Decreased Enteric Bacterial Composition and Diversity in South American Crohn’s Disease Vary With the Choice of Treatment Strategy and Time Since Diagnosis

Abstract

Background and Aims

The symptomology of Crohn’s disease [CD], a chronic inflammatory disease of the digestive tract, correlates poorly with clinical, endoscopic or immunological assessments of disease severity. The prevalence of CD in South America is rising, reflecting changes in socio-economic stability. Many treatment options are available to CD patients, including biological agents and corticosteroids, each of which offers variable efficacy attributed to host genetics and environmental factors associated with alterations in the gut microbiota.

Methods

Based on 16S rRNA gene sequencing and taxonomic differences, we compared the faecal microbial population of Brazilian patients with CD treated with corticosteroid or anti-tumour necrosis factor [anti-TNF] immunotherapy. Faecal calprotectin and plasma sCD14 levels were quantified as markers for local and systemic inflammation, respectively.

Results

Anti-TNF treatment led to an increased relative abundance of Proteobacteria and a decreased level of Bacteroidetes. In contrast, corticoid treatment was associated with an increase in the relative abundance of Actinobacteria, which has been linked to inflammation in CD. Disruption of the faecal microbiota was related to decreased bacterial diversity and composition. Moreover, the choice of clinical regimen and time since diagnosis modulate the character of the resulting dysbiosis.

Conclusions

Enteric microbial populations in CD patients who have been treated are modulated by disease pathogenesis, local inflammatory microenvironment and treatment strategy. The dysbiosis that remains after anti-TNF treatment due to decreased bacterial diversity and composition abates restoration of the microbiota to a healthy state, suggesting that the identification and development of new clinical treatments for CD must include their capacity to normalize the gut microbiota.

1. Introduction

Inflammatory bowel disease [IBD] is characterized by chronic inflammation of the gastrointestinal [GI] tract,^1–11 and for which the severity, clinical course and response to treatment are modulated by host genetics, environment and the enteric microbiota.^6,12–14 Crohn’s disease [CD] affects all segments of the GI tract with symptoms such as pain, diarrhoea, nausea and vomiting that do not always correlate with clinical, endoscopic or immunological assessments of disease severity.^5,15,16 The incidence of CD is 200 of every 100 000 adults in North America and 127 per 100 000 in Europe,^3,13,17 while its prevalence in Central and South America and elsewhere worldwide has risen from 240 to 350 per 100 000 people.^3,6,13,18,19 The increased disease occurrence in South America probably reflects changes in socio-economic stability,^{6,9,10,13,20,21} which may in turn be associated with changes in diet,^13,22,23 environmental exposure,^10,13,24 rates of infection (and the corresponding use of antibiotics), autoimmune disease, and the intestinal microbiome.^10,17,23

Many treatment options have been developed, including a broad range of biological agents that are often more effective than steroids and immunomodulators.^5,14,25–27 Anti-tumour necrosis factor α [anti-TNF-α] agents, such as infliximab [Remicade] and adalimumab [Humira], the only biological agents available in Brazil, are unevenly effective for the induction and maintenance phases of CD.¹⁰ The variable effectiveness of biological and pharmacological treatments may be attributed to host genetics and environmental factors,^13,14 both of which are similarly associated with alterations in the gut microbiota, known as dysbiosis.^{7,10,11,17,28,29} Many studies in North America and Europe have shown that initiation and progression of CD are strongly associated with dysbiosis,^{1,5,7,9,18,25,30} but it is not clear whether this is a cause or consequence of the disease process.¹⁷

The enteric microbiota consists of millions of bacterial, archaeal, fungal and viral communities.^7,12 The composition of the microorganisms varies along the GI tract, with the highest concentration of bacterial communities in the distal colon.²³ The microbiota serves essential functions involving intestinal homeostasis, host metabolism, proper development of the immune system and epithelial barrier, and immune regulation.^{6,7,10,12,17,19,22,23} Dysbiosis occurs because of environmental and host factors including diet, hygiene, use of medication and genetic susceptibility.^{1,3,5,7,10–12,17,23,28} While genome-wide association studies [GWAS] have identified many molecular signatures specific to immune dysfunction in CD, it is notable that they also identified genomic loci that interact with commensal or pathogenic bacteria.^{2,7,9,17,28,31,32} Complementary studies identified microbial communities that are altered in CD, highlighted by a reduction in microbial diversity.^{1,5,6,12,17,33} In the healthy gut, the core microbiota is dominated by Bacteroidetes and Firmicutes with lower relative abundance of Actinobacteria, Proteobacteria and Verrucomicrobia.^{7,10,12,17,28} In contrast, in North American and European CD patients, the faecal microbiota consists of an elevated abundance of Proteobacteria^{5,10,12,19,29,31,34,35} with a decrease in Firmicutes.^{3,10–12,19,29,31,35,36} Importantly, several studies have reported significant differences in their findings regarding levels of Bacteroidetes in CD,^{5,7,10,11,19,29,31,37,38} which may be explained by possible modulation of the microbiota by factors such as host genetics, disease progression, overall patient health, environmental exposure, and (critically for this our study) therapeutic strategy,¹⁴ medical history and country of origin.^{3,5,8,10,18,26}

The vast majority of reports on dysbiosis in CD patients are based on a North American and European population and are not controlled for medication regimens.^{8,11,12,16,27,29,33,36,39–43} Therefore, due to the dramatically increased incidence of CD in developing countries,^13,18,20 we designed this study to characterize the microbiota of South American CD patients, with a particular focus on contrasting two clinical regimens: corticosteroid versus anti-TNF-α. The goal of this study was to understand the impact of different CD treatments on the enteric microbiota in a South American population and to further define potential confounding factors that may promote and influence disease progression and treatment.

2. Materials and Methods

2.1. Study subjects

A total of 47 adult patients previously diagnosed with CD and 31 healthy volunteers were recruited from Hospital Israelita Albert Einstein in São Paulo, Brazil. Disease activity and treatment strategy was determined by the Crohn’s Disease Activity Index [CDAI] (remission [CDAI < 150], active [CDAI ≥ 150]) and calculated during their evaluation visit. All anti-TNF-treated patients initially started with corticoid treatment and based on their CDAI scores and faecal calprotectin levels transitioned to anti-TNF agents. All patients were tested for HIV-1/2, HTLV-1/2, syphilis, HCV, HBV and the absence of autoimmune disease. The study was approved by the Ethics Committee of the Hospital Israelita Albert Einstein and written informed consent was obtained from each participant.

2.2. Sample collection and processing

Peripheral blood was collected with EDTA and centrifuged at room temperature. The plasma was transferred to vials and stored at −80°C until use. The stool samples were collected in a dry container and part of the sample was forwarded to the clinical laboratory for quantification of faecal calprotectin. The remaining stool samples were stored at −80°C for future DNA extraction.

DNA was extracted from stool using the Invitrogen Purelink Genomic DNA Kit, as described by the manufacturer [ThermoFisher Scientific]. Briefly, the samples were digested with Proteinase K in digestion buffer at 55°C. After digestion, the residual RNA was removed by RNAse A. The extract was mixed with ethanol and binding buffer, and DNA was bound to the silica-based membrane in the column. DNA was eluted from the column with low salt elution buffer.

2.3. 16S rRNA library preparation and quantification

Bacterial DNA was quantified using the Qubit Broad Range assay [ThermoFisher Scientific]. A concentration of 5 ng/µL was used for library preparation following the Illumina MiSeq ‘16S Metagenomic Sequencing Library Preparation’ protocol targeting the V3 and V4 region of the 16S rRNA gene. Samples were quantified and normalized to 4 nM before library pooling. PhiX was used as an internal control and a MiSeq reagent kit v3 2× 250 cycles was used. Fastq files were used for downstream analysis.

2.4. Quality control and pipeline analysis

Processing of Fastq files was performed using QIIME.⁴⁴ Low-quality sequences were filtered out and reverse primers were truncated. Sequences were clustered to collapse similar sequences [with at least 97% identity between sequences] in order to build operational taxonomic units [OTUs]. Chimeras were then identified and removed. Machine learning-based taxonomic classification using a Naïve Bayes classifier was then performed with GreenGenes as the reference database. A feature [count] table was then built by collapsing features that share the same taxonomic assignment.

2.5. Faecal calprotectin and sCD14

Quantification of faecal calprotectin was performed using a commercial sandwich ELISA kit [BÜHLMANN fCAL], as described by the manufacturer [Buhlmann laboratories AG]. Briefly, a monoclonal capture antibody highly specific for heterodimeric calprotectin and polymeric complexes coated the plate. The standards and samples were incubated at room temperature for 30 min. After washing, the horseradish peroxidase-conjugated detection antibody was added. After incubation, the plate was washed again, followed by addition of the colorimetric substrate tetramethylbenzidine. Absorbance was measured at 450 nm by spectrophotometry.

Soluble CD14 [sCD14] was quantified using a commercial sandwich ELISA Kit [Quantikine ELISA], as described by the manufacturer [R&D Systems]. The plasma was diluted 200-fold, and the plate was precoated with a monoclonal antibody specific for human CD14. After sample incubation, a horseradish peroxidase-conjugated polyclonal antibody specific for human CD14 was added. After adding the substrate and stop solution, optical density was measured at 450 nm by spectrophotometry.

2.6. Statistical analysis

Analysis of variance [ANOVA] was used to test for differences between the groups for the percentage of the relative abundance of the core microbiota, diversity indices and immune mediators [calprotectin and sCD14] followed by a Tukey comparison. R software (v 3.6.0 [2019-04-26]—‘Planting of a Tree’) was used for the principal components analysis [PCA]. A multivariate ANOVA [MANOVA] was used to calculate the statistical significance between treatments for the PCA data, in addition to the Wilcoxon test for paired analysis. Pearson’s correlation coefficient was used to estimate the association between two continuous measurements [e.g. age, sex, time since diagnosis and disease state]. The Chi-square test was performed to assess for differences due to disease state between treatment groups. Multivariable linear regressions were performed to assess the effect of treatment on the diversity indices and the relative abundance of the core microbiota after adjusting for the effects of the following confounders: age, sex, time since diagnosis and disease state. Statistical analysis was performed using SAS, R software (v 3.6.0 [2019-04-26]—‘Planting of a Tree’) and GraphPad PRISM v8.1.1. All tests were two-sided, and a p-value less than 0.05 was considered statistically significant.

3. Results

3.1. Decreased bacterial diversity and composition with CD treatment

To characterize the gut microbiota of South American CD patients [Table 1] and to compare anti-TNF and corticoid clinical regimens, DNA from human faecal samples was sequenced for bacterial 16S rRNA gene content and profiled by studying the microbial community taxonomic composition. Unbiased analysis of the β-diversity among the three groups showed differences in the microbial composition at the phylum level [p = 0.0003] [Figure 1A]. A significant difference was further observed when comparing anti-TNF treatment and the non-IBD control [PC1 p = 0.01, PC2 p = 0.002] [Supplementary Figure 1A]. No significant differences were observed between the corticoid treatment and the non-IBD control [Supplementary Figure 1B], suggesting anti-TNF treatment may curtail microbial restoration after initial corticoid treatment. A principal component analysis [PCA] comparison between anti-TNF and corticoid treatments showed a difference in PC2 that approached significance [p = 0.063] [Supplementary Figure 1C]. Significant changes at the family level [p = 0.00005] were also observed between non-IBD controls and both clinical regimens [Supplementary Figure 2].

Table 1.

Demographic and inflammatory parameters of the study cohort

	Non-IBD control	Anti-TNF	Corticoid	p-value
Donors	31	31	16
Sex [F/M]	24/7	20/11	12/4
Age [years] [range]	35 [19–65]	38 [19–58]	39 [20–65]
CDAI	N/A	14–1088	37–549
Calprotectin [µg/g]	105.10 ± 15.87	685.21 ± 111.81	707.03 ± 158.66	<0.0001****
sCD14 [ng/mL]	1353.26 ± 85.73	1155.66 ± 129.31	1158.60 ± 147.90

IBD, inflammatory bowel disease; TNF, tumour necrosis factor; CDAI, Crohn’s Disease Activity Index. The analyses used to test for statistically significant differences are described in the Methods section. A p-value less than 0.05 was considered statistically significant; p-values ≥0.05 are not shown. ****ANOVA with multiple comparisons was used to determine statistical significance for calprotectin levels, p < 0.0001.

Figure 1.

Differences in faecal bacterial composition in anti-TNF-treated Crohn’s disease. Principal component analysis [PCA] was used to identify changes in the bacterial composition among all the treatment groups [p = 0.0003]. PCA was plotted using R software. Statistical analysis was performed using MANOVA and Wilcoxon tests. IBD, inflammatory bowel disease; TNF, tumour necrosis factor.

The dysbiosis observed with various CD treatment strategies in Figure 1 was reflected in additional measurements of bacterial diversity. Samples obtained from CD patients exhibited decreased diversity when compared to non-IBD controls based on the Shannon [Figure 2A], Phylogenetic Diversity [Figure 2B], Simpson [Supplementary Figure 3A] and Chao1 [Supplementary Figure 3B] indices. The Shannon diversity index showed decreased bacterial diversity in CD patients treated with either anti-TNF [p < 0.0001] or corticoid [p = 0.0003] drugs when compared to the index from non-IBD controls [Figure 2A]. Additionally, anti-TNF treatment showed a lower Shannon bacterial diversity [coefficient: −0.60, p = 0.035] when compared to that of the corticoid treatment [Figure 2A]. This decrease in diversity in the anti-TNF treatment group was further analysed by segregating the patients based on the anti-TNF formulation: adalimumab or infliximab [Supplementary Figure 4]. No difference in the Shannon diversity index was observed between adalimumab and infliximab. The decrease in Shannon index diversity between anti-TNF and corticoid treatment was maintained only within the infliximab group [p = 0.0295].

Figure 2.

Crohn’s disease treatment regimens associate with decreased bacterial diversity. Decreased α-diversity was observed using the Shannon [A] and phylogenetic diversity [B] indices in the anti-TNF and corticoid treatment compared to non-IBD controls. Box and whiskers graphs show the median and quartiles for each cohort. ANOVA with multiple comparisons and Tukey correction was used to identify statistical significance. Tables with significant p-values of the multiple comparison analysis are shown beneath each graph. IBD, inflammatory bowel disease; TNF, tumour necrosis factor.

In considering the phylogeny of faecal microbes, anti-TNF [p < 0.0001] and corticoid [p < 0.0001] treatment revealed decreased diversity when compared to the non-IBD control [Figure 2B]. Anti-TNF treatment showed less diversity [coefficient: −7.03, p = 0.1304] than the corticoid treatment when the two regimens were compared [Figure 2B]. Other measures of alpha diversity (Simpson index [dominant species] and Chao1 index [richness]) showed similar results [Supplementary Figure 3]. Overall these findings indicate that anti-TNF treatment of CD patients results in decreased diversity and composition of the bacterial communities in the gut of a South American population, whereas, in general, the microbiome after corticoid treatment largely approximates that of the non-IBD cohort.

3.2. Anti-TNF and corticoid treatments alter the faecal gut microbiota

To characterize the decreased bacterial diversity, phylogeny and composition observed in Figures 1 and 2, we identified bacterial communities associated with this dysbiosis. The gut microbiota in this study consisted of 21 bacterial phyla [Figure 3A] with predominant organisms from the core microbiota: Bacteroidetes, Firmicutes, Actinobacteria, Proteobacteria and Verrucomicrobia. Many of these phyla were shared between the clinical regimens, while certain ones were found mostly in the non-IBD control group or the anti-TNF treatment patients, such as Tenericutes [p < 0.0001] and Fusobacteria [p = 0.042], respectively [Supplementary Figure 5]. Among the identified phyla, we observed alterations in the core microbial communities of Bacteroidetes and Proteobacteria^{7,10,12,17,28} [Figure 3B, C]. The decrease in the relative abundance of the phylum Bacteroidetes was significant in CD patients treated with anti-TNF [p = 0.0226] agents when compared to the non-IBD control, while CD individuals taking corticoids did not show any statistical differences in the abundance of Bacteroidetes when compared to that of non-IBD control and anti-TNF treatment [Figure 3B]. In contrast, Proteobacteria was significantly increased in CD patients treated with anti-TNF in comparison to the non-IBD control individuals [p = 0.0005] and patients using corticoid treatment [p = 0.0475] [Figure 3C]. Firmicutes, Actinobacteria and Verrucomicrobia showed no statistical changes in the gut microbiota compared to the non-IBD control and between treatments [Figure 3D–F]. These results indicate that different CD treatment regimens impact members of the gut microbiota distinctly.

Figure 3.

Bacterial composition differs with Crohn’s disease treatment regimens. Sequencing analysis of bacterial 16S rRNA in DNA isolated from faecal samples revealed 21 bacteria phyla from a South American Crohn’s disease cohort [A], in which each column represents an individual donor and the top eight most abundant bacteria. The relative abundance of the core microbiota, namely Bacteroidetes [B], Proteobacteria [C], Firmicutes [D], Actinobacteria [E] and Verrucomicrobia [F], was compared among the treatment groups. Bacteroidetes was significantly decreased in the anti-TNF treatment group compared to non-IBD control [B] [p = 0.0226]. Proteobacteria was significantly increased in the anti-TNF group compared to non-IBD control [p = 0.0005] and corticoid treatment [p = 0.0475] [C]. Firmicutes, Actinobacteria and Verrucomicrobia showed no statistical difference in relative abundance between treatments [D–F]. Box and whisker graphs denote the ANOVA results and tables representing the Tukey correction are as described in the legend to Figure 2. IBD, inflammatory bowel disease; TNF, tumour necrosis factor.

3.3. Confounders affecting bacterial diversity due to CD treatment regimens

To identify potential confounders influencing the microbial diversity of treated CD patients, a multivariate linear regression analysis was performed. Analysis of the Shannon index [Table 2] in anti-TNF- and corticoid-treated CD patients and non-IBD controls showed that sex [p = 0.43] and age [p = 0.22] were not predictive of the diversity observed. Overall, treatment regimen was the only significant parameter that predicted a decrease in the Shannon index [p < 0.0001], with a decrease of 1.56 for the anti-TNF treatment [p < 0.0001] and a decrease of 0.92 for the corticoid treatment [p < 0.0001] when compared to the non-IBD control group [Table 2]. Interestingly, a comparison of anti-TNF to corticoid treatment, after controlling for the effects of sex [p = 0.43], age [p = 0.81], time since diagnosis in years [p = 0.51] and disease state [p = 0.59], was marginally significant in predicting the Shannon index [p = 0.06].

Table 2.

Multivariable regression analysis of diversity indexes

Diversity index	Factors	Multivariable regression		Factors	Multivariable regression [paired analysis of treatments]
		Coefficient	p-value		Coefficient	p-value
Shannon	Sex [male vs female]	0.16	0.43	Sex [male vs female]	0.25	0.43
	Age [per year increase]	−0.01	0.22	Age [per year increase]	−0.003	0.81
	Treatment		<0.0001	Time since diagnosis [per year increase]	−0.02	0.51
	Anti-TNF vs non-IBD control	−1.56	<0.0001	State [remission vs active]	−0.16	0.59
	Corticoid vs non-IBD control	−0.92	<0.0001	Treatment [anti-TNF vs corticoid]	−0.59	0.06
Simpson	Sex [male vs female]	0.03	0.18	Sex [male vs female]	0.05	0.18
	Age [per year increase]	0.0001	0.38	Age [per year increase]	−0.002	0.38
	Treatment		0.0003	Time since diagnosis [per year increase]	0.001	0.71
	Anti-TNF vs non-IBD control	−0.09	<0.0001	State [remission vs active]	−0.02	0.53
	Corticoid vs non-IBD control	−0.04	0.13	Treatment [anti-TNF vs corticoid]	−0.06	0.10
Chao1	Sex [male vs female]	56.04	0.59	Sex [male vs female]	−81.34	0.47
	Age [per year increase]	−6.49	0.15	Age [per year increase]	1.21	0.82
	Treatment		<0.0001	Time since diagnosis [per year increase]	−17.19	0.05
	Anti-TNF vs non-IBD control	−768.25	<0.0001	State [remission vs active]	−59.97	0.56
	Corticoid vs non-IBD control	−482.16	<0.0001	Treatment [anti-TNF vs corticoid]	−233.35	0.03
Phylogenetic diversity	Sex [male vs female]	2.32	0.45	Sex [male vs female]	−1.73	0.55
	Age [per year increase]	−0.12	0.35	Age [per year increase]	0.08	0.57
	Treatment		<0.0001	Time since diagnosis [per year increase]	−0.19	0.38
	Anti-TNF vs non-IBD control	−25.49	<0.0001	State [remission vs active]	−1.14	0.67
	Corticoid vs non-IBD control	−18.06	<0.0001	Treatment [anti-TNF vs corticoid]	−6.07	0.03

IBD, inflammatory bowel disease; TNF, tumour necrosis factor. Multivariable linear regressions were performed to address the effect of treatment regimen on each diversity index, adjusting for confounders that include sex, age, time since diagnosis [years] and disease state [active vs. remission]. A p-value less than 0.05 was considered statistically significant.

Bacterial richness via the Chao1 index [Table 2] showed that when comparing treatments to the non-IBD control group, sex [p = 0.59] and age [p = 0.15] were not predictive of changes in microbial richness. Overall, treatment regimen was significant in predicting the Chao1 index [p < 0.0001], with a decrease of 768.25 for anti-TNF treatment [p < 0.0001] and a decrease of 482.16 for corticoid treatment [p < 0.0001] when compared to the non-IBD control group. Compared to corticoid treatment, the Chao1 index decreased by 233.35 for the anti-TNF treatment [p = 0.03]. Paired analysis of clinical regimens showed that factors that are not predictors of the Chao1 index include sex [p = 0.47], age [p = 0.82] and disease state [p = 0.56]. However, time since diagnosis significantly negatively correlated with bacterial richness [r = −0.3556; p = 0.02] [Figure 4], indicating that repleting a normal microbiome becomes less likely with advancing disease progression. In contrast, differences in treatment responsiveness between active disease and remission [Table 3] do not correlate with any of the alpha diversity parameters studied for either clinical regimen. These analyses revealed two distinct factors that drive the decreased bacterial diversity found in CD patients: treatment regimen and time since diagnosis.

Figure 4.

Time since diagnosis is associated with decreased bacterial richness. Multivariable regression analysis indicates that time since diagnosis is a predictor of the Chao1 index [Table 2]. A decrease in bacterial richness is observed in patients who have been diagnosed with Crohn’s disease for longer periods of time [years] regardless of treatment: anti-TNF [red dots] and corticoid [blue dots]. Statistical analysis was performed using a Pearson correlation [r = −0.3556; p = 0.02]. TNF, tumour necrosis factor.

Table 3.

Chi-square test between active and remission states

Treatment	Active	Remission
Anti-TNF	14 [46.67%]	16 [53.33%]
Corticoid	7 [43.75%]	9 [56.25%]

Chi-square test p-value was 1; the percentage of the disease state is shown in parentheses.

3.4. Dysbiosis of the core microbiota is driven by intrinsic and extrinsic factors including sex and treatment regimen, respectively

Multivariable linear regression analysis of confounding factors that may influence the core microbiota in treated CD patients was performed. Analysis of the relative abundance of Bacteroidetes [Table 4] showed that age [p = 0.79] is not a predictor for this phylum, while sex [p = 0.03] and treatment regimen [overall p = 0.01] are significant. The increase in Bacteroidetes in males was 11.53% [Table 4]. Anti-TNF treatment showed a decrease in the relative abundance of Bacteroidetes of 16.07% [p = 0.003] when compared to that of the non-IBD control. However, there was no statistically significant difference between the corticoid group and the non-IBD control group [p = 0.29]. Paired analysis of treatment groups showed that age [p = 0.56], time since diagnosis [p = 0.69], disease state [p = 0.93] and treatment regimen [p = 0.21] were not predictive of the Bacteroidetes percentage, while sex [p = 0.08] was somewhat significant.

Table 4.

Multivariable regression analysis for the relative abundance of the core microbiota

Core microbiota	Factors	Multivariable regression		Factors	Multivariable regression [paired analysis of treatments]
		Coefficient	p-value		Coefficient	p-value
Actinobacteria	Sex [male vs female]	−0.11	0.93	Sex [male vs female]	1.20	0.52
	Age [per year increase]	−0.07	0.19	Age [per year increase]	−0.02	0.79
	Treatment		0.08	Time since diagnosis [per year increase]	−0.20	0.16
	Anti-TNF vs non-IBD control	2.06	0.08	State [remission vs active]	−2.35	0.18
	Corticoid vs non-IBD control	2.88	0.04	Treatment [anti-TNF vs corticoid]	−0.66	0.71
Bacteroidetes	Sex [male vs female]	11.53	0.03	Sex [male vs female]	13.43	0.08
	Age [per year increase]	0.06	0.79	Age [per year increase]	0.21	0.56
	Treatment		0.01	Time since diagnosis [per year increase]	0.22	0.69
	Anti-TNF vs non-IBD control	−16.07	0.003	State [remission vs active]	0.63	0.93
	Corticoid vs non-IBD control	−6.87	0.29	Treatment [anti-TNF vs corticoid]	−8.81	0.21
Firmicutes	Sex [male vs female]	−6.93	0.18	Sex [male vs female]	−7.98	0.28
	Age [per year increase]	−0.06	0.77	Age [per year increase]	−0.09	0.79
	Treatment		0.47	Time since diagnosis [per year increase]	−0.04	0.94
	Anti-TNF vs non-IBD control	−6.39	0.22	State [remission vs active]	−5.60	0.41
	Corticoid vs non-IBD control	−2.77	0.66	Treatment [anti-TNF vs corticoid]	−2.14	0.76
Proteobacteria	Sex [male vs female]	−5.20	0.25	Sex [male vs female]	−6.91	0.30
	Age [per year increase]	0.15	0.45	Age [per year increase]	0.09	0.77
	Treatment		0.0007	Time since diagnosis [per year increase]	−0.36	0.47
	Anti-TNF vs non-IBD control	17.69	<0.0001	State [remission vs active]	6.16	0.31
	Corticoid vs non-IBD control	4.03	0.47	Treatment [anti-TNF vs corticoid]	12.06	0.06

IBD, inflammatory bowel disease; TNF, tumour necrosis factor. Multivariable linear regressions were performed to address the effect of treatment on the core microbiota, adjusting for confounders that include sex, age, time since diagnosis [years] and disease state [active vs remission]. A p-value less than 0.05 was considered statistically significant.

Analysis of the phylum Proteobacteria [Table 4] showed that sex [p = 0.25] and age [p = 0.45] were not predictive of its relative abundance. Treatment regimen was significant in predicting the relative abundance of Proteobacteria [overall p = 0.0007], with anti-TNF treatment showing a 17.69% [p < 0.0001] increase in Proteobacteria compared to the non-IBD control group. There was no statistically significant difference between the corticoid group and the non-IBD control group in Proteobacteria percentage [p = 0.47]. After controlling for the effect of sex [p = 0.30], age [p = 0.77], time since diagnosis [p = 0.47] and disease state [p = 0.31], treatment regimen approached significance in predicting the relative abundance of Proteobacteria between treatments [p = 0.06].

Analysis of Actinobacteria [Table 4] showed that sex [p = 0.93] and age [p = 0.19] were not predictive of this bacterial phylum. While overall treatment was not significant in predicting the percentage of relative abundance of Actinobacteria [p = 0.08], there was a statistically significant difference between the corticoid group and non-IBD control group with an increase of 2.88% [p = 0.04]. However, there was no statistically significant difference between the anti-TNF treatment group and the non-IBD control group [p = 0.08]. Paired comparison between treatments showed that sex [p = 0.52], age [p = 0.79], time since diagnosis [p = 0.16], disease state [p = 0.18] and treatment type [p = 0.71] were not predictive of the percentage of Actinobacteria. Similarly, the percentage of relative abundance of Firmicutes was not predicted by sex [p = 0.18], age [p = 0.77] or treatment [p = 0.47] when compared to the non-IBD control. Sex [p = 0.28], age [p = 0.79], time since diagnosis [p = 0.94], disease state [p = 0.41] and treatment type [p = 0.76] were also not predictive of Firmicutes percentage with a paired analysis between treatments. Analysis of the Verrucomicrobia phyla was excluded due to the low levels found in this study. Overall, the dysbiosis observed in treated CD patients associated with a decreased percentage in the relative abundance of Bacteroidetes was dependent on the sex of the patients [Table 4]. In contrast, the increase in the relative abundance of Proteobacteria was due to the different treatment regimens. Thus, both intrinsic and extrinsic factors can drive the dysbiosis observed in CD.

3.5. CD treatment does not improve local inflammation

Alterations in faecal microbiota impact the health of the gut leading to intestinal permeability, bacterial translocation, and local and systemic inflammation.¹⁷ Calprotectin and sCD14 levels, often used as markers of local and systemic inflammation, respectively, were measured in faecal and blood samples from these patient populations [Table 1]. Neither of the CD treatment regimens was able to restore faecal levels of calprotectin to that of non-IBD controls [Supplementary Figure 6A], while no evidence of systemic inflammation was detected, in that similar levels of sCD14 were observed in treated CD patients and non-IBD controls [Supplementary Figure 6B]. Together, these results suggest that a major [as with anti-TNF] or residual [as with corticoid] disruption of the microbiome in CD maintains an elevated state of mucosal inflammation in the colon.

4. Discussion

Dysbiosis is widely reported to occur in CD from North American and European populations due to the impact of both disease pathogenesis and environmental cues.^{1,5,7,9,18,25,30} In this study we asked two questions: [i] Are the changes in the faecal microbial community in CD patients from Brazil different from those in North America and Europe? [ii] Does the treatment strategy modulate and, quite critically, normalize the gut microbiota in CD patients?

Regarding the first question, many studies have identified decreased bacterial diversity and similar microbial communities that are altered in North American/European CD. For example, in CD patients increased levels of Proteobacteria were found,^{5,10,12,14,19,29,31,34,35} which are associated with evasion of pathogen recognition and bacterial clearance and increased intestinal permeability by breaching the intestinal barrier.^45–49 This mechanism allows certain Proteobacteria species to colonize the gut environment, eventually leading to inflammation.^45,50 In agreement with the North American/European studies, we found increased levels of Proteobacteria in the anti-TNF-treated cohort, which is not observed in the corticoid-treated individuals. Based on our observations, anti-TNF treatment shows a greater disruption to the gut microbiota than corticoid treatment or, conversely, corticoid therapeutic regimens maintain baseline levels of Proteobacteria before the patient transitions to anti-TNF treatment.

In addition to the changes in levels of Proteobacteria, we found decreased levels of Bacteroidetes in CD patients who were treated with anti-TNF agents. Overall, the levels of Bacteroidetes in CD remain controversial, as levels depend on the patient’s clinical history, sample collection and processing.^7,38 A meta-analysis of nine studies evaluating 706 IBD patients identified a trend toward lower levels of Bacteroidetes especially in the active phase of the disease.^38,51 This observation was described recently in a Chinese cohort, which established a negative correlation with CDAI,⁵² suggesting that Bacteroidetes may have a negative impact on inflammation. Surprisingly, we found that the response of the microbiota to treatment is associated with the patient’s sex. The decrease in Bacteroidetes after anti-TNF treatment is observed predominantly in females, suggesting that this particular phylum is more susceptible to loss when women are treated with anti-TNF drugs. Our results may provide an explanation for the discrepancy in the CD literature concerning Bacteroidetes,^{5,7,10,11,19,29,31,37,38} in that previous studies did not partition their patients based on sex and clinical regimen.

One of the limitations of this study is the absence of an untreated CD cohort, due to the rapid introduction of treatment. Nonetheless, extrapolating from the microbial profile of the two treatment regimens, we propose that in parallel with the published changes in the microbiome of North American and European CDpatients,^{1,5,7,9,18,25,30,46} we observe in South American patients a similar increase in Proteobacteria and decrease in Bacteroidetes. The similarities between these distinct geographical populations suggest three possible interpretations. [i] The immunological changes in the CD host are dominant in determining the composition of the faecal microbiome, and thus disease pathogenesis is the driving force for dysbiosis in CD. [ii] The two major ethnic populations in São Paulo, Brazil, are of European and indigenous descent. Due to the demographics of the CD patient population at Hospital Israelita Albert Einstein, the vast majority of CD patients bear a predominant European genetic background, and thus the genetics in this unique South American cohort closely resemble those of North American and European CD patients. [iii] The microenvironment, both dietary and sociological in São Paulo, although quite distinct from that in previous studies, contributes very little to faecal microbial dysbiosis in CD. In the future, we will expand these studies to include clinical trial participants from South America who are in the placebo control group to eliminate any bias in the observed effect due to treatment regimens.

Due to the complexity in the pathogenesis of CD, the treatment regimens available are broad and quite variable, depending on age, disease progression, disease presentation and clinical experience.^{3,5,8,10,18,26} Each treatment is designed to initiate the induction and maintenance of remission in the patient,^5,25–27 and the variability in their effectiveness can be attributed to host genetics and the environment.^6,12,13 Our goal here was to contrast the faecal microbiota of CD patients under two of the most widely used regimens, with quite distinct mechanisms of action: corticoids and anti-TNF treatment.^14,53,54 A previous study showed significant differences in the microbial composition between patients responding and not responding to treatment,¹⁴ but this was not replicated in the two Brazilian cohorts in the present study. In addition, we analysed our findings taking into consideration major confounders, such as age, sex, time since disease diagnosis [years] and disease state [active vs. remission] as potential factors for the expected dysbiosis. Several studies have found decreased diversity in the enteric microbiota of CD patients.^{11,29,36,39,53,55} Our results highlight that each clinical regimen leads to changes in the core microbiota, which is reflected by decreased bacterial diversity and composition. Differences between anti-TNF and corticoid treatment are observed via the Shannon diversity index, suggesting that both the number of microbial species and their distribution in the community are different between treatments. Interestingly, both anti-TNF and corticoid treatment seem to share dominant species, with similar phylogenetics via the Simpson and phylogenetic diversity analysis. We propose that symbiotic interactions among microbes stabilize the bacterial communities to a certain degree that is far from the community observed in non-IBD individuals.

By analysing the interrelationship between bacterial communities, we were able to obtain a clear view on the modulation and normalization of the gut microbiota after each clinical treatment. In contrast to other studies,^53,54,56 we found that corticoid treatment has a similar, but not identical bacterial composition as non-IBD controls and that transition to anti-TNF treatment does not restore gut microbial diversity, independent of disease severity. An explanation for this observation is similar to what we observe in the gut in the presence of antibiotics: susceptible microorganisms are prevented from maintaining dominance and thus other microorganisms [pathobionts] gain a competitive growth advantage.

Multivariant analysis identified that time since disease diagnosis [years] is a strong predictor of a decline in bacterial richness [Chao1 index] in the gut, regardless of treatment. Thus, the inability of either treatment to fully normalize the gut microbiota, nor to fully eliminate local inflammation [calprotectin], to a healthy state may reflect the permanence of the dysbiotic population over time. These findings also indicate that as alternative treatments are developed an additional clinical parameter in their evaluation should include the identification of therapeutics that normalize gut microbiota.

In conclusion, we have identified here that the enteric microbiota of the GI tract resident in treated CD patients is modulated by disease pathogenesis, the local inflammatory microenvironment and the therapeutic treatment strategy. This dysbiosis leads to decreased bacterial diversity, phylogenicity and composition, which, in turn, impair restoration of the gut microbiota to a healthy state. This study provides essential information for evaluating the complexity of CD and its associated treatment options, calling for the development and identification of new clinical treatments that will target normalizing the microbiota in CD patients.

Funding

This work was funded by grants from the National Institutes of Health: National Institute on Drug Abuse [NIDA] [DP1 DA-037997], [3DP1 DA-037997-04S1] and [R01 DA-043253], National Institute of Allergy and Infectious Disease [NIAID] Center for AIDS Research [CFAR] [P30 AI-036219], and Conselho Nacional de Desenvolvimento Científico e Tecnológico [CNPq]; Instituto Nacional de Ciência e Tecnologia de Investigação em Imunologia [INCT] [465434/2014-2] and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior [CAPES], Brazil. L.D.G. [1667212] received a fellowship from CAPES and K.I.C. [311988/2015] received established investigator fellowships from CNPq.

Conflict of Interest

The authors have no relevant conflicts of interest for this study.

Author Contributions

A.C.L.: Conceived and designed the study, data acquisition, analysis and interpretation of the data, drafted the article, final approval of version submitted. L.D.G.: data acquisition and processing, analysis of data [sCD14 and calprotectin], final approval of version submitted. A.T.: data processing, analysis, and interpretation of 16S rRNA gene and taxonomic classification, final approval of version submitted. S.J.P.: contributed reagents, instrumentation and expertise on sequencing, final approval of version submitted. M.Q.M.: contributed reagents, instrumentation and expertise on sequencing, final approval of version submitted. R.P.S.: contributed expertise in data analysis, final approval of version submitted. K.I.C.: Conceived and designed the study, data acquisition, final approval of version submitted. A.D.L.: Conceived and designed the study, data interpretation, drafting and editing of the article, final approval of version submitted.