The gut microbiota may drive the radiosensitising effect of a high fibre diet

Background Patients with pelvic tumours, including bladder, cervix and rectal cancers, receiving radiotherapy are often given additional radiosensitising chemotherapy to improve cure rates, at the expense of increased toxicity. With an ageing population, new approaches to radiosensitisation are urgently required. Inhibition of Histone deacetylase (HDAC) is a promising mechanism of radiosensitisation. Tumour cells may accumulate butyrate which is produced by the gut microbiota via fibre fermentation to sufficient levels to cause HDAC inhibition, due to the Warburg effect. We hypothesised that mice fed a high fibre diet would have improved tumour control following ionising radiation, compared to mice fed a low fibre diet, and that the effects of the diet would be mediated through the gut microbiota. Results The faecal (n=59) and caecal (n=59) microbiomes from CD1 nude mice, implanted with RT112 human bladder cancer cell line flank xenografts at the same time as starting low and high fibre (soluble, insoluble, mixed) diets, were similar and they shared the three major taxa (> 80% abundance): Bacteroidales, Clostridales and Verrucomicrobiales. Significantly higher abundance of Bacteroides acidifaciens (p<0.01) was seen in the gut microbiome of the soluble high fibre (HF) group after 2 weeks of diet. Principal coordinate analysis showed a notable cluster effect within groups, indicating that the diets indeed modified the gut microbiome. Survival analysis by log-rank test showed soluble HF conferred survival benefits, with delayed tumour growth after irradiation (n=32, p=0.005). Comparison of the gut microbiomes in the soluble HF group between responders (n=4) and non-responders (n=4) to radiation revealed significantly higher abundance of B. acidifaciens in responding mice (p<0.05). Predictive metagenomic profiling showed the gut microbiome in responders was enriched for carbohydrate metabolism. To investigate the correlation between specific bacterial taxa and tumour response to radiation, all mice fed with different diets (n=32) were pooled together, and univariate linear regression revealed a statistically significant positive correlation between the survival time of mice and abundance of B. acidifaciens (R2=0.5278, P<0.001). Conclusions High fibre diets sensitised RT112 xenografts to radiation by modifying the gut microbiome and this phenotype was positively correlated with B. acidifaciens abundance. These findings might be exploitable for improving radiotherapy response in human patients.

increasing the response of tumour cells to radiation (radiosensitisation). Use of HDAC 83 inhibitors can result in acetylation of histone and non-histone proteins and can inhibit the 84 DNA damage response [11]. We previously found that in mice, the pan-HDAC inhibitor, 85 panobinostat, in combination with ionising radiation (IR) delayed the growth of RT112 86 bladder cancer xenografts but did not worsen local acute (3.75 days) and late (12 weeks) 87 toxicity in the surrounding intestinal tract [12]. The sparing of normal tissue in the gut relative 88 to the tumour is commensurate with the normal gut colonocytes using butyrate as their 89 primary energy source, so butyrate cellular concentrations do not accumulate to levels which 90 can lead to HDAC inhibition. 91 92 As a proof of principle of efficacy of this approach, we tested the hypothesis that mice fed a 93 high fibre diet would have improved tumour control following ionising radiation, compared 94 to mice fed a low fibre diet, and that the effects of the diet would be mediated through the 95 gut microbiota. The aims of this study were: (a) to evaluate the tumour growth in mice treated 96 with high or low fibre diets; (b) to examine the impact of the diet on the microbiome before 97 and after irradiation, and (c) to correlate diet-induced microbiome changes with tumour 98 growth and response to radiation treatment. 99 100

101
The environmental microbiome had minimal impact on gut microbiome analysis and 102 dietary fibre content did not affect tumour growth to 50 mm 3 103 Female CD1 nude mice were injected subcutaneously with RT112 bladder carcinoma cells, 104 and at the same time, they started to receive a modified diet, namely, one of the following: 105 low dietary fibre (0.2% cellulose), low fibre with butyrate in drinking water, high soluble fibre 106 (10% inulin), high insoluble fibre (10% cellulose) and high mixed fibre (5% cellulose, 5% inulin) 107 ( Figure 1A and Table S1). By quantification of bacterial loads using PCR and gel electrophoresis, 108 compared to known E. coli colony forming unit (CFU) numbers, our mouse samples contained 109 more than 10 4 bacterial CFUs which appeared to override contaminating species in the 110 sample microbial communities ( Figure 1B). The PBS negative control was processed 111 identically to the luminal content and tissue samples from the start of the DNA extraction. 112 The nucleic acid amount detected in the PBS negative controls was extremely low, compared 113 to that in the gut microbiota ( Figure 1C). Furthermore, the community microbiome in this 114 negative control was very different from the gut microbiome of the mice ( Figure 1D). 115 Therefore, the environmental microbiome had minimal impact on the analysis of the gut 116 microbiomes of interest in this study. 117 118 By investigating samples collected when the tumours reached 50 mm 3 and 350 mm 3 , the 119 faecal (hereinafter called gut microbiome) and caecal microbiome were found to have similar 120 bacterial components ( Figure S1). The mice were culled when their tumours reached 500 mm 3 . 121 Butyrate levels in the faeces, validated by high-performance liquid chromatography (HPLC), 122 were generally higher in the low fibre with butyrate and high soluble fibre groups (p=NS; 123 Figure 1E). The mean time for tumours to reach 50 mm 3 was 12.8  1.4 days ( Figure 1F). 124 lower Shannon's index (p<0.001 for Kruskal-Wallis test) ( Figure 3B). In terms of beta diversity, 149 principal coordinate analysis using Bray-Curtis dissimilarity showed a notable clustering effect 150 among different groups, which indicates that samples within groups were more similar to 151 each other than to those from the other groups ( Figure 3C). An increased 152 Firmicutes/Bacteroides ratio is considered to represent radiation-induced gut dysbiosis [13, 153 14]. Interestingly, only the low fibre group treated with radiation had a trend towards an 154 increased Firmicutes/Bacteroides ratio compared to non-irradiated control, but this 155 phenomenon was not seen in the high fibre diet groups, implying that dietary fibre may be 156 protective and therefore reduce gut dysbiosis caused by radiation treatment ( Figure S2A). 157 The diets continued to affect the bacterial composition of the gut microbiome as mice got 158 older, up to the time the tumours reached 350 mm 3 . The taxonomic cladogram of LEfSe 159 (Linear discriminant analysis Effect size) of the gut microbiome showed that the high soluble 160 fibre diet increased S24-7 ( Figure S2B). Studying the highest abundance bacterial taxa found 161 in the first cohort in this cohort, the relative abundance of B. acidifaciens, an acetate-162 producing bacteria [15,16], was distributed evenly in different diet groups, except in the 163 soluble HF group with treated with radiation (p=0.200 for Kruskal-Wallis test) ( Figure 3D). For 164 the highest abundance bacterial taxa in the second cohort, the relative abundance of 165 Bacteroidales S24-7 was significantly higher in the mixed HF and soluble HF groups compared 166 to LF and insoluble HF (p=0.001 for Kruskal-Wallis test) ( Figure 3E). 167 168 Soluble high fibre causes increased growth delay in irradiated bladder cancer cell xenografts 169 and responses are influenced by the gut microbiota composition. 170 To investigate the effect of different diets on the tumour response in mice irradiated when 171 the tumour had grown to 50 mm 3 , tumour growth was monitored to 350 mm 3 . Slopes of the 172 tumour growth curves were obtained using linear regression to indicate the tumour 173 progression rates. The high soluble fibre diet group had the lowest growth rate. The slopes 174 were 4.4 ± 1.3 for LF, 16.1 ± 1.7 for mixed HF, 28.7 ± 1.3 for insoluble HF and 0.4 ± 1.5 for 175 soluble HF ( Figure 4A and Figure S3 for individual irradiated mouse tumour growth curves). 176 Kaplan Meier survival curves for time to treble tumour volume showed that the soluble HF 177 group had the longest median survival time (7.5 days for LF, 7 days for mixed HF, 10 days for 178 insoluble HF, 11.5 days for soluble HF; p=0.005, log-rank test) ( Figure 4B). 179 180 Among the eight mice fed the soluble high fibre diet, four mice were classified as responders 181 as they had shallower slopes to the tumour growth curves; their slopes were 7.3 ± 1.2, -0.9 ± 182 0.6, -4.8 ± 0.5, -5.2 ± 0.9. The other four mice were classified as non-responders with steeper 183 slopes to the tumour growth curves, namely 34.6 ± 3.0, 31.1 ± 2.6, 23.3 ± 1.0, 33.8 ± 2.9 using 184 linear regression ( Figure 4C). In terms of alpha diversity, no significant difference in Shannon's 185 index was shown between responders and non-responders ( Figure 4D). In terms of beta 186 diversity, principal coordinate analysis of Bray-Curtis dissimilarity showed the gut microbiome 187 of responders and non-responders were more similar within groups than between groups 188 ( Figure 4E). 189

Differences in composition of the gut microbiome between responders and non-responders 191
Linear discriminant analysis showed that the mice responding to the soluble high fibre diet 192 for a slower tumour growth rate were enriched with Bacteroidaceae, Flavobacterium,193 Flavobacteriales,Lactococcus,Streptococcus,Streptococcaceae,Allobaculum,194 Erysipelotrichales. The non-responding tumour-bearing mice were enriched with 195 Bifidobacterium, Bidifobacteriaceae,Bifidobacteriales,Parabacteroides,196 Porphyromonadaceae, Lactobacillus, Lactobacillaceae and Lactobacillales ( Figure 5A). In 197 terms of effect size, B. acidifaciens and Bacteroidaceae had the largest enrichment in 198 responders, and Parabacteroides and Porphyromonadaceae had the largest enrichment in 199 non-responders ( Figure 5B). To further explore these findings, the discrete false-discovery 200 rates within all taxonomic levels were calculated ( Figure 5C As B. acidifaciens was the 'top hit' for responders in the soluble HF group, we explored how 210 specific bacterial taxa affected mouse survival time. The correlation between B. acidifaciens 211 abundance and time to culling was investigated across the diet groups. Some mice in the non-212 IR cohort lived as long as those in the IR cohort, ie. >40 days, which may be a reflection of 6 213 Gy being a relatively low dose of radiation ( Figure S4A). In the IR cohort, the time of culling 214 positively correlated with B. acidifaciens abundance (R 2 =0.528, p<0.001). However, a similar 215 correlation was not seen in the non-IR cohort (R 2 =0.085, p=0.357). Using the time for tumours 216 to treble in volume as the outcome measure, mice with high B. acidifaciens abundance had a 217 significantly prolonged median survival time (Log-rank test: p<0.001) ( Figure 6A). A similar 218 finding was seen in the IR cohort (p=0.003), but not in the non-IR cohort (p=0.236). To further 219 identify specific unfavourable bacterial taxa which might offset the effect of radiation, the 220 Parabacteroides genus, was selected, as one of the top two 'hits' for non-responders in the 221 soluble HF group. This was investigated for the association between its abundance and time 222 to culling ( Figure S4B). In the IR cohort, the time to culling negatively correlated with the 223 abundance of Parabacteroides genus ( Figure S4B; R 2 =0.164, p=0.022). However, a similar 224 correlation was not seen in the non-IR cohort (R 2 =0.084, p=0.360). Using the time for tumours 225 to treble in volume as the outcome measure, mice in the low Parabacteroides genus 226 abundance group had no significant difference in median survival time compared to the high 227 abundance group (Log-rank test: p=0.374) ( Figure 6B). B. acidifaciens (p=0.200 for  Wallis test) and Parabacteroides genus (p=0.005 for Kruskal-Wallis test) abundance was 229 evenly distributed among all cages, which suggested that the existence of this taxa in the gut 230 microbiome was not a cage-specific effect ( Figure S5 and Table S2). cancer xenografts [18]. In our study, no significant difference for the time for tumours to 261 reach 50 mm 3 was seen in mice from different groups. This demonstrated that diet and the 262 diet-modified microbiome had no effect on tumour growth, up to that point. 263

264
The tumours were irradiated when they reached a volume of 50 mm 3 and monitored until 265 they reached 350 mm 3 . As the mice aged during the time required for tumour growth, all mice 266 were enriched for the S24-7 family. This indicated that their gut microbiomes had altered over 267 the tumour growth period. Although the gut microbiomes became more heterogeneous, a 268 notable cluster effect still existed in samples within groups. More recently it has emerged that one of the major mechanisms by which dietary fibre intake 287 protects against cancer is the modulation of the gut microbiota [23], via short chain fatty acid 288 production and immunomodulation. Dysbiosis has been linked to colorectal cancer, and a 289 recent study proposed stage-distinct alterations in the tumourigenesis of colorectal cancer 290 [24]; Atopobium parvulum and Actinomyces odontolyticus were significantly enriched in 291 multiple polypoid adenomas and intramucosal carcinomas, and Fusobacterium nucleatum 292 increased continuously from the intramucosal carcinoma to more advanced stages. 293 Helicobacter pylori is the most famous and early-defined link between a bacterium and cancer 294 [25]. Nowadays, H. pylori eradication has become one of the most promising treatments in 295 the prevention of gastric cancer [26]. 296   297 In this study, in CD1 nude mice, short-term high fibre intake alone and its modifying influence 298 on the gut microbiota had no effect on tumour progression rate (to 50 mm 3 ). All of these recent studies proposed that a modified gut microbiota can augment the efficacy 307 of anti-tumoural treatment. However, most of them are limited to chemotherapy and 308 immunotherapy [30]. To date, this is the first study to suggest that a high soluble fibre diet, 309 with the related modified gut microbiome, can act as a radiosensitiser. Innate immunity plays 310 a role in shaping the gut microbiota and adaptive immunity has a strong impact on tumour 311 progression or treatment response. Our studies were conducted in immunodeficient CD-1 312 nude mice, which implies that the radiosensitising effect of dietary fibre here was 313 independent of immunomodulation. This does not rule out the tantalising possibility that this 314 radiosensitisation could be further enhanced by immunomodulatory effects in an 315 immunocompetent environment. This is worthy of further investigation. 316 317 B. acidifaciens abundance was enhanced by a high soluble fibre diet in this study and B. 318 acidifaciens was identified as a potential radiosensitiser because its abundance was positively 319 correlated with tumour response to radiation and survival time in the IR cohort. This 320 bacterium was first isolated in 2000 and was so named because its reduces the pH level of 321 showing 98 µM) and all tumour butyrate levels could not be resolved above the signal to noise 367 of the chromatogram (limit of quantification 10 µM) but this does not rule out that the effect 368 comes from butyrate.

Overview of study design 401
In this study, we aimed to evaluate the tumour growth after irradiation in mice fed high or 402 low fibre diets and to correlate diet-induced microbiome changes with tumour growth. Two 403 cohorts of CD1 nude female mice were studied. Two types of samples were collected, namely 404 faeces and caecal contents. The samples from the first cohort were collected when the 405 tumours had grown to 50 mm 3 , and consisted of 15 mice in five groups (n=3 per group) fed 406 diets comprising: low fibre (< 0.2% cellulose), low fibre with butyrate (98% sodium butyrate, 407 Sigma Aldrich), high mixed fibre (5% cellulose, 5% inulin), high insoluble fibre (10% cellulose), 408 high soluble fibre (10% inulin) (see Table S1 for details). These diets were formulated by  Yu Ke at Research Diets, Inc. on 11/6/2018. The samples from the second cohort were 410 collected when the tumours had reached 350 mm 3 either without (n=3) or following (n=8) 411 radiation, and consisted of four groups fed diets comprising: low fibre, high soluble fibre, high 412 insoluble fibre and high mixed fibre. 16S rRNA gene sequencing was performed on a MiSeq 413 platform to investigate the bacterial components of the mouse gut microbiome. 414 Metagenomic analysis was conducted using a QIIME2 platform to identify the features 415 (differentially abundant taxa or diversity index) that correlated to the phenotype. 416 Mice and mouse diets 417 CD1-nude female mice (Charles River Laboratories, USA 6-7 weeks) were housed in a 418 temperature-controlled environment with a 12-h reversed-phase light/dark cycle (lights on 419 21:00 h) and provided with food and water ad libitum. At 7 to 8 weeks of age, mice were 420 injected subcutaneously with RT112 bladder cancer cells (DSMZ, Germany) and started 421 receiving either a low fibre diet (2 g cellulose/3850 kcal), a high insoluble fibre diet (100 g 422 cellulose/3850 kcal), a high soluble fibre diet (100 g inulin/3850 kcal) or a high mixed fibre 423 diet (50 g cellulose + 50 g inulin / 3850 kcal) for a maximum time of 9 weeks or until they were 424 culled when the tumours reached 350 mm 3 . Faeces, caecal contents, and proximal and distal 425 colons from the first cohort were taken when the tumour reached 50 mm 3 (each group n=3) 426 without irradiation to investigate the microbiome at baseline. Faeces and caecal contents 427 from the second cohort were taken when the tumour reached 350 mm 3 after IR (each group 428 n=8) or without IR (each group n=3) or at the end of study (9 weeks after xenograft) to study 429 the association between the gut microbiome composition and tumour response. Thereafter, 20 µL of sample or standard was taken and 10 µL of internal standard (valeric acid, 456 Alfa Aesar, UK) added prior to the addition of 5 µL 15% percholoric acid. Samples were mixed 457 and centrifuged at 12,000 g for 15 min at 4°C followed by direct injection (10 µL All 16S rRNA gene-based metagenomic analysis was conducted using a QIIME2 platform [57]. 491 Quality filtered sequences with >97% identity were clustered into bins known as Operational 492 Taxonomic Units (OTUs), using open-reference OTU picking. The relative abundance of each 493 OTU was obtained from all samples. In the taxonomic analysis, the microbiome at the phylum, 494 class, order, family, genus and species levels was classified with reference to the Greengenes 495 database [58]. 496 The analysis pipeline was as follows: 497 i. All sequences were trimmed to a length of 240, since the quality dropped above 498 this length based on the sequence quality plots. 499 ii. De-noised sequencing errors by using the "Deblur" plugin in QIIME2 [59]. 500 iii. Taxonomic assignment was performed with Greengenes [60] by the "feature-501 classifier" command. 502 iv. To visualise the differences in microbial composition between gut contents and 503 tissue, a taxonomic profile was generated by conducting differential abundance 504 analysis using balances in gneiss. 505 v. To identify the features characterising the differences between groups, the LEfSe 506 method of analysis was performed to compare abundances of all bacterial clades 507 [61]. By validation using the Kruskal-Wallis test at the α setting of 0.05, effect size 508 was obtained by LDA (linear discriminant analysis) based on the significantly 509 different vectors resulting from the comparison of abundances between groups. 510 vi. To validate the significance of enrichment of bacterial taxa among different 511 groups, discrete false-discovery rates (DS-FDR) were calculated [62] 512 vii. A phylogenetic tree was generated by using the "phylogeny" plugin in QIIME2. 513 viii. To investigate the alpha and beta diversity, the diversity commands of "alpha-514 group-significance" and "beta-group-significance" were used to obtain Shannon's 515 index, and Bray-Curtis dissimilarity. A principal coordinates (PCoA) plot was 516 obtained by using the Emperor Tool based on the results of Bray-Curtis 517 dissimilarities. 518 ix. The OTU table was rarefied using the "alpha-rarefraction" command in QIIME2. 519 The alpha rarefraction plot showed the richness of the samples with increasing 520 sequence count. 521 x. To predict the metagenome functional profiles, PICRUSt, a bioinformatics 522 software package, was used to collapse predicted functions (KEGG Orthology; KO) 523 based on 16S rRNA surveys into higher categories (KEGG pathway) after picking 524 OTUs and normalisation [63]. 525

Statistical analysis 526
Power calculations for the number of mice per group were done using G*Power software 527 version 3. 1.9.4 [64]. Alpha diversity and relative abundance of specific bacterial taxa were 528 compared using the Kruskal-Wallis test following by Dunn's multiple comparison test. All mice 529 were classified into high, intermediate or low diversity groups based on tertiles of distribution. 530 Time to treble in volume was defined as the interval (in days) from the date of irradiation 531 (growth to 50 mm 3 ) to the date for the tumour to treble in volume. Tumour growth curves 532 were analysed for each group, and their slopes were compared using one-way ANOVA. The 533 LEfSe method of analysis was applied to determine the difference in bacterial taxa, using the 534 Kruskal-Wallis test. Significantly different taxa presented from the previous comparison 535 applying LEfSe method were used as input for LDA, which produced an LDA score. Volcano 536 plots showed the significance of the taxa which are different among different groups, with 537 log10 (FDR-adjusted p-values) on the y-axis and median-adjusted effect sizes on the x-axis. In 538 addition, mice were also classified as having high, intermediate and low abundance of B. 539 acidifaciens or high and low abundance of parabacteroides genus based on the relative 540 abundance of these taxa in the gut microbiome sample. All analyses were conducted in 541 QIIME2 and GraphPad Prism version 8.0 (La Jolla, CA). 542