Germline genetic biomarkers to stratify patients for personalized radiation treatment

This retrospective analysis was approved by our institutional review board (UCSD HRPP#200495). Thirty-seven HNSCC patients who underwent Tempus xT somatic tumor testing paired with normal matched specimens and received RT with available dosimetric data were selected for this study. All selected patients in our study cohort (n = 37) were treated at the Moores Cancer Center at the University of California San Diego between 2009 and 2021.

The CAP/CLIA validated Tempus xT test is ordered by a clinician to provide predictive, prognostic, and therapeutic management for patients. Patients were consented for the test in accordance with federal, state, University, and UCSD Human Research Protection Program policies. Normal matched specimen sources accepted for testing include blood or saliva, collected at time of ordering the test.

We acquired patient characteristics and radiation data for these 37 patients including age at diagnosis, gender, smoking history, and human papillomavirus (HPV) status. Treatment parameters including pre-RT surgical resection, radiation dose, and induction/concurrent systemic therapy were recorded as well. Staging information was collected according to the American Joint Committee on Cancer (AJCC) classification edition in effect at the time of diagnosis, ranging from 6 to 8th edition [24].

Patient charts were utilized to report early and late rtAE endpoints for mucositis, dysphagia and xerostomia. Toxicities were recorded using Common Terminology Criteria for Adverse Events (CTCAE) v.5.0, which were scored and reported by the treating physician on the day of service during therapy and in follow up. Early toxicity endpoints were recorded as the highest CTCAE grade experienced during therapy or within 6 weeks of completing therapy. Late toxicity endpoints were recorded as the highest CTCAE grade experienced from 6 months post-RT to the time of most recent follow up.

The relationship between categorical variables and each outcome was analyzed with a chi-square test where p-values ≤ 0.05 were significant. Differences in patient characteristics were compared using Fisher’s exact test or chi-square as appropriate. The effect size was calculated as the ratio of variant to wild type samples in group A divided by the ratio of variant to wild type samples in group B: (Mut A/WT A)/(Mut B/WT B). Differences in OS, locoregional recurrence (LRR), locoregional failure free survival (LFS), progression free survival (PFS), and metastasis free survival (MFS) were compared between Group A and Group B as well as Group C and group D by Kaplan–Meier (KM) survival analysis with log-rank testing for significance. Analysis was performed using SPSS V22.0 [25].

Ribonucleic acid (RNA) sequencing (RNA-seq) FastQ files were processed for alignment and quality control. The reads were trimmed, and low-quality reads were removed using Trimgalore v 0.6.3_dev [26] with the “paired” parameter and length of 76 bps. Trimmed FastQ sequences were aligned to the human reference genome (GRCh38) using STAR aligner v2.7.1a. Bam files were sorted by coordinate by using option “–outSAMtype BAM SortedByCoordinate”. Alignment quality control (QC) and read mapping statistics were obtained from Picard v2.20.3 tools using function “CollectMultipleMetrics” [27]. FastQC v0.11.8 was used to perform QC checks on the raw sequencing data.

Variant genotypes per patient were first validated in DNA sequencing samples using IGV [28]. Each patient’s total depth and allele frequency were recorded and an empirical assessment of genotype was made. If 100% of reads supported an alternative variant in germline DNA sequencing then the patient was deemed homozygous for that particular variant. If < 100% and > 0% of germline reads supported the variant, the patient was deemed heterozygous. Otherwise, the patient was deemed homozygous for the reference allele. This process overrode variant caller germline categorizations where disagreement was present.

Each genotype was annotated with total RNA read depth and RNA variant allelic fraction (VAF). Bam-readcount [29] was used to measure the depth and VAF of alternate alleles at the locus of each respective variant for each patient.

GnomAD [30] was queried to determine the population minor allele frequency (MAF) of variants of interest. Population MAF was then broken into major ethnic groups as follows: European non-Finnish, African/African American, Latino/Admixed American, Ashkenazi Jewish, South Asian, and East Asian.

The variant protein sequence was deposited in the Baker lab folding algorithm [31]. The deduced model was presented by ChimeraX [32].

For optimal assessment of normal tissue function associated with RT toxicity, we build a next generation sequencing (NGS) database that included somatic tumor and paired normal tissue [33, 34]. NGS was performed using the same sequencing platform for all samples to maximize consistent data output for discovery of germline variants implicated in RT toxicity. Our dataset included 37 HNSCC patients (n = 37) with NGS data; 34 of which were HNSCC patients treated with curative intent and analyzed for clinical outcomes. Their characteristics are summarized in Table 1. Excluded patients include one with metastatic disease at diagnosis and two with locoregionally advanced cutaneous SCC. Metastatic disease was excluded due to inability to evaluate time to certain outcomes, such as metastasis free survival. Patients with cutaneous SCC were excluded due to differences in treatment paradigm, disease course and prognosis. While these patients were not evaluated for clinical outcomes they were retained for NGS analyses to assess for potential biomarkers associated with differing radiation AEs.

There are previously reported findings in HNSCC pointing to differential tissue response in acute and late toxicity, where acute toxicity is associated with inflammation while necrotic developments may drive late toxicity outcomes [35]. To isolate contributing genetic factors in early and late responses, the patient cohort was divided into two groups (Group A and Group B) based on their toxicity profile. Specifically, for this grouping, patients in Group A had significant increase in late toxicity while patients in Group B did not experience a late stage change in toxicity. Based on these criteria 18 patients were designated as Group A and 19 patients were designated as Group B.

We tested for germline variant association with RT toxicity outcomes, restricting to germline variants that were present in at least 25% MAF with significant p-value ≤ 0.05 between Groups A and B. Among the top 30 germline variants for Groups A and B, we identified 5 that met our selection criteria: human leukocyte antigen (HLA-A), tuberous sclerosis complex 2 (TSC2), marker of proliferation Ki-67 (MKI67), interferon induced protein with tetratricopeptide repeats 2 (IFIT2), and interleukin 10 receptor subunit alpha (IL10RA).

We found 13 patients representing 37.1% of our cohort that carried the HLA-A Arg68Lys/Val91Met variant with 10 patients in Group A and 3 in Group B (p-value = 0.052, Table 2).

All patients with the HLA-A Val91Met variant (rs79361534) also carried the HLA-A Arg68Lys variant (rs707910). This particular variant is documented in the HLA-A polymorphism database with a low MAF in a population of 1000 [36]. We found new characteristics of this HLA-A variant that have not been reported previously. This particular variant co-caries two isolated single nucleotide changes resulting in two amino acid substitutions at protein location 68 and 91. This has not been reported before since the two changes are listed as occurring separately and resulting in two HLA-A polymorphism genes (rs79361534 and rs707910). To examine the impact of these changes on the structure and function of the encoded protein, we performed a 3D modeling analysis (Fig. 1). We found that the valine substituted by methionine at position 91 is situated in the protein peptide binding groove consisting of two α-helixes and thus it is reasonable to conjecture that this amino acid substitution could be a factor in modified peptide binding affinity resulting in neo-antigen presentation. The Arginine substitution with Lysine at position 68 may introduce a new ubiquitin binding site, in addition, this variant was predicted as deleterious by using the Sorting Intolerant from Tolerant (SIFT) algorithm [37]. This HLA variant was not significantly associated with differences in PFS (p = 0.402), LRR (p = 0.173), MFS (p = 0.769) or OS (p = 0.757, Table 3). The effect size for this variant was high (6.67) signifying higher odds of late RT toxicity and potentially serving as a biomarker for RT stratification.

Our data analysis identified an intronic C>T TSC2 variant (rs1800720) flanking a splice donor site and occurring in 10 patients (28.6%) in our n37 cohort, with a trend towards a significant enrichment in Group A. The variant was detected in 8 Group A patients versus 2 in Group B (p = 0.0577, Table 2). This TSC2 variant was particularly enriched in our cohort (MAF = 0.189) relative to the gnomAD data (MAF = 0.0968) [30]. Interestingly, the MAF in our cohort closely aligns with the African/African American gnomAD population (MAF = 0.202).

This specific TSC2 splice region is identified as an associated germline variant in tuberous sclerosis syndrome. TSC2 in complex with TSC1 has tumor suppressor functions via regulation of the mechanistic target of rapamycin kinase (mTOR) signaling pathway [38]. Aberrant activation of the mTOR pathway has been widely implicated in HNSCC [39] and it is possible that this TSC2 variant is associated with HNSCC via effects on the mTOR pathway. Clinical outcome KM analysis for this variant is associated with lower risk of LRR (p = 0.018), however no association was found with PFS (p = 0.216), MFS (p = 0.702) and OS (p = 0.685, Table 3). This is the first time to our knowledge that the finding of this germline variant is reported in HNSCC.

We detected 9 germline variants of significance associated with early versus late toxicity that did not meet our biomarker selection criteria due to less than 25% frequency of occurrence (Table 2). These were Fanconi anemia complementation group D2 (FANCD2) pPro714Leu (rs3864017); dynein cytoplasmic 2 heavy chain 1 (DYNC2H1) pGly304Leu (rs12146610; rs72989734); protein phosphatase 1 regulatory subunit 15A (PPP1R15A) pAla32Thr (rs3786734); SLX4 structure-specific endonuclease subunit (SLX4) pSer71Asn, pAsn457Lys, pArg204Cys (rs77985244, rs74319927, rs97842542); major histocompatibility complex, class II, DQ alpha 2 (HLA-DQA2) splice site (rs4398729); major histocompatibility complex, class II, DM beta (HLA-DMB) pSer71Asn, pAsp49Val, pThr28Ala (rs17617321, rs17617333, rs17583782); major histocompatibility complex, class II, DM alpha (HLA-DMA) pArg210His, pGly181Ala (rs41555121, rs6926628); Fanconi anemia complementation group A (FANCA) pVal6Asp, (rs1800282); and melanocortin 1 receptor (MC1R) pArg160Trp (rs1805008). The known functions of these genes cluster into two main functional families, 3 out of 8 are DNA repair genes associated with Fanconi anemia: FANCD2, SLX4, and FANCA [40‐45], while 3 out the 8 are HLA genes comprising the major histocompatibility complex (MHC) involving immunity. The remaining genes are associated with skeletal neogenesis [46‐48], protein synthesis homeostasis [49] and skin pigmentation [50]. All these germline variants were previously reported in the general population validating our findings, however their structure function-relationship in their individual encoded proteins awaits characterization. We report here that these gene variants occur in HNSCC patients and could be associated with late toxicity. Based on our study design, we found that FANCD2, DYNC2H1, PPP1R15A and HLA-A variants are associated with a significant increase in late toxicity, while SLX4, HLA-DQA2, HLA-DMB, HLA-DMA, FANCA, and MC1R variants are associated with fewer late RT toxicity events.

In addition to early versus late toxicity gene variant effects, we analyzed cumulative toxicity associations, Group C (Grade 2–4) and Group D (Grade 0–1) in the same cohort of patients (n = 37, Table 4).

Not surprisingly, the TSC2 splice site variant (rs1800720) detected in our previous study design displayed significant cumulative toxicity association. In addition, we detected a new variant GEN1, a Holliday junction 5′ flap endonuclease (GEN1) pLys839Glufs (rs149936944) that was associated with significant protection from RT therapy induced toxicity. We found that 9 patients carried this variant and that 2 of them were bi-allelic (Fig. 2). We confirmed this finding by analyzing the RNA-seq data. The reference GEN1 protein is 908 amino acids long. The variant we detected has a frame shift mutation introducing a stop codon at position 839 resulting in truncated messenger RNA (mRNA) missing 69 amino acids at the C-terminus of the GEN1 protein with potential for nonsense-mediated RNA decay. The presence of homozygosity (Fig. 2) of this variant implicates GEN1 redundancy, however, it is possible that this change leads to an alternate biological function. Indeed, GEN1 redundancy has been reported by Wang et.al [51]. They found that GEN1 and essential meiotic structure-specific endonuclease 1 (EME1) play redundant roles in meiotic recombination in a mouse model and that deletion of both genes confer synthetic lethality in mice [51]. In addition, a GEN1 knockout mouse is viable [52, 53].

We found that the nuclear receptor corepressor 2 (NCOR2) Gln510dup variant (rs35831183), showed a strong association with Group D (p = 0.008151).

We also examined clinical outcomes between the two sets of groups and the implication of certain variants. For this portion of the analysis only those patients with HNSCC treated with definitive or post-op radiation/chemoRT were included. A total of 34 patients were included with radiation treatment dates ranging from January 2009 to January 2021. There were 14 patients (41.2%) that were treated in the post-operative setting. Radiation dose was available for 31 (91.2%) of patients, ranging 60–70 Gy (Gy). 27 patients (79.4%) received concurrent systemic therapy, most frequently cisplatin (70.4%). Median age at treatment was 62.5 years (range 37–85, Table 1), with a median follow up of 42.5 months (range 7–205). Between Groups A vs. B and Groups C vs. D there were no significant differences in age, stage, site of primary, HPV status, use of surgery or concurrent chemotherapy. Patient characteristics can be found in Table 1. 11.7% of patients had Grade 3 early mucositis or dysphagia and 29.4% of patients had Grade 3 dysphagia; no Grade 4 or 5 AEs were recorded.

Patients in Group A had significantly better OS in comparison to Group B (p = 0.017, Fig. 3). There were also significant differences in PFS (p = 0.001), LFS (p = 0.041) and MFS (0.010). As noted previously, alterations in TSC2 were more prevalent in elevated late toxicity Group A, and patients with variants in TSC2 also had significantly improved LFS (p = 0.018, Table 3). FANCD2 was also more prevalent in Group A and associated with improved PFS (p = 0.039) and MFS (p = 0.018), as was PPP1R15A with PFS (p = 0.008) and LFS (0.028). There were five HLA-DMB/HLA-DMA variants only present in patients in Group B which were associated with a higher risk of LRR (p = 0.041). Within Groups C and D the major HLA-F N353L variant was significantly more prevalent in Group D and associated with more frequent distant metastases (p = 0.046).