Response prediction in patients with gastric and esophagogastric adenocarcinoma under neoadjuvant chemotherapy using targeted gene expression analysis and next-generation sequencing in pre-therapeutic biopsies

Formalin-fixed paraffin-embedded (FFPE), pre-therapeutic biopsies from 36 Caucasian patients diagnosed 2009–2016 for AEG/AS with different histopathological regression grades according to Becker et al. (2003) (Becker-1, n = 11; Becker-2, n = 7; Becker-3, n = 18) toward completed neoadjuvant FLOT or FLOT-like (oxaliplatin substituted by cisplatin; 5-FU substituted by capecitabine) chemotherapy were included in this study (Table 2). Patients initial clinical stage was ≥ cT2 or/ and cN + cM0. The tumor cell content of the pre-therapeutic biopsy samples was between 10 and 90% and no microdissection was performed. The study was approved by the Local Ethics Committee of the Charité—Universitätsmedizin Berlin (Approval No. EA4/075/15).

Genomic DNA and total RNA was extracted from FFPE biopsies employing the Maxwell 16 System DNA IQ Casework Pro Kit or, respectively, Maxwell 16 System RNA IQ Casework Pro Kit, each according to the manufacturer’s protocols (Promega Corporation, Madison, USA) as previously described (Bonnet et al. 2018; Vlahovic and Kubat 2012). Extracted DNA and RNA were quantified with the Qubit System using the Qubit DNA Assay or, respectively, Qubit RNA Assay (ThermoFisher Scientific).

To detect Epstein–Barr virus (EBV)-encoded RNA (EBER), we performed an in situ hybridization (ISH) assay as already described (Anagnostopoulos et al. 1992; Fan and Gulley 2001). A ready-to-use EBER-ISH probe (BOND #PB5089) was used together with the Leica Bond-maX autostainer (Leica Biosystems, Illinois, USA) according to the standard BOND protocol. Immunohistochemical staining (IHC) was performed using sections derived from FFPE primary tissue samples with the help of the Leica Bond-maX autostainer (Leica Biosystems, Illinois, USA) according to the manufacturer’s instructions. After heat-induced epitope retrieval, the sections were incubated with anti-HER2/neu (4B5) (VENTANA—#790-4493) or anti-FGFR2 antibody (Sigma Aldrich—#HPA035305). Horseradish peroxidase-labeled anti-rabbit using the Bond Polymer Refine Detection Kit (Leica Biosystems, Illinois, United States) was employed to convert the chromogen substrate. Staining was performed with appropriate positive and negative controls. Evaluation of staining intensity was performed by a trained pathologist according to four level scoring (0—no staining, 1—light staining, 2—moderate staining, 3—strong staining) as previously described (Jia et al. 2016).

For the detection of microsatellite instability, a fluorescent PCR-based assay (MSI Analysis System Version 1.2, Promega) was used according to the manufacturer's instructions. The MSI Analysis System included fluorescently labeled primers for co-amplification of seven markers including five mononucleotide repeat markers (BAT-25, BAT-26, NR-21, NR-24 and MONO-27) and two pentanucleotide repeat markers (Penta C and Penta D). Additionally, an Internal Lane Standard 600 (ILS 600, Promega) was used. In brief, 20 ng of genomic DNA was used for the PCR under following conditions: initial denaturation for 11 min at 95 °C and 1 min at 96 °C, ramp 100% to 94 °C for 30 s, ramp 29% to 58 °C for 30 s, ramp 23% to 70 °C for 1 min, for 10 cycles, followed by ramp 100% to 90 °C for 30 s, ramp 29% to 58 °C for 30 s, ramp 23% to 70 °C for 1 min, for 20 cycles, and final elongation at 60 °C for 30 min. The PCR products were separated by capillary electrophoresis using a 3500 Genetic Analyzer. Data were analyzed with the GeneMapperR software (Applied Biosystems) by comparing sample marker patterns with the pattern of a positive amplification control (K562 High Molecular Weight DNA). A shift of two or more mononucleotide markers was considered as MSI-high, a shift of one mononucleotide marker was considered as MSI-low.

RNA samples (n = 36) underwent target-specific expression analysis using the NanoString nCounter gene expression system (NanoString Technologies, Seattle, WA, USA) as has been described previously (Geiss et al. 2008; Perez et al. 2019; Daum et al. 2020; Kulkarni 2011). nCounter PanCancer Immuno Profiling Panel (Cesano 2015) and the PanCancer Pathway Panel (Omarini et al. 2018) were employed to assess the expression level of a total of 1390 genes, including endogenous controls. The nCounter assay analysis was performed as previously described using 300 ng of total RNA, based on the manufacturer’s protocol (Kulkarni 2011).

Next-generation sequencing (NGS) was performed to identify genomic aberrations within genes of TruSight Tumor 170 (TST170) DNA and RNA assay (Illumina, Catalog no. 20028821). The enrichment-based targeted TST170 panel covers the coding regions of 170 tumor hot spot genes and allows the identification of SNPs, CNVs, and fusion genes. For TST170 analysis, DNA and RNA samples from the same tissue samples were used resulting in a total 24 biopsy samples (Becker-3, n = 14; Becker-2, n = 4; Becker-1, n = 6).

For this purpose, DNA and RNA concentrations were measured by fluorometric quantification on Qubit 3.0 (Thermo Fisher Scientific, Inc.) utilizing the Qubit Broad Range DNA assay and the Qubit Broad Range RNA assay (Thermo Fisher Scientific, Inc.), respectively. The quality of the total RNA was assessed by analyzing all samples on a TapeStation 4200 system with the RNA ScreenTape (Agilent Technologies) and the DV200 value was calculated for all samples. Only samples resulting in DV200 values of ≥ 20% were included in the library preparation process. RNA was diluted to 7 ng/µl in nuclease-free water and 8.5 µl of diluted RNA was processed according to the manufacturer’s recommendations. Genomic DNA was diluted to 7 ng/µl in TEB (Illumina, Inc.) and 12 µl of diluted samples were further diluted with 40 µl TEB for fragmentation. All DNA samples were fragmented to 90–250 bp (with a peak at ~ 125 bp) in microTUBE-50 AFA Fiber Screw-Cap tubes (PN: 520166; Covaris, Ltd.) using the Covaris ME220 system (Covaris, Ltd.) with the following settings: 75 W peak incident power, 25% duty factor, 1000 cycles/burst, 225 s. Further sample processing was performed for fragmented genomic DNA and purified cDNA simultaneously according to the instructions of the manufacturer. Prior to normalization, libraries were quantified by Qubit 3.0 and the Qubit High Sensitivity DNA assay (Thermo Fisher Scientific, Inc.). Libraries with concentrations ≥ 3 ng/μl qualified for bead-based normalization and pooling as well as sequencing. A total of eight samples, eight RNA libraries and eight DNA libraries originating from the same FFPE tissue were sequenced on one high-output NextSeq 500 (Illumina, Inc.) paired-end run with 101 cycles per read (2 × 101) and 8 cycles per index read.

For a general overview of the mutation status and to identify potential classifiers between different responder and non-responder groups, a bioinformatic pipeline powered by Bioinformatics Solution Center (BSC) was used. Briefly, the in house pipeline starts with raw reads data (FASTQ files), followed by quality control (QC) and trimming step, hg19/GRCh37 mapping using hisat2 2.1.0 program and finishing with variant calling by freebayes 1.3.1 program (Garrison and Marth 2012; Kim et al. 2019a).

Although to the small number of cases in our collective, we performed a statistical analysis for different data sets. To evaluate the statistical significance of immunohistochemical staining intensity for FGFR2 and HER2 expression between Becker groups we using the t test with Welch's correction (two-tailed) for unpaired comparisons. A p value less than 0.05 were considered as significant.

The Nanostring RNA expression data set was used with nSolver Software 4.0 using the Advanced Analysis Version 2.0.115 provided by NanoString. To calculate differential expression patterns, the samples were grouped according to their Becker scores (fringe groups: Becker-1 vs. Becker-3). The p value calculation was performed without adjustment and statistical significance was accepted at p < 0.01.

To determine differences within the total number of detected SNV (TST170 Data set), we performed in addition a Kruskal–Wallis test (Kruskal and Wallis 1952). It has to be noted that p value adjustment by false discovery rate procedure of Benjamini and Yekutieli was renounced, because of the small number of samples within this collective.

We also determined the MSI and EBV status among the Becker groups and observed a random distribution among the patients (Fig. 2). However, both are rare events, only 1 of 32 investigated pre-therapeutic biopsies were EBV positive and only 4 of 29 were MSI-high.

As expected, analysis of the TST170 gene panel data revealed that individual molecular events are unable to predict the treatment response in pre-therapeutic biopsies. However, the total number of detectable SNV within the Becker-3 group is significantly higher compared to Becker-1 (p < 0.1818) (Fig. 2). Interestingly, we identified 7 different CNV events affecting the genes for FGFR2, EGFR, FGF3, FGF4, CCBD1, FGF19 and FGF10, which are also exclusively present in Becker-3 patients (f5 out of 14). This tendency of accumulation of genomic events in patients of the Becker-3 group (seven fusion events in six patients: DIAPH3-FLT1, ERBB2-EGFR, FGFR2-TACC2, EML4-BRCA1, GLYCTK-ALK, KIF5B-ARHGAP21 and JAK2-LCOR) was furthermore demonstrated for chromosomal translocations as depicted by RNA-NGS for the detection of the corresponding fusion transcripts (Fig. 2).

To see if any gene is more frequently mutated in one of the Becker groups, the total number of SNV per gene was calculated within each sample and within the groups (Fig. 3). Top mutated genes were ETV1, SMO, FANCL, NBN and EP300 the least represented genes were MYD88, MCL1 and CEBPA (data not shown). In Fig. 4 the top five genes are given together with unadjusted p values and the total counts of SNPs in these genes were shown.

To have a general overview of similarity or clustering/stratification among individual genotypes, the principal component analysis (PCA) was undertaken in PLINK 1.9 software. Top principal components were extracted from the variance-standardized relationship matrix of genotypes. Genotype was called for all SNV loci, imputed as homozygous reference allele when alternative allele was not called in a sample. However, no clustering of samples according to Becker grouping was identified, but only (small) deviations of several samples from the central cluster cloud could be detected (Appendix Fig. 8).

For comparison between Becker groups, the variant call files (VCF) were merged, and only biallelic SNV loci were kept. Total numbers of loci that were used and association between SNPs and Becker groups was calculated in PLINK by 1df Chi-square allelic test (standard case/control association analysis). The results of association of alleles with specific Becker groups are shown in Fig. 5. The Manhattan plots yield four significant SNP locus (MSH6, DDR2, EP300 and FGF3 gene) that differ between responder (Becker-1) and non-responder (Becker-3) group (Fig. 5).

For RNA expression, we performed a target-specific analysis using the NanoString nCounter platform. Two panels (PanCancer Pathway Panel and PanCancer Immuno Profiling Panel) consisting of a total of 1390 target genes were used to identify differentially expressed genes (Fig. 6). Principal component analysis (PCA) demonstrates no clustering (Fig. 6A, D). Among the investigated genes no significant differential expression was observed between responder (Becker-1) and non-responder (Becker-3) groups (Fig. 6B, E). Top differentially expressed genes are summarized in Fig. 6C, F. However no significant differences in RNA expression were observed, rather the higher ERBB2 expression (Log2 fold change 2,9) in Becker-1 group was affected by high ERBB2 expression in two individual Becker-1 samples.

To get a general impression of RNA expression within the whole collective, we downsampled the NanoString data to 24 samples, and expression analysis in the DESeq2 package (in R 4.0.2) was done for NanoString (770 genes) data and TST170 (55 genes) data. The matrix of NanoString and TST170 experiment raw expression data were imported to R for DESeq2 analysis. DESeq2 statistical model internally corrects/normalizes the data (Love et al. 2014).

To explore the count matrix, the top 20 expressed genes heat map was produced with the clustering of samples for NanoString Data Set (Fig. 7A) and TST170 Data Set (Fig. 7B). Similar to the previous analysis in nSolver software by NanoString, clusters did not represent Becker groups. However, it can be seen that in B2M-, RPS27A-, KMT2A and NOTCH2-gene represent the highest expressed genes within the whole collective.