EGFR806-CAR T cells selectively target a tumor-restricted EGFR epitope in glioblastoma

Targeting solid tumor antigens with chimeric antigen receptor (CAR) T cell therapy requires tumor specificity and tolerance toward variability in antigen expression levels. Given the relative paucity of unique cell surface proteins on tumor cells for CAR targeting, we have focused on identifying tumor-specific epitopes that arise as a consequence of target protein posttranslational modification. We designed a CAR using a mAb806-based binder, which recognizes tumor-specific untethered EGFR. The mAb806 epitope is also exposed in the EGFRvIII variant transcript. By varying spacer domain elements of the CAR, we structurally tuned the CAR to recognize low densities of EGFR representative of non-gene amplified expression levels in solid tumors. The appropriately tuned short-spacer 2nd generation EGFR806-CAR T cells showed efficient in vitro cytokine secretion and glioma cell lysis, which was competitively blocked by a short peptide encompassing the mAb806 binding site. Unlike the nonselective Erbitux-based CAR, EGFR806-CAR T cells did not target primary human fetal brain astrocytes expressing wild-type EGFR, but showed a similar level of activity compared to Erbitux-CAR when the tumor-specific EGFRvIII transcript variant was overexpressed in astrocytes. EGFR806-CAR T cells successfully treated orthotopic U87 glioma implants in NSG mice, with 50% of animals surviving to 90 days. With additional IL-2 support, all tumors were eradicate without recurrence after 90 days. In a novel human induced pluripotent stem cell (iPSC)-derived teratoma xenograft model, EGFR806-CAR T cells infiltrated but were not activated in EGFR+ epidermal cell nests as assessed by Granzyme B expression. These results indicate that EGFR806-CAR T cells effectively and selectively target EGFR-expressing tumor cells.

The 5 cell lines processed for RNA-Seq analysis are listed in Supplementary Table 1. Numbers for mapped reads to the reference as well as only the EGFR locus (or engineered EGFRvIII) are shown in Supplementary  Table 1. The epidermal cell line A431 has the highest relative expression of EGFR, which correlates with its high expression levels as measured by Western blotting and surface analysis ( Figure 2). A431, T98, U251 and U87 cell lines express wild type EGFR but not EGFRvIII. Raji cells transduced with EGFRvIII served as controls and only expressed EGFRvIII and not wild type EGFR transcripts. As follows from Table S1, our alignment strategy sufficiently distinguishes reads mapping to EGFR versus EGFRvIII.
According to GRCh38.p12 annotation, the EGFR gene is composed of 59 exons that generate 11 transcript forms. Supplementary Figure 4 shows a schema of the 5' region of the EGFR locus. Concurrent exons were consolidated to yield exonic (or coding) regions that are partitioned by intervening introns.
TapeStation trace analysis of the RNA-Seq library region gave an approximate range of 200-720 bp. This includes sequencing adapters that add up to 121 bases, so the actual template sizes are between 80-600 bp. As can be seen from Supplementary Figure 4, EGFRvIII has a defining feature, in that it is the only transcript with exonic region 12 adjacent to region 1 (per GRCh38.p12 annotation). Major transcripts 1-6 also incorporate both regions 1 and 12; however, the intervening coding regions add up to 666 bases for transcripts 1 and 3, which skip region 11, and 801 bases for the others (3).
Sequence analysis was accomplished by capturing all reads originating in region 1 (upstream anchor reads) then interrogating if the corresponding mate pair reads mapped to region 12 (or further downstream). Supplementary Figure 5 shows the distribution of mate pair mappings to exonic regions. As observed, the highest proportions of mate pairs map to exonic region 5, mirroring the transition from region 1 to 5 for major transcripts 1-6 (Supplementary Figure 4). The progressive decrease in mate pair frequency downstream of exonic region 5 (regions 6 and 7) follows the template size distribution for sequencing, as expected. Given our template sizes sequenced, with a maximum of approximately 600 bp, it would be very unlikely for any fragmented template sourced from transcripts 1-6 to contain both regions 1 and 12 (or regions downstream of 12). Thus, any template featuring the expected transition from region 1 to 12 in EGFRvIII that would support the existence of this variant, would have been detected sensitively by our approach. Importantly, there is no evidence of such transitions from region 1 to 12 (Supplementary Figure 4), and therefore no detectable levels of EGFRvIII variant in any of the cell lines analyzed, using this extremely sensitive sequencingbased analysis.

RNA-Seq
For each cell line, 5 × 10 6 cells were washed in PBS, pelleted, snap frozen and stored at −80°C. Total RNA was isolated using the RNeasy Mini kit (Qiagen) according to the manufacturer's recommendations and quantified with NanoDrop 2000 (Thermo Fisher Scientific, Waltham, MA). Total RNA integrity was assessed on TapeStation 4200 using R6K Screen Tape (Agilent Technologies Inc., Santa Clara, CA). RIN values obtained were above 9, indicating good RNA preparations. Additional quantitation for RNA-Seq library preparation was performed using a Trinean DropSense96 spectrophotometer (Caliper Life Sciences, Hopkinton, MA).
RNA-Seq libraries were prepared from total RNA with TruSeq RNA Sample Prep v2 Kit (Illumina Inc., San Diego, CA, USA). Library size distributions were validated using the D1000 Screen Tape on TapeStation 4200 (Agilent Technologies, Santa Clara, CA, USA). Additional library QC, blending of pooled indexed libraries, and cluster optimization was performed using Invitrogen Qubit ® 2.0 Fluorometer (Life Technologies-Invitrogen, Carlsbad, CA, USA). RNA-seq libraries were pooled and clustered onto a S2 flow cell for sequencing on Illumina NovaSeq 6000 targeting a 110 × 90 base pairedend read length. Library preparation and sequencing was performed at the Fred Hutch Genomics Core, Seattle, WA.

Alignment of RNA-Seq reads
Genome sequence and comprehensive gene annotation pertaining for EGFR was obtained from GENCODE, Human Genome Reference GRCh38.p12 (https://www.gencodegenes.org/human/). The engineered EGFRvIII transcript sequence was appended to the genome reference. Sequenced reads were then aligned to this composite reference with STAR version 2.5.3a using a 2-pass mapping strategy [1,2]. Starting from aligned reads in bam files, subsequent analysis was executed in R leveraging the Bioconductor framework (http:// bioconductor.org/). Analysis was constrained to reads overlapping the EGFR gene on chromosome 7.