Effects of a high-fiber, high-fruit and high-vegetable, low-fat dietary intervention on the rectal tissue microbiome

Abstract

Background

Emerging evidence suggests that bacteria residing in colorectal tissue are plausibly associated with colorectal cancer. Prior studies investigated the effects of dietary interventions on the fecal microbiome, but few assessed colorectal tissue microbiome endpoints. We investigated the effects of a high-fiber, high-fruit, high-vegetable, and low-fat dietary intervention on the rectal tissue microbiome in the Polyp Prevention Trial (PPT).

Methods

PPT is a 4-year randomized clinical trial with intervention goals of consuming (1) at least 18 g of fiber per 1000 kcal/day; (2) at least 3.5 servings of fruits and vegetables per 1000 kcal/day; and (3) no more than 20% of kcal/day from fat. Using 16S ribosomal RNA gene sequencing, we characterized bacteria in rectal biopsies collected at baseline and the end of years 1 and 4 (n = 233 in intervention arm and n = 222 in control arm). We estimated effects of the intervention on alpha and beta diversity and relative abundance of a priori–selected bacteria using repeated-measures linear mixed-effects models.

Results

The intervention did not statistically significantly modify rectal tissue alpha diversity. Compared with the control arm, relative abundance of a priori–selected Porphyromonas (absolute intervention effects [standard errors] at T1 vs T0 = –0.24 [0.07] and T4 vs T0 = –0.12 [0.07]; P = .004) and Prevotella (absolute intervention effects at T1 vs T0 = –0.40 [0.14] and at T4 vs T0 = –0.32 [0.15]; P = .01) were more strongly decreased in the intervention arm.

Conclusion

The PPT intervention did not influence rectal tissue microbiome diversity or the relative abundance of most bacteria, except for 2 oral-originating bacteria that were previously associated with colorectal cancer presence.

Introduction

The gut contains trillions of microbes that are plausibly associated with the development and progression of multiple health conditions, particularly colorectal cancer (CRC).^1-4 CRC is the second overall leading cause of cancer death in the United States, with more than half of CRCs attributable to modifiable lifestyle factors including diet and other exposures.⁵ In turn, these exposures may influence the composition of the gut microbiome, which may be targeted to inform primary and secondary prevention of CRC.^6,7

Multiple intervention studies found effects of dietary patterns on the gut microbiome.^6,8-11 For example, diets high in fruits, vegetables, and fiber increased fecal alpha diversity and abundance of short chain fatty acid–producing bacteria and lowered the abundance of potentially carcinogenic bile acid–producing bacteria in fecal samples.^6,8-10 However, the majority of previous dietary interventions collected fecal samples to study changes in the gut microbiome.^6,8-10 Accumulating evidence supports a colorectal tissue-specific microbiome that is distinct from fecal samples.^12,13 In comparison with its luminal fecal counterparts, the tissue may reflect the colonic bacterial community that inhabits the epithelial crypts and is therefore potentially more representative of the resident microbes.¹⁴ The tissue microbiome has been strongly associated with the presence of CRC and with CRC outcomes.^3,4,15-20 For example, multiple bacteria in tumor and normal tissue collected from the colon or rectum—such as oral-originating Porphyromonas, Prevotella, and Fusobacterium—have been associated with CRC presence and with CRC recurrence in human and mechanistic animal studies.^3,14,19 To our knowledge, to date, there are no studies among humans investigating the effects of diet on the gastrointestinal tissue microbiome in human interventional settings.

Herein, we investigated the effects of a high-fiber, high-fruit, high-vegetable, and low-fat dietary intervention on the rectal tissue microbiome in the Polyp Prevention Trial (PPT).

Methods

Study design and population

The PPT²¹ is a 4-year multicenter, randomized, controlled trial that investigated the effect of a high-fiber, high-fruit, high-vegetable, and low-fat diet on colorectal adenoma recurrence. The original study included men and women aged older than 35 years, with at least 1 histologically confirmed colorectal adenomatous polyp removed 6 months prior to baseline. The exclusion criteria and changes in dietary intake within the intervention arm are detailed elsewhere.²¹ All participants provided written informed consent, and the study was approved by the institutional review boards at the National Cancer Institute and participating centers (OH91C0159-B), with the original trial registered under identifier NCT00339625.

Dietary intervention

The specific goals of the dietary intervention included (1) limiting fat to 20% of kcal/day; (2) consuming at least 18 g of fiber per 1000 kcal/day; and (3) consuming at least 3.5 servings of fruits and vegetables per 1000 kcal/day. Dietary goals were calculated based on each participant’s total energy intake as determined by the baseline food frequency questionnaire, and participants received a communication with their individual dietary goals and behavior-modification techniques. The intervention group continued to receive dietary counseling by a nutritionist throughout the trial. Individuals in the control arm were provided with general dietary guidelines from the National Dairy Council with no additional nutritional or behavioral information.

At each annual follow-up visit, participants answered questionnaires collecting demographic, behavioral, and medical history information and had dietary goal achievement determined by (1) nutritionist assessment and (2) their responses to a 4-day food record followed by the Block/ National Cancer Institute food frequency questionnaire modified to assess more detailed intakes of low-fat and high-fiber foods. Each year the investigators also administered unscheduled 24-hour dietary recall questionnaires to a newly selected 10% random sample. The food frequency questionnaire ascertained dietary intakes over the past year and average serving sizes. Compared with the 4-day food record and the 24-hour recall, the food frequency questionnaire slightly overestimated fat and underestimated fiber, fruit, and vegetable intake.^22,23

As described previously in prior manuscripts by our team,^24,25 in addition to our overall analysis, we explored limiting our analysis to those who self-reported adhering to the dietary intervention (super-compliers; n = 48) and comparable goal-achieving control arm participants (n = 64), among whom the strongest effects of the diet intervention on adenoma recurrence were observed. Briefly, based on annual food frequency questionnaire responses, super-compliers were defined as those who completed 9-12 goals over 4 years of the above-described fat, fruit and vegetable, and fiber goals.^24,25 For example, the strictest complier (ie, meeting 12 goals) would have met each of the fat, fiber, and fruit and vegetable goals every year of follow-up for the 4-year trial. To create a comparable comparison group, we selected control arm participants who completed the study and had no missing data on dietary goals over follow-up. Then, using their annual food frequency questionnaire, we summed the dietary goals (eg, consuming <20% of kcal/day from fat) met at each follow-up and ranked the sum. Approximately 21% of intervention participants with microbiome data were super-compliers, therefore we similarly selected the top 21% of ranked control arm participants into this study and considered them goal-achieving.

Rectal tissue microbiome analysis

Additional details on the biopsy collection and microbiome analyses were provided previously.²⁶ Briefly, rectal biopsies were obtained and frozen immediately without a fixative at baseline, year 1, and year 4 from 455 participants. Then, they were lysed using an enzymatic cocktail, homogenized in a Bead Ruptor (Omni International, Inc, Kennesaw, GA, USA), and centrifuged. The Animal Tissue DNA Extraction Kit (AutoGen, Holliston, MA, USA) was used for DNA extraction. The V4 region of the 16S ribosomal RNA gene was polymerase chain reaction–amplified for 30 cycles, and 2 x 250 bp paired end sequencing was performed on the Illumina MiSeq v2 using the 500-cycle kit (Illumina, San Diego, CA, USA).

Using the DADA2 pipeline 1.2.1,²⁷ sequence variant tables and phylogenetic trees were generated based on pair-end sequence reads. After merging and error correction, amplicon sequence variants (ie, 100% operational taxonomic units) were identified. Taxonomy was assigned to the resulting amplicon sequence variants using the SILVA v123 database, and nonbacterial sequences were removed.

Observed amplicon sequence variants, Shannon Index, and the Faith phylogenetic diversity were computed based on the whole community (ie, the full amplicon sequence variants table) using QIIME 1.9.1. Based on rarefaction curves for alpha diversity, we rarefied the alpha and beta diversity metrics to 8000 reads; this reduced the specimen sample size from 1059 to 1030 rectal biopsies. For this study, we selected taxa a priori based on prior associations with CRC including Bacteroides, Fusobacterium, Porphyromonas, Parvimonas, Peptostreptococcus, Gemella, Prevotella, Solobacterium, and Dialister genera in addition to the order Clostridiales.^3,4 Further, in exploratory analyses, we considered genera present in 50% of the population at a mean relative abundance of more than 0.1% (n = 85).

Statistical analysis

We compared baseline participant characteristics between study arms using χ² tests for categorical variables, analysis of variance (ANOVA) for normally distributed continuous variables, and Kruskal–Wallis tests for nonnormally distributed variables.

Mean values for the alpha diversity metrics and relative abundances of bacteria were calculated for the intervention and control groups for baseline, year 1, and year 4. We then conducted analyses estimating the effect of the dietary intervention on mean year-1 and year-4 alpha diversity and taxa relative abundance and presence using repeated-measures linear mixed-effects models to calculate absolute intervention effects, which can be defined as (T1 or T4 mean microbiome metric_intervention—baseline mean microbiome metric_intervention) – (T1 or T4 mean microbiome metric_control—baseline mean microbiome metric_control). All mixed models included a random effect for participant, the intercept, indicators for intervention group and follow-up time (baseline or follow-up), and an intervention*follow-up interaction term. The P values for the intervention*follow-up interaction term, representing overall intervention effects, were calculated using likelihood ratio tests. We also explored including covariates (eg, age, smoking history, sex) in the model based on causal diagrams and previous literature but found that their inclusion did not meaningfully change our findings, so these were not included in the main results for the mixed effects models because of the randomized nature of this study. To estimate effects of the intervention on beta diversity, we used permutational multivariate analysis of variance (PERMANOVA) tests including indicators for intervention group, follow-up time (baseline or follow-up), and an interaction term for the intervention*follow-up (from which the P value was taken) using the Adonis function in package Vegan²⁸ with a strata statement to account for repeat measures.

In stratified analyses, we explored limiting our study population to super compliers and goal-achieving participants. We also repeated the above mixed effects model analyses stratifying the intervention arm by completion of (1) fiber goals T1 and T4, (2) fat goals at T1 and T4, and (3) fruit and vegetable goals at T1 and T4.

To assess whether habitual pretrial diet assessed via baseline food frequency questionnaire was associated with the gut microbiome metrics, we calculated partial Spearman correlations among the microbiome metrics and pretrial fat, fiber, and fruit and vegetable intake. Pretrial dietary intakes were energy adjusted using the nutrient density method. We included covariates for age at random assignment; sex; intervention arm; body mass index; baseline adenoma characteristics; education; study center; baseline smoking status; family history of CRC; regular nonsteroidal anti-inflammatory drug or aspirin use; and total energy, alcohol, fat, fruits and vegetables, and fiber intake.

P values were considered statistically significant at an alpha threshold of .05, with the exception that in the exploratory analyses, we adjusted for multiple testing using Bonferroni-corrected alpha levels for relative abundance (P = .05 dividing by 78 taxa = 0.001). All statistical tests were 2-sided.

Results

Characteristics of the 455 participants included are listed in Table 1 by intervention arm. There were 333 individuals with baseline samples, and of these, 225 had baseline, year-1, and year-4 follow-up samples available. Additionally, 41 individuals had baseline and year-1 samples only (but no year-4 sample), and 20 had baseline and year-4 samples only (but no year-1 sample). Those in the intervention group on average met 5.2 goals over the 4 years, whereas those in the control group met an average of 1.2 goals in our study population (P < .001). Overall, there were no other major differences between the 2 arms of the study.

Table 1.

Baseline characteristics of 455 participants of the Polyp Prevention Trial, 1991–1998

Characteristicsa	Intervention arm (n = 233)	Control arm (n = 222)	P b
Age at random assignment, mean (SD), y	61.21 (10.22)	61.90 (9.74)	.46
Center, No. (%)
California	53 (22.7)	43 (19.4)	.65
New York, Pennsylvania, Illinois, North Carolina, Virginia	117 (50.2)	119 (53.6)
Utah	63 (27.0)	60 (27.0)
Male, No. (%)	162 (69.5)	155 (69.8)	1.00
Postgraduate college, No. (%)	65 (27.9)	74 (33.3)	.25
Family history of colorectal cancer, No. (%)	64 (27.5)	61 (27.5)	1.00
Aspirin and nonsteroidal anti-inflammatory drug use, No. (%)	90 (38.6)	87 (39.2)	.98
Smoking status, No. (%)
Current smoker	37 (15.9)	27 (12.2)	.39
Former smoker	103 (44.2)	95 (42.8)
Never regular smoker	93 (39.9)	100 (45.0)
Body mass index, mean (SD), kg/m2	27.53 (3.95)	28.01 (4.07)	.28
Advanced or multiple adenomas at baseline, No. (%)	119 (51.1)	109 (49.1)	.50
Adenoma located in colon at baseline, No. (%)	161 (87.0)	154 (84.2)	.52
Time between baseline colonoscopy and biopsy, mean (SD), d	133 (51.5)	125 (53.6)	.15
Time between year 1 colonoscopy and biopsy, mean (SD), d	53.8 (80.9)	49.8 (71.2)	.62
Time between year 4 colonoscopy and biopsy, mean (SD), d	13.6 (49.7)	11.9 (36.1)	.73
Biopsy collected after baseline colonoscopy, No. (%)	169 (72.5)	164 (73.9)	.83
Biopsy collected after year 1 colonoscopy, No. (%)	165 (88.2)	154 (85.1)	.46
Biopsy collected after year 4 colonoscopy, No. (%)	117 (76.5)	93 (66.0)	.06
Dietary intervention goals met, mean (SD) No. goals	5.21 (3.53)	1.22 (1.80)	<.001

^aContinuous variables are presented as means (SD), and categorical variables are presented as No. (percentages based on the full sample size as the denominator).

^b P values were calculated using χ² test for categorical variables, analysis of variance for normally distributed continuous variables, and Kruskal–Wallis test for nonnormally distributed continuous variables.

Though rectal tissue alpha diversity generally decreased over time—more strongly so in the intervention arm—the effects of the intervention on alpha diversity were not statistically significant (Table 2). There were no major differences in the effects by sex. All PERMANOVA P values for the effects of the dietary intervention on beta diversity were not statistically significant for Bray–Curtis (P = .64), unweighted Unifrac (P = .77), or weighted Unifrac (P = .90) distance matrices. As shown in Figure 1, the relative abundance of 2 a priori oral cavity–originating genera, Porphyromonas and Prevotella, was more strongly decreased in the intervention arm than the control arm. For example, from baseline to year 1, in the intervention arm, Porphyromonas decreased from an average relative abundance of 0.40 (95% confidence interval [CI] = 0.31 to 0.49) to 0.24 (95% CI = 0.15 to 0.32) in the intervention arm, whereas it increased from 0.19 (95% CI = 0.10 to 0.27) to 0.26 (95% CI = 0.18 to 0.35) in the control arm (absolute intervention effects [SD] = –0.24 [0.07] and T4 vs T0 = –0.12 [0.07]; P_arm*visit = .004). Similarly, Prevotella relative abundance decreased on average more substantially in the intervention arm than the control arm (absolute intervention effect [SE] at T1 vs T0 = –0.40 [0.14] and at T4 vs T0 = –0.32 [0.15]; P = .01) (see Table S1 for exact estimates). Among the genera selected for exploratory analyses, there were no changes in relative abundance bacteria reaching statistical significance, though there were suggestions of stronger decreases in Ezakiella and Oscillibacter relative abundance in the intervention arm (Table S1).There were no statistically significant effects on presence of a priori bacteria in the tissue (Table S2).

Table 2.

Effects^a of the high-fiber, high-fruit and vegetable, and low-fat dietary intervention on rectal tissue bacteria alpha diversity, Polyp Prevention Trial, 1991–1998 (n = 455)

Microbiome metric	Baseline mean (95% CI)a	Year 1 mean (95% CI)a	Year 4 mean (95% CI)a
Overall
Observed amplicon sequence variants
Intervention	201 (192 to 210)	185 (176 to 193)	180 (171 to 189)
Control	192 (183 to 201)	184 (175 to 193)	180 (171 to 189)
Intervention effectb	Referent	−8.77 (7.26)	−9.46 (7.48)
Pc			0.37
Shannon
Intervention	5.54 (5.38 to 5.69)	5.21 (5.07 to 5.36)	5.15 (5.00 to 5.31)
Control	5.37 (5.22 to 5.53)	5.25 (5.10 to 5.40)	5.20 (5.05 to 5.36)
Intervention effectb	Referent	−0.20 (0.14)	−0.21 (0.15)
Pc			0.27
Faith Phylogenetic Diversity
Intervention	22.03 (21.34 to 22.71)	20.88 (20.22 to 21.53)	20.60 (19.92 to 21.29)
Control	21.49 (20.79 to 22.19)	20.90 (20.23 to 21.57)	20.44 (19.73 to 21.14)
Intervention effectb	Referent	−0.56 (0.56)	−0.37 (0.58)
Pc			0.61
Men
Observed amplicon sequence variants
Intervention	200 (190 to 211)	189 (179 to 200)	184 (173 to 195)
Control	189 (179 to 200)	181 (171 to 191)	180 (169 to 191)
Intervention effectb		−2.87 (8.43)	−7.13 (8.77)
Pc			0.72
Shannon
Intervention	5.53 (5.35 to 5.71)	5.31 (5.14 to 5.48)	5.23 (5.04 to 5.41)
Control	5.36 (5.17 to 5.54)	5.25 (5.07 to 5.42)	5.23 (5.05 to 5.42)
Intervention effectb		−0.11 (0.17)	−0.18 (0.17)
Pc			0.58
Faith Phylogenetic Diversity
Intervention	21.87 (21.06 to 22.69)	21.10 (20.33 to 21.88)	20.91 (20.09 to 21.74)
Control	21.22 (20.41 to 22.04)	20.49 (19.71 to 21.28)	20.52 (19.68 to 21.36)
Intervention effectb		−0.04 (0.67)	−0.26 (0.69)
Pc			0.92
Women
Observed amplicon sequence variants
Intervention	203 (186 to 219)	173 (157 to 190)	171 (155 to 187)
Control	198 (181 to 215)	192 (175 to 209)	179.62 (163 to 197)
Intervention effectb		−23.16 (14.07)	−13.58 (14.15)
Pc			0.26
Shannon
Intervention	5.56 (5.28 to 5.84)	4.97 (4.70 to 5.25)	5.00 (4.73 to 5.28)
Control	5.43 (5.13 to 5.72)	5.26 (4.98 to 5.54)	5.13 (4.84 to 5.41)
Intervention effectb		−0.42 (0.28)	−0.26 (0.28)
Pc			0.33
Faith Phylogenetic Diversity
Intervention	22.37 (21.13 to 23.60)	20.34 (19.11 to 21.57)	19.94 (18.72 to 21.16)
Control	22.16 (20.85 to 23.46)	21.93 (20.67 to 23.18)	20.26 (18.99 to 21.54)
Intervention effectb		−1.80 (1.05)	−0.54 (1.05)
Pc			0.21

^aMeans and 95% confidence intervals are least squared means from linear mixed effects models, with a random effect for subject and an interaction term for visit*intervention arm. Abbreviation: CI = confidence interval.

^bBeta coefficients and standard errors are from linear mixed effects models, with a random effect for participant and an interaction term for visit*intervention arm; the P_interactions for effect modification by sex were not statistically significant (all P > .05).

^c P values were estimated using likelihood ratio tests.

Figure 1.

Volcano plot of intervention effects comparing baseline and (A) year 1 relative abundances and (B) year 4 relative abundances for a priori–selected bacteria in the Polyp Prevention Trial, 1991–1998.

Upon limiting our analyses to super compliers and comparable control arm participants, the findings for the dietary effects were generally similar. The effects on alpha diversity were similarly weak (Table S3). The magnitude of the change in the relative abundance of Porphyromonas and Prevotella was similarly decreased in the intervention arm, though these findings were not statistically significant in the smaller sample size. Further, among this subgroup, a priori–selected Gemella was more strongly decreased in the intervention arm at year 4 (absolute intervention effect at T4 vs T0 = –0.41; P = .05). We also explored limiting the analyses to those who adhered to the fiber goals, fat goals, or fruit and vegetable goals at year 1 and year 4. Among those who adhered to the dietary goals at year 1 and year 4, the effect sizes were generally similar in magnitude. One exception was that among those who adhered to the fat goals at year 1 and year 4, there were statistically significant decreases in Fusobacterium relative abundance (absolute intervention effect at T4 vs T0 = –1.04; P = .01; Table S4).

The associations of longer-term fiber, fat, and fruit and vegetable intake as assessed via baseline food frequency questionnaires with the microbiome metrics are shown in Table 3. Overall, self-reported intakes of these dietary factors were not strongly associated with the microbiome metrics; however, fruit and vegetable intake was inversely associated with Faith Phylogenetic Diversity (R = –0.12; P = .03) and Prevotella and Porphyromonas relative abundance (R = –0.11; P = .06 for both); and fruit and vegetable intake was positively associated with relative abundance of order Clostridiales.

Table 3.

Partial Spearman correlations^a among pretrial fat, fiber, and fruit and vegetable intake and the baseline microbiome metrics, Polyp Prevention Trial, 1991–1998 (n = 333)

Microbiome metric	Fiber		Fat		Fruits and vegetables
	R 2	P	R 2	P	R 2	P
Shannon	−0.02	.72	0.06	.28	0.04	.51
Observed amplicon sequence variants	−0.00	.97	0.02	.68	−0.01	.88
Faith phylogenetic diversity	0.08	.16	−0.01	.81	−0.12	.03
Bacteroides	−0.04	.45	0.07	.22	−0.01	.89
Dialister	0.09	.11	0.06	.32	−0.04	.45
Fusobacterium	0.03	.61	−0.02	.66	−0.05	.41
Gemella	−0.02	.69	0.00	.95	−0.03	.58
Parvimonas	−0.05	.33	0.01	.90	−0.02	.73
Peptostreptococcus	0.06	.27	0.04	.52	−0.06	.25
Porphyromonas	0.06	.32	0.01	.90	−0.11	.06
Prevotella	0.06	.31	0.02	.73	−0.11	.06
Solobacterium	0.03	.59	0.10	.07	0.01	.91
Clostridiales	−0.00	.94	−0.01	.92	0.12	.04

^aAdjusted for age at random assignment; sex; intervention arm; body mass index; baseline adenoma characteristics; education; study center; baseline smoking status; family history of colorectal cancer; regular nonsteroidal anti-inflammatory drug or aspirin use; and total energy, alcohol, fat, fruits and vegetables, and fiber intake.

Discussion

In this study, we found that a high-fiber, high-fruit, high-vegetable, and low-fat dietary intervention had minimal effects on rectal tissue bacterial alpha or beta diversity, though there seemed to be a trend toward stronger reductions in alpha diversity among those in the intervention arm. We did however find that the dietary intervention more strongly decreased relative abundance of Porphyromonas and Prevotella and that longer-term fruit and vegetable intake was inversely associated with relative abundance of these bacteria.

Our study is among the first demonstrating influences of dietary patterns on the composition of the mucosal microbiome. The majority of prior dietary interventions to date studied fecal sample endpoints, with some demonstrating relatively quick effects of drastic dietary changes.^6,8-10 These prior studies found that dietary patterns similar to the high-fiber, high-fruit, high-vegetable, and low-fat diet discussed herein generally tended to increase fecal alpha diversity and to increase relative abundance of short chain fatty acid–regulating bacteria including Faecalibacterium, Roseburia, and especially relative abundance of Prevotella.^9-33 In fact, some evidence exists that Prevotella in fecal samples may be a marker of a more prudent, higher carbohydrate diet.^29,34,35 Our findings suggested opposite effects on tissue alpha diversity and relative abundance of Prevotella, which were both decreased by the intervention. This perplexing phenomenon may be biologically plausible as some evidence suggests that bacterial residence may differ between the mucosa and the lumen.³⁶ As further illustration, tissue and fecal microbiomes may have different health-related implications. For example, higher fecal alpha diversity is generally associated with healthier states,³⁷ however, the same may not be true for tissue in which higher bacterial invasion can signal a dysfunctional gut barrier.^38,39 In summary, the microbiome of the mucosa may reflect the effects of diet differently than the microbiome of stool.

As described above, there are few studies that have investigated effects of diet on the colorectal tissue microbiome among humans to allow more direct comparisons of our findings. However, in an animal model, it was demonstrated that a fiber-rich diet promotes the abundance of fiber-degrading microbes supporting intact intestinal mucosal barrier function, whereas a fiber-free diet degrades the intestinal mucosal barrier, potentially bringing bacteria in close contact with the epithelium.³⁹ It is plausible that on a fiber-rich diet consistent with the PPT intervention, fewer microbes are able to invade the epithelium leading to decreases in alpha diversity and in relative abundance of potentially pathogenic bacteria, which were both observed in our study. Overall, our study adds to the limited evidence in the literature for the effects of a high-fiber, high-fruit, high-vegetable, and low-fat dietary intervention on the rectal tissue microbiome.

The findings from our study have potential implications that can be leveraged clinically on the accumulation of further evidence. Although fecal samples offer high value as a noninvasively collected biospecimen that reflects luminal exposures, some literature suggests that the tissue microbiome might be more reflective of the contributions of bacteria to carcinogenesis and may therefore be highly prognostic.^14,19 Prior meta-analyses of the associations of the tissue microbiome with CRC presence and of the fecal microbiome with CRC presence identified many overlapping bacteria that were differentially abundant among individuals with CRC.⁴⁰ Oral-originating taxa like Porphyromonas and Prevotella have historically been more abundant among CRC cases, indicating potential pathogenic functions.⁴⁰ These bacteria were moderately correlated (rho = 0.39; P_adjust = .04, for Porphyromonas and rho = 0.49, P_adjust = .01, for Prevotella) in paired fecal and mucosal samples (n = 32).¹² Additionally, in our prior study in this cohort, we found that rectal tissue alpha diversity and Bacteroides relative abundance was inversely associated with adenoma prevalence, but no other bacteria were strongly associated with adenoma recurrence either prospectively or cross-sectionally.²⁶ Understanding whether dietary effects extend beyond stool passing through the digestive tract to influence the mucosa is crucial toward advancing targeted microbial interventions.

Our study had a number of strengths, including that it was conducted within a well-characterized, randomized intervention trial setting. We had detailed diet and lifestyle information, allowing for stratification by adherence patterns and assessment of associations of longer-term dietary intakes with the rectal tissue microbiome. There were also some limitations. Our analyses only represent a snapshot of the colorectal tissue found in the rectum. The microbiome of rectal tissue may differ from colon tissue because of spatial variations in oxygen, pH, and water content.⁴¹ Despite these regional differences, numerous studies have shown that microbial communities within a single individual are more similar to each other than to those of other individuals, regardless of the sampling site.^41-43 Additionally, from a practical perspective, rectal biopsies offer a feasible option for studying the tissue-associated microbiome.^3,44,45 There was a somewhat small sample size of individuals who adhered to the dietary intervention over the 4 years, perhaps attenuating our findings. It is possible that dietary exposures over longer periods of time (more than 4 years or earlier in the life course) may more strongly influence the tissue microbiome.³⁵ We found that longer-term self-reported food frequency questionnaire intakes were weakly associated with the microbiome metrics. Individuals in this study underwent baseline colonoscopies, removing existing adenomas and inherently disrupting the gut microbiome. It is unknown whether this could have attenuated the effects of the dietary intervention on the microbiome. However, all participants underwent baseline colonoscopies prior to tissue collection meaning that these effects would likely be nondifferential, likely attenuating our findings. We conducted multiple tests, so chance findings are possible even though we took an a priori approach and adjusted for multiple testing.

In summary, our findings suggest that an intervention emphasizing a diet high in fiber and fruits and vegetables and low in fat had minimal impact on rectal tissue microbiome diversity and the relative abundance of most bacterial species, except intriguingly, 2 oral-originating bacteria previously linked to CRC presence. Future intervention studies, including paired tissue (collected throughout the colon and rectum) and fecal samples collected from larger, diverse study populations for longer periods of time are warranted to inform potential primary prevention strategies for CRC.

Author contributions

Doratha A. Byrd, PhD, MPH (Writing—original draft; Writing—review & editing), Maria Gomez, MPH (Writing—original draft; Writing—review & editing), Stephanie Hogue, MPH (Writing—original draft; Writing—review & editing), Yunhu Wan, PhD (Writing—original draft; Writing—review & editing), Ana Ortega-Villa, PhD (Writing—original draft; Writing—review & editing), Andrew Warner, PhD (Writing—original draft; Writing—review & editing), Casey Dagnall, PhD (Writing—original draft; Writing—review & editing), Kristine Jones, PhD (Writing—original draft; Writing—review & editing), Belynda Hicks, PhD (Writing—original draft; Writing—review & editing), Paul Albert, PhD (Writing—original draft; Writing—review & editing), Gwen Murphy, PhD, MPH (Writing—original draft; Writing—review & editing), Rashmi Sinha, PhD (Writing—original draft; Writing—review & editing), and Emily Vogtmann, PhD MPH (Writing—original draft; Writing—review & editing).

Supplementary material

Supplementary material is available at JNCI: Journal of the National Cancer Institute online.

Funding

This study was supported by funding from the Intramural Research Program of the National Cancer Institute at the National Institutes of Health. This project has been funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health, under Contract No. 75N91019D00024. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government.

Conflicts of interest

The authors have no conflicts of interest to disclose.

Data availability

The sequencing and meta data that support the findings of this study are openly available in the National Center for Biotechnology (NCBI) Sequence Read Archive (http://www.ncbi.nlm.nih.gov/ bioproject/PRJNA810087; bioproject ID PRJNA810087). This analysis was not preregistered in an independent, institutional registry.