Transitions in oral and gut microbiome of HPV+ oropharyngeal squamous cell carcinoma following definitive chemoradiotherapy (ROMA LA-OPSCC study)

Background Oral and gut microbiomes have emerged as potential biomarkers in cancer. We characterised the oral and gut microbiomes in a prospective observational cohort of HPV+ oropharyngeal squamous cell carcinoma (OPSCC) patients and evaluated the impact of chemoradiotherapy (CRT). Methods Saliva, oropharyngeal swabs over the tumour site and stool were collected at baseline and post-CRT. 16S RNA and shotgun metagenomic sequencing were used to generate taxonomic profiles, including relative abundance (RA), bacterial density, α-diversity and β-diversity. Results A total of 132 samples from 22 patients were analysed. Baseline saliva and swabs had similar taxonomic composition (R2 = 0.006; p = 0.827). Oropharyngeal swabs and stool taxonomic composition varied significantly by stage, with increased oral RA of Fusobacterium nucleatum observed in stage III disease (p < 0.05). CRT significantly reduced the species richness and increased the RA of gut-associated taxa in oropharyngeal swabs (p < 0.05), while it had no effect in stool samples. These findings remained significant when adjusted by stage, smoking status and antibiotic use. Conclusions Baseline oral and gut microbiomes differ by stage in this HPV+ cohort. CRT caused a shift towards a gut-like microbiome composition in oropharyngeal swabs. Stage-specific features and the transitions in oral microbiome might have prognostic and therapeutic implications.


BACKGROUND
The human microbiome has recently emerged as a promising biomarker in cancer. 1 The microbiome inhabiting the orogastrointestinal tract has been implicated in the carcinogenesis of many tumour types and in modulating responses to anti-cancer therapies, including immunotherapy, although the mechanisms are not yet well understood. [2][3][4] A few studies have shown differential oral microbial composition in the saliva of patients with oral cavity and oropharyngeal tumours when compared to healthy individuals, while specific commensals have been associated with lower risk of developing head and neck squamous cell carcinomas (HNSCC). [5][6][7] Oral microbiome composition seems to vary across different primary sites (e.g. oral cavity vs oropharynx) or according to stage, human papillomavirus (HPV) status and treatment received (e.g. surgery vs chemoradiotherapy (CRT)), suggesting a role as a tumour-specific biomarker in this disease, with potential impact on treatment efficacy and toxicity. [8][9][10] However, the evaluation of the oral microbiome in HNSCC has thus far been limited to retrospective and heterogeneous cohorts of patients, while the gut microbiome is yet to be investigated.
Among HNSCC, the incidence of oropharyngeal squamous cell carcinoma (OPSCC) has dramatically increased over the past decade, with HPV-related disease being most prevalent. 11,12 HPVpositive (HPV+) OPSCC are a biologically distinct disease with increased treatment responsiveness and survival when compared to HPV-negative tumours. 13 As such, multiple studies are evaluating de-escalation strategies in the locoregionally advanced (LA) setting to reduce treatment toxicity without compromising survival. 14 However, HPV+ tumours are heterogeneous and not all have a favourable prognosis. 15 Beyond clinical and pathological factors such as smoking history, tumour, node, metastasis (TNM) staging and HPV status, there remains an unmet need for new biomarkers that provide accurate risk stratification of this patient population. In this regard, HPV+ LA-OPSCC represents a unique setting to evaluate and compare both tumour-associated and gut microbiomes and their potential effect on treatment.
ROMA LA-OPSCC is the first study to prospectively characterise both oral and gut microbiomes and to evaluate the impact of definitive CRT on their composition in a homogeneous cohort of newly diagnosed HPV+ LA-OPSCC.

METHODS
Patient population and study design ROMA LA-OPSCC (NCT03759730) is a single-centre, non-interventional, investigator-initiated feasibility study designed to evaluate the oral and intestinal microbiome in a prospective cohort of patients with HPV+ LA-OPSCC treated with definitive CRT. Patients with previously untreated histologically proven OPSCC (tonsil, base of tongue, soft palate) candidates for definitive concurrent CRT with single-agent cisplatin (CDDP) as per standard of care were eligible. HPV status was determined by p16 immunohistochemical staining and classified as positive if nuclear and cytoplasmic staining in ≥70% tumour cells. In situ hybridisation to confirm the presence of high-risk HPV DNA was performed in equivocal cases. All patients were staged and treated according to eighth edition TNM staging criteria. Treatment and follow-up assessments were conducted according to institutional protocol (Supplement). Saliva, oropharyngeal swabs over the tumour site and stools samples were collected before treatment (up to 3 weeks prior to the start of radiotherapy) and at completion of CRT (up to 3 weeks following last day of radiotherapy; Supplementary Fig. 1). Patients were evaluable for analysis if samples were provided at least at one time point. The study was approved by the institutional research ethics board. All patients provided written, signed, informed consent to participate. Treatment and follow-up All patients received intensity-modulated radiotherapy to a gross tumour dose of 70 Gy in 35 fractions over 7 weeks (2 Gy/fraction). Concurrent CDDP (three-weekly at 100 mg/m 2 on RT days 1, 22 and 43 or weekly at 40 mg/m 2 for 7 weeks) was delivered according to institutional protocol. The choice of a three-weekly versus weekly schedule was based on patient's Eastern Cooperative Oncology Group Performance scale and comorbidities as assessed by medical oncologist. All patients had a prophylactic gastrostomy tube placed within 3-4 weeks from the start of radiation as per institutional standard practice. Follow-up after treatment completion was conducted according to institutional protocol. Local and regional recurrences were confirmed histologically, while distant metastases were diagnosed by unequivocal clinical/radiologic evidence +/− histologic confirmation. Clinical data were abstracted prospectively (M.O.) for all patients enrolled in the study.
Sample collection and microbiome analysis Saliva, oropharyngeal swab over the tumour site and stool samples were collected using the ZymoBIOMICS DNA/RNA Mini Prep TM kits (Zymo Research, Irvine, CA). Sampling and storage protocol are available as Supplementary Data (Laboratory Manual). Processing and analysis of the samples was conducted at the Centre for Genome Evolution and Function (CAGEF) of the University of Toronto. DNA was extracted using the ZymoBIOMICS DNA Micro Kit TM . 16S rRNA gene sequencing (Supplement) was performed on saliva (n = 46), oropharyngeal swabs (n = 46) and stool samples (n = 46). Briefly, the V4 hypervariable region of the 16S rRNA gene is amplified using an universal forward sequencing primer and a uniquely barcoded reverse sequencing primer to allow for multiplexing. 16 Amplicon sequencing was performed on an Illumina MiSeq platform (Illumina, CA, USA) with V2 chemistry as described in Schneeberger et al. 17 . Taxonomic profiling of 16S data sets was performed using the UNOISE pipeline. 18 Shotgun metagenomics sequencing was only performed on oropharyngeal swabs (n = 46) and stool samples (n = 46). Libraries were constructed using the Illumina Nextera Flex kits (Illumina, USA) using 150 ng DNA as input. A total of 1.94 Billion reads were generated on an Illumina NovaSeq 6000 platform (Illumina, USA) using a SP flow cell and reagents according to the manufacturer's protocol at the Princess Margaret Genomics Centre. A median of 2.3E + 07 [1.35E + 07-4.01E + 07] reads for stool samples and of 9.16E + 05 [2.17E + 05-3.09E + 07] reads for oropharyngeal samples were remaining after host read removal with Kneaddata v. 0.7.2 (https://bitbucket.org/biobakery/biobakery/wiki/kneaddata). Taxonomic profiles resulting from shotgun datasets were generated using Metaphlan2 with the Chocophlan database v. 293. 19,20 Alpha diversity and beta diversity were measured using the Phyloseq package (ref. 20 ; v. 3.9) and VEGAN v. 2.5.5 21 in R v.

22
Statistical analysis ROMA LA-OPSCC is a signal-finding study. Descriptive statistics were used to summarise clinical and microbiome characteristics. Mixed model regression was conducted to explore the potential demographic and clinical factors that are related to the microbial change during CRT. For microbiome analyses, summary statistics were described including within-patient community composition (taxonomic relative abundance), alpha diversity (compositional diversity within-sample) using Shannon index (SDI; a composite metric of both richness and evenness) and Berger-Parker index (BP; an indicator of dominance in the community), as well as betadiversity (inter-sample similarity) of baseline and end of treatment samples. Alpha diversity measures were compared between groups using Mann-Whitney (MW) tests. LEFSE was used to measure the differences in relative abundances (non-parametric Kruskal-Wallis tests) and the effect size (linear discriminant analysis) between groups. Beta diversity was measured using the Bray-Curtis dissimilarity index and group comparisons were conducted using permutational multivariate analysis of variance (PERMANOVA). Interaction between treatment effect (changes in composition pre-and post-CRT) and other variables including use of antibiotics, G-tube dependency, grade of mucositis, smoking status, tumour location, stage and T staging were measured using PERMANOVA. Assuming a significance level for alpha of 0.01 to adjust for multiple comparisons of key taxa, alpha diversity, and beta diversity, our study with 22 patients' microbiome samples had at least 85% power to identify significant differences between pre-and post-CRT, given an effect size of 0.7 standard deviation (SD) of the paired mean difference. The power analysis is based on two-sided paired t tests. Description of baseline oral and stool microbiome in HPV+ LA-OPSCC A total of 132 samples collected from the 22 evaluable patients (100% compliance in sample acquisition) were analysed. Taxonomic composition of oropharyngeal swabs and saliva samples by 16S rRNA gene sequencing were similar (R 2 = 0.06; p = 0.827; Supplementary Fig. 3), thus shotgun metagenomic sequencing was only conducted in oropharyngeal swabs and stool. All subsequent results are based on shotgun metagenomic sequencing analyses. Taxonomic composition differed by sampling site (oropharyngeal swabs vs stool samples: R 2 = 0.276; p = 0.001; Fig. 1a, b). Oral communities comprised mostly oropharyngeal anaerobes and facultative anaerobes, including Prevotella, Veillonella, Streptococcus and Actinomyces species while stool communities were composed mainly of obligate anaerobic Bacteroides species. The number of species was higher in the stool vs oral communities (p < 0.0001) but they had overall similar diversity (SDI mean = 3.3 for stool and 3.12 for oropharyngeal samples; BP mean = 0.19 for stool and 0.2 for oropharyngeal samples; Fig. 1c). Four patients (R05, R17, R23 and R26) had a high proportion (>10% of the community) of Bacteroides species in their oropharyngeal swabs more typical of the lower intestinal tract.

Clinical characteristics and outcome
Differential baseline oral and stool microbiome composition by stage Taxonomic composition of oropharyngeal swabs significantly differed across stage III vs stage I-II patients (p < 0.05): four genera were enriched in patients with stage III, including Fusobacterium (Fusobacterium nucleatum), Gemella (Gemella morbillorum and Gemella haemolysans), Leptotrichia (Leptotrichia hofstadii) and Selenomonas (Selenomonas sputigena and Selenomonas infelix) (Fig. 2a). Taxonomic composition of stool samples also differed in stage III vs stage I-II disease, with significant enrichment of two phyla, Actinobacteria and Proteobacteria, and 18 species (p < 0.05; Fig. 2b).
In the univariate analysis, no effect on baseline oropharyngeal swab microbiome composition was seen by smoking or primary tumour location, although a trend was observed by T staging (p = 0.06; Supplementary Table 2). Impact of CRT on the oral and stool microbiome Oral microbiome. We compared the composition of oropharyngeal swabs pre-and post-CRT (Fig. 3). We observed the formation of three distinct clusters based on collection time point baseline vs post-CRT (Fig. 3a): cluster 1 (15 baseline vs 1 post-CRT samples) was characterised by high relative abundance of species from the Veillonella, Prevotella and Streptococcus genera; cluster 2 (4 baseline vs 11 post-CRT samples) was characterised by high abundances of Streptococcus species, Prevotella melaninogenica, Neisseria flavescens and Rothia mucilaginosa, among others; and cluster 3 (3 baseline vs 10 post-CRT samples) was characterised by high abundances of species from the Bacteroides, Faecalibacterium, Prevotella (Prevotella copri), Collinsella, Alistipes and Parabacteroides genera. Overall, the number of species was significantly reduced in post-CRT oropharyngeal swabs (MW; p = 0.006). Alpha diversity did not change post-CRT (SDI mean = 3.12 at baseline and 3.09 at the end of treatment; BP mean = 0.2 at baseline and 0.2 at the end of treatment; MW; p SDI = 0.716; p BP = 0.944) nor did bacterial density (8.8E + 09 16S copies/ml at baseline and 2.6E + 09 16S copies/ml at the end of treatment; p = 0.15) ( Supplementary Fig. 4). Intra-patient changes in community composition post-CRT are summarised in Fig. 3b. Most patients (65%) were classified in cluster 1 at baseline while only 1 patient (4.5%) was classified in cluster 1 post-CRT. Out of the 15 patients in cluster 1 at baseline, 8 transitioned to cluster 2 and 6 to cluster 3 post-CRT. None of the patients who grouped in cluster 2 and 3 at baseline shifted to cluster 1 after CRT and remained within the cluster 2 or 3.
Clinical characteristics associated with cluster subgroups are summarised in Table 1. Out of the 15 patients grouping in cluster 1 at baseline, the majority were former/non-smokers (93%), with tonsillar primary (67%) and stage I-II disease (67%). Baseline oral composition from 3 out of the 4 current smokers of the cohort belonged to cluster 2 or 3. No clear pattern was seen between cluster transitions post-CRT by TNM, stage, smoking status, use of antibiotics, grade of mucositis or gastrostomy tube dependency post-CRT. One of the two patients (patient R05) with cluster 3-type oral microbial composition at baseline and post-CRT experienced biopsy-proven locoregionally and distant recurrence.
Overall, the compositional changes between baseline and post-CRT consisted of a shift towards gastrointestinal tract-like communities (Fig. 4). Oropharyngeal swabs post-CRT clustered closer to stool samples, with significant changes in taxa composition when compared to baseline (R 2 = 0.1; p = 0.001). The dissimilarity observed between oropharyngeal swabs and stool samples was reduced in post-CRT samples (PERMANOVA; R 2 = 0.115; p = 0.001) compared to baseline samples (PERMA-NOVA; R 2 = 0.203; p = 0.001). Functional analyses associated with these taxonomic findings were attempted but could not be performed due to insufficient sequencing depth ( Supplementary  Fig. 5).  Table 3 and Supplementary Fig. 7).
The impact of CRT (defined as collection time point: baseline and post-CRT) on the oral microbiome remained significant when adjusting by potential confounding factors, including smoking status, TNM, stage subgroups, maximum grade of mucositis, gastrostomy tube dependency 3-4 weeks post-CRT and use of antibiotics (Supplementary Table 4).

DISCUSSION
ROMA LA-OPSCC is the first study to prospectively characterise both oral and gut microbiomes in HPV+ OPSCC patients treated with definitive CRT. We found that both oral and stool community composition differed by disease stage at baseline and that the oral but not stool microbiome composition changed after CRT. The shift in oropharyngeal taxonomic composition after treatment was largely driven by an increase in the relative abundance of gutassociated obligate anaerobes. The results of this study provide a step forward in the understanding of both microbiomes in this disease and may be used as a benchmark as new treatments are being investigated in this patient population.
The composition of the oral microbiome in our cohort was comparable to that of other retrospective cohorts involving patients with oral cavity and oropharyngeal tumours. 8,23 Guerrero-Preston et al. reported differential taxonomic composition in HPV+ OPSCC when compared to HPV-negative OPSCC and oral cavity cancer, with higher prevalence of Veillonella, Prevotella, Streptococcus and Gemella genera. 8,9 In our cohort involving HPV+ disease exclusively, we did observe a similar taxonomic composition at the genus level. Our shotgun metagenomics analysis revealed differential oral microbial composition across stages, and patients with stage III had significantly higher relative abundance of F. nucleatum species. F. nucleatum had been previously described in heterogeneous cohorts involving HNSCC patients treated with surgery and/or radiation, but it has also been recently associated with advanced disease, chemotherapy resistance and adverse prognosis in other tumour   types, such as oesophageal carcinoma. 9,24-26 In our cohort, the stage differences in oral composition seemed to be associated with larger primary tumours. Patients with stage III HPV+ OPSCC are known to be at higher risk of recurrence despite definitive concurrent chemoradiation and new treatment intensification approaches including immunotherapy that are being explored in this setting (e.g. NCT03040999). 27 It remains to be tested whether these findings have prognostic implications and therefore could be used for risk stratification in this patient population. Interestingly, while smoking history seems to have a role as prognostic biomarker for HPV+ disease and has also been highlighted to correlate with oral dysbiosis, we did not observe differences in oral microbiome composition according to smoking status in our overall cohort or by stage. [28][29][30] We evaluated the changes post-CRT on both oral and stool microbiomes. Two studies involving patients with HNSCC and  III  I/II  III  I/II  III  I/II  III  I/II  III  I/II  III  I/II  III  I/II  III   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I/II   III   I Baseline  End of treatment   R10  R15  R24  R04  R01  R06  R11  R19  R25  R20  R20  R18  R12  R09  R07  R26  R07  R04  R16  R25  R21  R12  R13  R11  R16  R10  R22  R13  R19  R06  R23  R09  R22  R26  R05  R05  R18  R17  R01  R17  R15  R24  R21  R23   CST3 CST2 nasopharyngeal carcinoma have described changes in the oral microbiota following radiation and an increase in opportunistic pathogens. 31,32 Our analysis revealed a significant and consistent impact of CRT in the overall oral communities among the cohort, with increases in the prevalence and relative abundance of obligate anaerobes (e.g. Bacteroides species). The cause of these shifts is unclear but may be due to treatment-induced tissue necrosis or other changes in the tumour-adjacent mucosa, direct effects of CRT on the microbes themselves or treatmentassociated immune or metabolic changes in the local tissues affecting microbial ecology. The potential biological and/or clinical impact of baseline and post-treatment composition or shifts after CRT remains unknown and long-term follow-up is required. Of note, one of the two patients harbouring a "gut-like" oropharyngeal taxa both at baseline and post-CRT experienced disease recurrence about a year after treatment completion. We did not observe any significant shift in the gut microbiome composition after CRT in our cohort. While radiation is a local therapy and thus it is not expected to specifically alter the gut microbiome, cytotoxic chemotherapeutic agents including cisplatin are known to induce damage of the intestinal mucosa and disrupt the microbiome, leading to increased risk of infections. 33 The heterogeneity of gut microbiome composition at baseline and the limited number of patients might have limited the detection of differences due to CRT or antibiotic use. Although there were intra-patient changes in gut microbiome composition in our study, these changes were patient specific and no common pattern was observed in the overall cohort.
There were no differences in the overall taxa composition between saliva and oropharyngeal swabs taken from the tumour site. This is particularly relevant as this patient population is characterised by radiation-induced xerostomia, 34 and thus the swab could substitute the collection of saliva, the collection of which can be a challenge following completion of CRT in this patient population. Zhang et al. reported differential taxa composition between saliva and tumour tissue from patients with oral cavity tumours, with significantly higher levels of F. nucleatum and Acinetobacter found in the tumour. 25 Whether microbiome data that are obtained from oropharyngeal swabs differ from those from tumour tissue was not assessable in our study.
The limitations of our study include: inability to account for all patient factors that may influence oral and stool microbial community composition, such as dietary habits and dental hygiene; short median follow-up for HPV+ OPSCC disease limiting the evaluation of the prognostic impact of microbiome signatures; lack of further sampling beyond 4 weeks from CRT limiting the evaluation of long-term oral and stool microbiome alterations post-CRT 31,35 ; small number of patients involved, which prevents statistical power for specific subgroup analysis. We used both 16S rRNA and shotgun sequencing techniques for two reasons. We first wanted to assess the level of agreement between different samples types retrieved in the same body compartment (saliva vs oropharyngeal swabs). For this exploratory analysis, 16S sequencing is sufficiently sensitive to compare the overall composition between sample types with a relatively low cost. Based on the high agreement between both sample types, we then selected oropharyngeal swabs for shotgun sequencing, as it has higher taxonomic resolution to observe CRTmediated changes at the species level. This combination of approaches allowed us to gain the greatest amount of highresolution microbiome compositional data at the lowest cost. A 'shallow' shotgun approach was used to characterise the taxonomic composition in the different sample types, which was sufficient for the detection of species above relative abundance of 0.05% but it did not allow us to conduct functional analyses. 36 This pilot study shows that prospective characterisation of both oral and stool microbiome is feasible in this patient population, with 100% compliance in sample acquisition and analysis. The stagespecific microbial features in the oral and gut communities from this cohort are hypothesis-generating and should be further investigated to evaluate their use as a biomarker for risk stratification in patients with HPV+ OPSCC. Additional correlation with HPV-related factors such as serotype or viral load in saliva and comparison with a matched-HPV-negative cohort are to be explored. These findings might serve as a 'control' for the microbiome landscape as therapeutic interventions such as immunotherapy are being incorporated into the treatment of these patient populations. Indeed, prospective evaluation of oral and intestinal microbiome is currently ongoing in the setting of an international prospective chemo-sparing approach evaluating definitive chemoradiation vs immunoradiotherapy in HPV+ intermediate-risk OPSCC (NCT03410615). The transitions observed in the composition of the oral but not gut microbiome following treatment might not only have prognostic value but also therapeutic implications to explore gut microbiome modulation strategies in this setting. In this regard, we are currently evaluating the feasibility of gut microbiome intervention in the context of CRT in patients with LA-OPSCC using an oral consortium of taxa associated with immune checkpoint inhibitor-responsiveness (NCT03838601).