Droplet digital PCR quantification suggests that higher viral load correlates with improved survival in HPV-positive oropharyngeal tumours

suggests that higher correlates with improved survival

The incidence of oropharyngeal cancer (OPC) has increased dramatically over the last two decades 53 including in the United Kingdom (1,2). A component of OPC is associated with HPV although the 54 amount varies. For example, a recent global analysis indicated <10% of OPC cases in Brazil were 55 positive for HPV compared to ~50% in the UK (3). While the extent of HPV driven OPC varies, data 56 converge on the fact that HPV positive (versus negative) status is independently associated with 57 better clinical outcomes (4,5). This has led to a recent change in tumour classification which 58 incorporates HPV status and also trials to determine the efficacy of de-escalated therapy in HPV 59 positive OPC patients (6,7). 60 61 Current approaches for determining HPV status of OPC are largely based on qualitative tests and 62 include immunohistochemistry for p16INK4a or HPV PCR for DNA or mRNA (8). Unfortunately, some 63 HPV-positive patients, so determined by these methods, still have very poor outcomes (9). Further, 64 these approaches do not quantify levels of infection i.e. viral load (VL). Given the increased incidence 65 of OPC, it is important to refine tools for improved risk stratification; one such candidate is the 66 measurement of viral load. 67 68 Current evidence indicates that VL in HPV-positive head and neck cancers varies widely within and 69 between anatomical sites (10)(11)(12)(13)(14). In addition, investigations into the physical status of HPV in OPC 70 indicate a landscape of integrated and episomal forms within a single lesion (15,16 reaction as a copy number reference, i.e. each ddPCR was a duplex reaction. All reaction runs 126 contained negative control wells in triplicate. In-reaction digestion of the DNA with restriction 127 enzymes was performed to enhance the partitioning of DNA into droplets. We confirmed that the 128 restriction enzymes selected for this (EcoRI and HindIII), would not cut within any of the viral or 129 control target sequences. Primer and probe concentrations were optimised by titration. Reaction 130 mixes were set up using ddPCR Supermix for Probes without dUTP (Bio-Rad), 0.7 µl of the RPP30 131 endogenous control assay, HPV16 L1 or E6 specific primers and probes at 300 nM and 200 nM (final 132 concentration) respectively, 10-100 ng of template DNA and 1 µl of restriction digest mix (consisting 133 of 4 U of both EcoRI and HindIII in 1x NEB Cutsmart buffer (NEB, UK)). Reactions were mixed with 134 Droplet Generation Oil on DG8 cartridges in the QX200 droplet generator (Bio-Rad) to generate 135 droplets. Thermal cycling conditions were: 95°C for 10 minutes followed by 40 x 94°C for 30 s and 136 60°C for 1 minute prior to final extension at 98°C for 10 minutes. Post amplification, droplets were 137 analysed on a QX200 Droplet Reader (Bio-Rad) and output data files were analysed using QuantaSoft 138 analysis software v1.7.4 (Bio-Rad). 139 140

Definition of low and high VL 141
The individual viral loads were ranked from smallest to largest and separated using tertiles. A priori, 142 the analysis planned to compare viral load in tertiles (low, medium and high) but low numbers of 143 deaths in the medium and high viral load groups meant that analysis was performed for low VL(viral 144 load in the lowest third) vs a combined medium/high VL category (viral load in the upper two-thirds); 145 this was performed for both L1 and E6. The VL threshold(s) for "medium/high" E6 VL and L1 VL were 146 >8.68 and >5.57 viral gene copies per cell respectively. Samples with VL lower than this were classed 147 as having a low VL. 148 149

E6 and L1 VL and clinical outcome 150
Analysis was performed on the cases which had ddPCR results (n=93 for L1 and n = 82 for E6). 151 Kaplan Meier plots were constructed for overall and progression free survival, stratified according to 152 the OPC cases having a low or medium/high VL. In addition, hazard to death and hazard to death or 153 recurrence were assessed and related to low or medium/high VL status using Cox's regression with 154 Firth's penalised likelihood given the small denominators. Follow up data were censored as of 155

Ratios of E6 and L1 VL and clinical outcome 160
The association(s) between E6/L1 ratio and the demographic and clinical variables were assessed 161 with significance determined using the Fishers exact test. Further, we modelled the association 162 between the distance of the E6/L1 ratio to 1 and the variables using a regression model (to avoid 163 imposing an "arbitrary" cut off). As the effect may have been different for E6/L1 >1 vs <1, tests for 164 interaction were performed to determine whether they could be included with the same model. All 165 statistical analyses were performed in R version 3.6.1. 166

Characteristics of cohort assessed for VL 170
Demographic and clinical variables of the cohort are presented in Table 1 in addition to VL status 171 separated as "low", "medium" or "high". The cohort contained 75 males and 18 females with an 172 average age of 57 (interquartile range of 52-66). All were squamous cell carcinomas and most cases 173 were from more deprived areas; 50/93 patients were SIMD 1 and 2. In relation to smoking and 174 alcohol, 53 had "ever" smoked and 19 were heavy drinkers. A total of 84 cases were TMN 3 or above 175 and most, (56/93), received chemo-radiotherapy as treatment.  Medium/high viral load was weakly associated with longer overall survival and progression free 211 survival although the relationship was not as strong as that observed for L1 (Figure 2). In the 212 univariate analyses, medium/high E6 VL was associated with a slightly higher overall survival 213 although this was not significant; HR 0.76 (95%CI 0.21-2.68) p=0.67 (Table 3, Figure 2). High alcohol 214 intake was associated with worse overall survival; HR 4.31 (95%CI 1.18-15.67)  We have demonstrated that ddPCR is an accurate and rapid method for determining HPV VL in OPC 228 patients, consistent with previous studies in smaller cohorts (28,29). The VL detected with both E6 229 and L1 genes displayed a wide range but was much more restricted than the 10 3 to 10 7 range 230 reported for 48 OPC patients (12) or 1->900 copies in 45 clinical samples, (29), which both detected 231 L1 gene copy numbers. 232

233
The ddPCR analysis indicated that a higher L1 VL was associated with better clinical outcomes. One 234 explanation is that cancers with a higher VL are virus "driven" -whereas those with lower VLs may 235 represent cancers where other drivers are responsible for the cancer due to impairment of viral 236 function through integration and/or epigenetic mechanisms. Notably, we detected very low levels 237 indeed of L1 with some as low as 0.001 copies per cell. Although such samples tested positive for 238 HPV16 in the original HPV PCR/genotyping test, we accept that the presence of the virus might not 239 drive tumorigenesis in these cases and that some tumours in the "low VL" group could be treated as 240 functionally HPV-negative. We did not test cancers for p16INK4a or E6 RT-PCR or perform in-situ 241 hybridisation for E6/E7 sequences, which can indicate transcriptional activity of virus. This would be 242 of interest for future work and could better clarify the relevance and activity of the virus of the "low 243 VL" group (30). 244 245 It is known that the mutational burden of HPV-positive OPC is lower than HPV negative cancer; 246 making it amenable to non-surgical treatment options (7). This may be because the majority of OPCs 247 retain viral episomes and there is a lack of insertional mutagenesis of cellular genes due to 248 integration (31). Therefore, the association of a higher VL with better outcome could be that 249 multiple viral episomes might allow full virus gene expression, particularly of the highly 250 immunogenic L1 protein. Theoretically, this would allow for greater antigen presentation and 251 immune checks, particularly in the tonsils, which are lymphoid tissue. 252 253 HPV genome status in OPC tumours can be episomal or integrated or exist as virus-human episomes 254 or integrants (32). While the relationship between integration status and clinical outcome is not fully 255 understood, only low copy numbers have been detected in cases with integrated HPV genomes (32). 256 Our samples had a mean viral load of 30.9 (L1) or 37.9 (E6), and VL with L1 detection was similar to 257 VL with E6 detection, suggesting the majority of samples had mostly episomal genomes, but this 258 requires confirmation. A surprising finding was that in the majority of samples, the ratio of E6 to L1 259 was not equal to one and some had greater numbers of E6 vs L1 copies; this may be explained by 260 amplification of E6/E7 genes either in episomal or integrated viral genomes due to recombination 261 events. Conversely, the samples which had more L1 than E6 copies may reflect the presence of full-262 length genomes alongside partially deleted viral genomes missing the E6 gene. We could not 263 demonstrate a relationship between L1:E6 ratios and clinical/demographic variables or outcomes, 264 but this may be an artefact of the small number of cases. Viral activity in OPCs, including level of HPV 265 gene expression is likely to provide further insight on clinical outcomes. While sequencing is 266 required to address this question, our data provide further proof of the instability of the HPV 267 genome in OPC. 268

269
In the adjusted analysis, VL was not independently associated with improved outcomes. This is 270 consistent with the fact that OPC is influenced by various behavioural, demographic and clinical 271 factors and their complex interplay. Consequently, L1 VL may be a proxy of one or a combination of 272 these but nevertheless represents a tool which can be applied objectively to ascertain risk-groups 273 within the HPV positive category. It is also feasible that VL may be used to indicate/inform treatment 274 We are grateful to staff at the Scottish HPV Reference laboratory for support with sample collation. 289 Conflicts: K Cuschieri (non-personal) K Cuschieri's institution has received research funding or gratis 290 consumables to support research from the following commercial entities in the last 3 years: 291 Cepheid, Genomica, LifeRiver, Euroimmun, GeneFirst, SelfScreen, Qiagen, Abbott, Hiantis and 292 Hologic. No other authors report a conflict of interest in relation to this work. 293