronectin type III domains in purple, and microexons 5 and 19 in red. (B) Schematic showing the pipeline for antibody production. (C and D) Flow cytometry histograms showing (C) 2D10 and (D) 3F8 mAb binding profiles when used on live CHO-K1 cells expressing “empty vector” (target null, blue), full-length (FL) NRCAM (unintended target, green), and Δex5Δex19 NRCAM (intended target, red). Control stainings with secondary (2° ) antibody only are shown for comparison. (E) Same staining performed on live KNS42 cells endogenously expressing Δex5Δex19 NRCAM (intended target, purple) or with the entire gene knocked out using CRISPR-Cas9 (target null, green). In both images, orange arrow points to the signal generated by Δex5Δex19 NRCAM. (F) Flow cytometry histograms showing 3F8 binding to various patient-derived pHGG cells, compared to secondary-antibody-only-stained samples (2°). (G) The strategy to test the therapeutic utility of the 3F8 antibody against glioma cells expressing the splice isoform of NRCAM. The composition of FcγRI-based UIR is shown on the left. Non-neoplastic cells expressing the FL isoform of NRCAM are depicted at the top right as being presumably resistant to 3F8 UIR treatment.

Figure 6.. mAb 3F8-mediated killing of glioma cells

(A) Killing of PDX3058 and KNS42 cells expressing indicated NRCAM isoforms. (B) Killing of adult glioblastoma U251 and TM31 cells and their NRCAM KO derivatives. (A and B) Shown on the x axis are Ab concentrations (in mg/mL), and on the y axis, the extent of tumor cell killing, as evidenced by reduced luciferase expression. “E:T” values refer to the ratio of effector (T) to target (glioma) cells. Two-way ANOVA was used for comparison across different treatment conditions and groups. ****p < 0.00001, ***p < 0.001, **p < 0.01, *p < 0.05.