BRAFV600E mutation impinges on gut microbial markers defining novel biomarkers for serrated colorectal cancer effective therapies

Colorectal cancer (CRC) harboring BRAFV600E mutation exhibits low response to conventional therapy and poorest prognosis. Due to the emerging correlation between gut microbiota and CRC carcinogenesis, we investigated in serrated BRAFV600E cases the existence of a peculiar fecal microbial fingerprint and specific bacterial markers, which might represent a tool for the development of more effective clinical strategies. By injecting human CRC stem-like cells isolated from BRAFV600E patients in immunocompromised mice, we described a new xenogeneic model of this subtype of CRC. By performing bacterial 16S rRNA sequencing, the fecal microbiota profile was then investigated either in CRC-carrying mice or in a cohort of human CRC subjects. The microbial communities’ functional profile was also predicted. Data were compared with Mann-Whitney U, Welch’s t-test for unequal variances and Kruskal-Wallis test with Benjamini–Hochberg false discovery rate (FDR) correction, extracted as potential BRAF class biomarkers and selected as model features. The obtained mean test prediction scores were subjected to Receiver Operating characteristic (ROC) analysis. To discriminate the BRAF status, a Random Forest classifier (RF) was employed. A specific microbial signature distinctive for BRAF status emerged, being the BRAF-mutated cases closer to healthy controls than BRAF wild-type counterpart. In agreement, a considerable score of correlation was also pointed out between bacteria abundance from BRAF-mutated cases and the level of markers distinctive of BRAFV600E pathway, including those involved in inflammation, innate immune response and epithelial-mesenchymal transition. We provide evidence that two candidate bacterial markers, Prevotella enoeca and Ruthenibacterium lactatiformans, more abundant in BRAFV600E and BRAF wild-type subjects respectively, emerged as single factors with the best performance in distinguishing BRAF status (AUROC = 0.72 and 0.74, respectively, 95% confidence interval). Furthermore, the combination of the 10 differentially represented microorganisms between the two groups improved performance in discriminating serrated CRC driven by BRAF mutation from BRAF wild-type CRC cases (AUROC = 0.85, 95% confidence interval, 0.69–1.01). Overall, our results suggest that BRAFV600E mutation itself drives a distinctive gut microbiota signature and provide predictive CRC-associated bacterial biomarkers able to discriminate BRAF status in CRC patients and, thus, useful to devise non-invasive patient-selective diagnostic strategies and patient-tailored optimized therapies.


Background
Colorectal cancer (CRC) is the third most common cancer and the second leading cause of cancer-related deaths in developed countries [1]. Although some risk factors are well outlined [2] and many comprehensive studies have established the molecular criteria for CRC's classification [1,3,4], the regulatory mechanisms of this tumor remain largely unrevealed.
The inherent extensive heterogeneity of CRC encompasses as many different histological and molecular bases associated with diverse clinical-pathological features. The adenoma-carcinoma sequence proposed by Vogelstein, in which the pre-neoplastic lesions accumulate stepwise molecular and morphological changes leading to cancer [5], has been long considered a unique model of CRC cancerogenesis until the description of the "serrated pathway" [6,7]. This pathway arises from serrated polyps, once considered benign, including hyperplastic polyps (HPs), sessile serrated adenomas/polyps (SSAs/ Ps) and traditional serrated adenomas (TSAs) [8,9]. Among these, HPs are the most frequent (80-90%) and usually display a low malignant potential, while TSAs account for less than 1% of serrated CRCs [10]. SSAs/SSPs, which account for the 5-20% of all the serrated lesions, display peculiar molecular features, including CpG island methylator phenotype (CIMP-H) and BRAF mutation [9,11]. CIMP-H results in transcriptional repression of p16 INK4a and MLH1 genes, whereas BRAF mutations, often consisting in the activating V600E substitution, causes aberrant activation of the MAPK signaling [12,13].
From a clinical point of view, BRAF-mutated CRC behave as a distinct subset compared to conventional adenomas, typically exhibiting lower response to conventional therapy, elevated invasiveness and the poorest clinical outcome, suggesting that initiation through this pathway might predict the aggressiveness of CRC [6,13].
It is nowadays recognized that a small subpopulation of self-renewing CRC cancer stem cells (CCSCs) drive the initiation and progression of CRC, the metastatic colonization and the disease relapse after therapy [14,15]. We have also very recently reported in our human CCSCs-based in vivo model that stemness underpins all the stages of CRC development, identifying CCSCs as a constitutive component in the establishment and dissemination of this tumor [16].
Disruption of the gut microbial community's homeostasis, i.e. dysbiosis, also plays an important role in the initiation and maintenance of CRC as well as in the response to therapies [17][18][19][20][21]. In recent years a plethora of studies revealed the existence of a human CRCspecific microbial signature, underlining the pathogenetic role of some microorganisms, such as Streptococcus gallolyticus, Bacteroides fragilis, Enterococcus faecalis, Fusobacterium spp., Fusobacterium nucleatum and Escherichia coli. Yet, the expansion of these "bad" bacterial species with the ensuing depletion of "good microbes" has been shown to be implicated in DNA damage, uncontrolled cell growth, inflammatory signaling pathways and, thus, in CRC promotion and progression [22][23][24]. Additionally, it has been speculated that gut microbiota shift might be related to different early precursors of CRC, but whether specific microbial profiles could discriminate between conventional and BRAF-mutated CRC still remains under-investigated [25][26][27].
Here, we report either in our CCSCs-based in vivo model or in CRC patients that BRAF V600E mutation can itself sustain a typical microbiota profile, thus identifying new putative tumor-associated bacterial markers for patient-tailored diagnostic and therapeutic purposes.

Methods
Primary CCSCs culture and analysis, immunochemistry, reagents, targeted and sanger sequencing, qPCR and methylation-specific PCR, bacterial DNA extraction are described in detail in the Supplementary Methods.

In vivo studies
All the animal experimental procedures were performed in compliance with the Guidelines for the Care and Use of Laboratory Animals. Animal protocols have been approved by the Ministry of Health (PR/15-297/2019-PR). In order to minimize any suffering of the animals, anesthesia and analgesics were used when appropriate. Orthotopical PDX was determined by injecting CCSCs into the wall submucosa of the ascending colon of Scid/ bg mice (Charles River Lab) [16]. Quantification of tumor growth was performed from ventral and dorsal views by In vivo Lumina (Xenogen, PerkinElmer Inc) [16,28]. Upon sacrifice at different time points according to the cell line originally injected, tissue samples from CCSCs-derived primary colon tumors and from mesenteric lymph nodes as well as lung, liver, spleen and brain metastases were collected and processed as previously reported [16,[28][29][30]. Spontaneous metastatic pulmonary lesions formation was performed by injecting 3 × 10 5 luc-CCSCs cells into the lateral tail-vein of Scid/ bg mice [16]. For the gut microbiota analysis, fresh fecal samples were collected from the cages at early-stage (7 days post-transplantation; DPT) and next to the median end-stage of the disease (late and end-stages, 37-43 DPT). Samples were immediately frozen and stored at − 80°C until DNA extraction.

Clinical patient's features
Patients with a confirmed diagnosis of CRC (33 cases) and healthy subjects (13 subjects) included in the control group were enrolled in this study at IRCCS "Casa Sollievo della Sofferenza" Hospital, under the Ethical committee approvals number N.175/CE and N.94/CE. All the subjects agreed to participate according to the ethical guidelines of the 2013 Declaration of Helsinki and signed an informed consent. Eligible subjects were 45-90 years old who did not undergo radio/chemotherapy or pharmacological/long-term antibiotic treatments. All the fecal samples were collected at the moment of diagnosis before any surgery or adjuvant treatment. Human CRC tissues were classified according to established staging system (AJCC and TNM) and diagnosis was confirmed by the pathologist. Histological data together with localization of colonic lesions are reported in Table 1. Fresh stool samples were collected by each participant in containers with DNA stabilization buffer (Canvax Biotech) and stored at RT for few days until DNA extraction. Information regarding subject's variables (i.e. age, gender and BMI) as well as dietary, lifestyle and smoking habits were assessed the same day of the stool sample collection.
Sequencing and analysis of 16S rRNA Library preparation and sequencing was performed with Illumina 16S Metagenomic Sequencing Library Preparation kit (Illumina Inc) accordingly to manufacture's instruction. Briefly, the V3-V4 hypervariable region of the bacterial 16S ribosomal RNA was amplified with primers selected from [31], containing appropriate Illumina overhang adapter sequences. Amplicons were further amplified to attach dual Illumina indices (Nextera XT Index Kit, Illumina Inc) and PCR products again purified. The pooled libraries were paired-end sequenced (2 × 300 cycles) in Illumina MiSeq platform (Illumina Inc). Sequences were demultiplexed and FASTQ files were generated. Raw sequencing data were then trimmed for quality and Illumina adapters were removed. After excluding host reads, reads were aligned and mapped to the NCBI taxonomy database of bacterial and archaeal 16S rRNA sequences, using Kraken2 software [32]. Rarefaction curves were generated by randomly subsampling the OTU tables to a depth of 61,888 and 12,609 sequences (for mouse and human samples, respectively) per sample 10 times before computing the observed species. Several a-diversity metrics, including Chao1 and Shannon index, the Simpson reciprocal and the observed genus and species, were computed. To assess b-diversity in xenogeneic CRCs, jackknifed Bray-Curtis distances (10 sub-samplings at a depth of 61,888 sequences per sample) was computed and the matrices visualized in PCoA plot. Core diversity analysis was performed on the OTU tables, including a-and b-diversity as well as taxonomic summary, as implemented in QIIME software [33]. To account for library size, OTU profiles were converted to relative abundances and then filtered for species confidently detectable. Specifically, microbial species that did not exceed a maximum abundance of 1 × 10 − 3 in at least one sample were excluded, together with the fraction of unmapped metagenomic reads. Hierarchical cluster analysis and visualization of the relative abundances were performed with Partek Genomics Suite v.6.6 software (Partek Inc.).

Functional profile prediction
Based on the 16S rRNA sequences, microbial communities' functional composition was predicted using PICRUSt software [34]. All sequences from each sample were searched against the Greengenes (gg_13_5) at the 97% identity (closed OTU picking method). OTU tables were normalized by dividing each OTU by the known/ predicted 16S rRNA gene copy number abundance and the prediction of the metagenome functional content was classified according to the Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology. The predicted metagenome BIOM table was analyzed and visualized using the Statistical Analysis of Taxonomic and Functional Profiles (STAMP) v. 2.1.3 software [35].

Statistical analyses
For in vitro studies, statistical tests were performed using GraphPad Prism v7.0 software and ANOVA tests according to the variance and distribution of data. Differential gene expression was assessed by the implementation of the ANOVA test available in Partek Genomic Suite 6.6 with FDR < 0.05. P-values< 0.05 were considered significant. Results from 16S rRNA gene sequences between or among groups were compared with nonparametric Mann-Whitney U, Student's t-test (or Welch's ttest for unequal variances) and Kruskal-Wallis test with Benjamini-Hochberg false discovery rate (FDR) correction for multiple comparisons at each level separately. FDR (q-value) < 0.10 was considered significant. Pearson correlation coefficient was used to assess association between gene expression and bacterial abundance at genus level in BRAF V600E vs healthy and BRAF wt mice (43DPT) (q-values < 0.1). P < 0.05 was visualized. To discriminate the BRAF status in CRC patients, a Random Forest classifier (RF) was used [36]. The number of decision trees was set to 500. Significantly different bacterial species' abundance between BRAF V600E and BRAF wt CRC cases, as emerged from Mann-Whitney test (P < 0.05), was extracted as potential BRAF class biomarkers and selected as model features. From 8-fold crossvalidation, mean test prediction scores were obtained and subjected to Receiver Operating characteristic (ROC) analysis. ROC curve was used to evaluate the diagnostic value of bacterial candidates in distinguishing BRAF-mutated from BRAF wt cases. A Fisher's exact test was also performed. RF was firstly applied on the whole set of selected features and then on each one of them to identify the most important ones, ranked by areas under ROC (AUROC) metric. The best cut-off values were determined by ROC analyses that maximized the Youden index (J = Sensitivity + Specificity -1 [31]. All statistical analyses were conducted using R software.

Molecular and pathophysiological features of BRAFmutated and BRAF wild-type CRC stem-like cells
By orthotopical injections of CRC stem-like cells (CCSCs) we have previously reported an in vivo model, which faithfully recapitulates human CRC features [16].
Here we characterized and confirmed in vitro and in vivo the phenotypic hallmarks of the three CCSCs lines isolated from CRC patient's tissue, either associated to serrated pathway (BRAF V600E CCSCs) or conventional CRC (BRAF wt CCSCs) [16]. To verify as to whether BRAF V600E CCSCs reflects the key characteristics of serrated CRC, we delineated the presence of BRAF mutation and the unchanged form of KRAS and NRAS and the CIMP-H phenotype [37]. As shown in Fig. 1a, BRAF V600E CCSCs are characterized by the presence of BRAF point mutation T1796A in exon 15 codon 599, reflecting a valine to a glutamic acid amino acid shift (V600E). Consistently, a methylated status of the promoter regions of p16 INK4a , MutL homolog1 (hMLH1), MGMT, MINT1 and MINT2 genes was shown (Fig. 1b).
To pinpoint the inherent "serrated" signature of BRAF V600E CCSCs, we next compared the transcriptional profile of CCSC lines to each other ( Fig. 1c-e). As expected, hierarchical clustering analysis based on the global gene expression clearly segregated the serrated BRAF V600E CCSCs from the other two CCSCs lines, which are quite similar (Fig. 1c). Consistently, among the genes preferentially expressed in BRAF V600E CCSCs, many are reported to be up-regulated in the BRAF V600E CRC pathway as well as to control matrix remodeling and epithelial-to-mesenchymal transition, inflammation and innate immunity, cell migration and invasion and transforming growth factor-b (Fig. 1d) [6,38]. Meanwhile, transcriptomic fingerprint of BRAF V600E CCSCs was identified by low levels of CDX2 and Wnt target genes [39]. The epithelial cell adhesion molecule EpCAM, CD44 and VIL1 protein expression was confirmed to be a hallmark feature of CCSCs with a "serrated" phenotype, as compared to the BRAF wt counterpart, whereas CDX2 marker was highlighted exclusively in the latter (Fig. 1e-f). B-catenin protein was mainly localized in the plasma membrane in BRAF V600E CCSCs cells (Fig. 1e) [16].
Strikingly, following orthotopical delivery of BRAF V600E CCSCs, tumors with histological architectures, that closely resemble the human serrated pathway, were detected, whereas the typical CRC morphology was identified in lesions from BRAF wt CCSCs-bearing mice ( Fig. 2a-b). Though all CCSCs gave rise to distant spontaneous metastatic lesions [16], only mice infused intravenously with BRAF V600E CCSCs exhibited pulmonary metastasis within 60 days after injection (Fig. 2c).
These findings lent to the conclusion that BRAF V600E CCSCs recapitulates the main features of serrated CRC, being a faithful model for in vivo studies of serrated tumorigenesis.

Microbiota profiles and functional composition of BRAFmutated and BRAF wt xenogeneic CRCs
To explore associations between gut microbiota composition and BRAF V600E CRC, we first exploited our xenogeneic CRC model, in terms of global alteration in the microbiota profiles of BRAF V600E and BRAF wt CRC-bearing mice vs control. Bacterial flora was analyzed either at early stage or next to the median endstage of the disease (Figs. 3 and 4, Supplementary Table S1-S2). Species richness was found higher in controls and BRAF V600E xenogeneic CRCs than in BRAF wt tumors (Fig. 3a). As indicated by the a-diversity Shannon index, xenogeneic BRAF-mutated CRC and controls displayed a higher microbial community diversity than BRAF wt CRC-carrier group, both at genus and species level (P < 0.0001 and P = 0.004, respectively, Kruskal-Wallis test) (Fig. 3b, top), whereas genus and species richness did not reach significance among the groups (Fig. 3b, middle and bottom). Next, we explored the signature of the gut microbiota in the CRC-carrier groups vs control observing in the former a remarkable clustering within the progression of the disease over time (Fig.  3c). Firmicutes and Bacteroidetes were found the most represented phyla in all groups, together with Proteobacteria, Tenericutes and Verrucomicrobia (Supplementary  Table S1-S2). Yet, the key marker of gut dysbiosis [40] Firmicutes/Bacteroidetes ratio, tended to be comparable between xenogeneic BRAF V600E CRCs and control, whereas was significantly lower in mice bearing BRAF wt tumors (Fig. 3d). Unsupervised hierarchical analysis at the family level (Fig. 3e) revealed a characteristic "healthy" microbial signature in controls segregated from that one of CRC-bearing mice, being BRAF V600E CRCs closer to tumor-free mice than BRAF wt tumors. Notably, an enrichment in several butyrate-producing Fig. 1 Phenotypic fingerprint of BRAF V600E and BRAF wt CCSCs. a Automatic sequencing electrogram showing a minor "A" peak at T1796 denoting a CCSCs' population retaining such a "T1796A" mutation with a T to A mutation in codon 599 (BRAF V600E CCSCs; arrowhead). b MGMT, p16 INK4a , MINT1, MINT2 and hMLH1 methylation was evaluated in BRAF V600E and BRAF wt CCSCs by using primers for methylated (M) and unmethylated (U) alleles of bisulfite-treated DNA. Normal and cancer lung tissues as positive controls. c Heat map of one-way hierarchical clustering of 33 differentially expressed genes in BRAF-mutated vs. BRAF wt CCSCs revealing a typical CRC serrated signature for the former as compared to the latter. A dual-color code represents genes up-(red) and down-regulated (blue), respectively. d Differentially enriched genes associated with cellular migration and invasiveness, matrix degradation and epithelial phenotype in BRAF V600E vs BRAF wt CCSCs, as confirmed by qPCR. ***P < 0.001, **P < 0.01, Mann-Whitney test. e By means of confocal imaging, widespread positivity for CD44, EpCAM and Villin1 markers and weak signal for CDX2 in BRAF V600E CCSCs was shown. Positive nuclear b-catenin staining was retrieved exclusively in BRAF wt CCSCs. Insets: higher magnifications. Scale bars, 50um, 25um, 20um and 10um. Quantification of each marker is shown in f. ***P < 0.001, one-way ANOVA Tukey's multiple comparison test. Data are mean ± SEM bacteria such as Clostridiales, Eubacteriaceae, Lachnospiraceae, Proteobacteria, Firmicutes and Streptococcaceae was observed in mice carrying BRAF V600E CRC as well as in control.
Concerning the shifts in the microbiota profile shared by the two BRAF wt CRC-carrier groups vs controls at 7 DPT, lower abundances of few genera either belonging to Firmicutes or to Bacteroidetes were detected (Fig. 4c and Supplementary Table S1). Unlike xenogeneic BRAF V600E CRC, at 43DPT, the relative bacterial abundance at the phylum and genus level seemed to be clearly different from that one of control (Fig. 4b,   When the microbial communities' functional composition was compared between CRC-bearing mice and control ( Fig. 4e and Supplementary Table S3), according to the level 2 KEGG module the relative abundance of immune system diseases (FDR = 0.012) category together with carbohydrate metabolism (FDR = 0.015) [41] emerged significantly higher in xenogeneic BRAF-mutated CRCs, whereas a remarkable enrichment of general metabolic functions was observed in BRAF wt CRCs.
Consistently, when comparing the two CRC-carrier groups, cell motility (FDR = 0.05) and transcription (FDR = 0.03) categories were significantly enriched in BRAF V600E CRCs, whereas metabolic functions were overrepresented in BRAF wt counterpart.
All of these data delineate a microbial signature associated to xenogeneic BRAF V600E CRC, reminiscent of that one of control.

Microbial taxa's abundance correlates with gene expression in BRAF V600E xenogeneic CRC
We then looked for a potential association between microbiota composition and the transcriptomic hallmarks of the BRAF V600E serrated pathway [6]. For this purpose, the relative abundance of bacterial genera Fig. 4 BRAF V600E CRC exhibits specific bacterial markers and a typical functional microbiota composition. Gut microbiota relative abundance (% similarity) at the phylum (a-b) and genus level (c-d) in gut microbiota of mice carrying BRAF V600E and BRAF wt CRC and controls at 7 DPT (a-c) and 43 DPT (b-d). e Significative relative abundance of predicted function (mean proportions %) and different abundance of predicted function (difference in mean proportion %) for specific KEGG modules level 2 in pairwise comparisons between controls, BRAF V600E (left; top) and BRAF wt CRCs (right) or between the two CRC groups (left; bottom). FDR-adjusted p value (q value) from Mann-Whitney test are shown associated with xenogeneic BRAF V600E CRC and the expression level of markers distinctive of BRAF V600E CCSCs (Fig. 1e) were correlated with each other. A considerable score of positive correlation was identified between genera along with Firmicutes phylum, i.e. Oscillibacter, Desulfallas, Anaerostipes and Ethanoligenens, together with Akkermansia, and genes involved in the BRAF V600E pathway, inflammation and innate immunity and epithelial-to-mesenchymal transition (Fig. 5a). Yet, a negative correlation trend was shown with almost all the Wnt target genes, CDX2 and genes involved in the TGFb pathway. Remarkably, a specular correlation trend was shown for genera along with Bacteroidetes phylum, such as Muribaculum and genera belonging to Bacteriodales, Muribaculaceae and Sphingobacteriaceae.
Data here suggest the existence of a bidirectional communication, involving inflammation, invasion and innate immune signaling, between the BRAF V600E lesion and the gut microbiota.

Gut microbiota fingerprint in serrated BRAF V600E and BRAF wt CRC patients
We next investigated the microbiota composition in a cohort of CRC patients who did not undergo any type of treatment (8 BRAF V600E and 25 BRAF wt CRC) and healthy controls [13]. Early-stage stage I BRAF wt CRCs were excluded (Fig. 6a and Table 1) [41]. As shown in Fig.  6b, no significant differences were observed between CRC cases and controls in terms of age or the body mass index (BMI). Variants in BRAF were confirmed c. 1799 T > A mutation (p.V600E) ( Table 1) and, as expected, a mutually exclusive missense mutation in KRAS and NRAS genes was found. When bacterial community properties were analyzed, the highest level of species richness was reported in BRAF-mutated and control groups vs BRAF wt cases (Fig. 6c). A-diversity and community richness were significantly different among the groups at genus level (P = 0.02 and P = 0.016, respectively, Kruskal-Wallis test) and, down at species level, displayed the lowest expression in BRAF wt cases (P = 0.031) (Fig. 6d).
When the relative abundance of functional category of CRC cases was compared to healthy subjects, according to the level 2 and 3 KEGG modules, samples from BRAF V600E CRC were confirmed to be closer to controls than BRAF wt CRC ones (Fig. 7e and Supplementary   Table S5), perfectly matching data in Fig. 4e. When comparing the abundance of predictive function between BRAF V600E and BRAF wt CRCs, translation (P = 0.02, Fig. 7 Identification of bacterial markers discriminating BRAF status in CRC patients. a-b Relative abundance of bacterial phyla (a) and genera (b) in fecal microbiota of healthy, BRAF V600E and BRAF wt subjects. c Heatmap of one-way hierarchical clustering of differentially represented species among the three cohorts (q-values < 0.10 from Mann-Whitney test). A dual-color code counts for species up-(red) and down-represented (blue), respectively. d Differences in the relative abundances of Hungateiclostridium saccincola (Hs), Prevotella enoeca (Pe), Sutterella megalospaeroides (Sum), Victivallales bacterium CCUG44730 (Vb), Prevotella dentalis (Pd) and Stenotrophomonas maltophila (Stm) (top) and Bacteroides dorei (Bd), Bacteroides ovatus (Bo), Lachnoclostridium phocaeense (Lp) and Ruthenibacterium lactatiformans (Rl) (bottom) markers in BRAF-mutated vs. BRAF wt CRC counterpart. P values from Mann-Whitney test are shown. e Differences in the relative abundance of predicted function for specific KEGG modules (level 2) in pairwise comparisons between healthy and BRAF V600E (top) or BRAF wt subjects (middle) and between BRAF V600E and BRAF wt CRC cases (bottom). q values are from Mann-Whitney test. f-g Receiver operating characteristic (ROC) curves for the single metagenomic classifiers Pe or Rl (f) and for the combination of the 10 markers Hs, Pe, Sum, Vb, Pd, Stm, Bd, Bo, Lp and Rl (g) in discriminating BRAF-mutated from BRAF wt CRC cases FDR = 0.13) and cell growth and death (P = 0.03, FDR = 0.13) categories were significantly enriched in the former, whereas metabolic functions were overrepresented in the latter (Fig. 7e). Several functions of the genetic information processing category, including mismatch repair, ribosome and aminoacyl-tRNA biosynthesis were significantly highly expressed in BRAF V600E vs BRAF wt cases, in which several functions of the metabolism, transport and catabolism categories were found, such as starch and sucrose, amino and nucleotide sugar metabolism and pentose phosphate pathway (Supplementary Table S5).
All of these data confirm that a distinctive microbiota's fingerprint can be distinguished between serrated BRAF V600E and BRAF wt CRC's patients, with the former strongly resembling healthy subjects.

Potential predictive biomarkers of BRAF status in CRC patients
We finally tested the predictive potential of the bacterial markers differentially represented in the two CRC groups to discriminate between BRAF-mutated and BRAF wt patients [22,41,42]. Among the 10 candidate species detected (Fig. 7d), Rl and Pe emerged as single factor with the best performance in discriminating BRAF status, as quantified by the area under the receiver operating characteristic (ROC) curve (AUC) of 0.74 and 0.72 (95% confidence interval, 0.53-0.95 and 0.51-0.93, respectively) (Fig. 7e). Performing ROC analysis at the best cut-off value that maximized the sum of sensitivity and specificity, Pe discriminated CRC patients based on their BRAF status with a sensitivity of 73%, specificity of 87.5%, negative predictive value (NPV) of 64% and positive predictive value (PPV) of 92% (P-value = 0.66, Fisher's Exact Test, 95% confidence interval), while Rl showed a sensitivity of 47%, specificity of 100%, negative predictive value (NPV) of 50% and positive predictive value (PPV) of 100% (P-value = 0.62, Fisher's Exact Test, 95% confidence interval). Strikingly, as depicted by the AUROC of 0.85 (95% confidence interval, 0.69-1.01) in Fig. 7f, the combination of all the 10 fecal markers reached a better performance in distinguish BRAF V600E subjects, with a sensitivity of 73.3%, specificity of 87.5%, NPV of 63.6% and PPV of 91.7% (P-value = 0.026, Fisher's Exact Test, 95% confidence interval).
Findings so far demonstrate that 10 candidate bacterial markers can discriminate BRAF status in CRC's patients, representing new opportunities for the improvement of non-invasive identification and diagnosis of BRAF V600E cases.

Discussion
In this work, we provided the unprecedent findings that BRAF V600E CRC subjects, who did not undergo any type of treatment, might be discriminated from the other CRC's cases in terms of their microbial composition, being closer to healthy condition than BRAF wild-type cases. These findings were observed in BRAF V600E CRCbearing mice (Figs. 3 and 4 and Supplementary Table  S2-S3), and, most important, confirmed in BRAF V600E CRC patients (Figs. 6 and 7 and Supplementary Table  S4-S5), pointing out that our in vivo model of serrated CRC recapitulates the main features of human disease.
We first analyzed gut microbial profile in terms of diversity and richness in CRC cases versus healthy controls and, in agreement with previous studies [43,44], we found that BRAF wt cases, whether they were mice or patients, displayed lower a-diversity and richness (Fig.  3 and 6 a-b and c-d). Otherwise, these differences were not observed in BRAF-mutated subjects, which almost behave the same way as controls also in terms of Firmicutes/Bacteroidetes ratio, whose deviation is considered a hallmark of gut dysbiosis ( Fig. 3 and 6d and e) [40]. Indeed, the down-representation of Firmicutes phylum observed in mice carrying BRAF wt CRCs, might also reflect the depletion of many "good" butyrate-producing bacterial families, such as Clostridiales, Eubacteriaceae, Lachnospiraceae, and Streptococcaceae, characterized by anti-inflammatory, immunoregulatory and metabolic functions [45,46]. Importantly, as regards the molecular and functional microbiota composition over time, although numerous differences were detected at the beginning of the disease in both of the two groups of CRC-carrying mice versus control, only BRAF V600E CRCs resembled healthy mice at the end-stage of the disease ( Fig. 4 and Supplementary Table S2). Although the healthy status mirrored by BRAF V600E is yet to be elucidated, our hypothesis is that the microbiota shift of BRAF-mutant CRC might be strongly related to the peculiar molecular profile of this tumor [6,11,38,39]. Yet, one of the most relevant aspect emerging from the delineation of BRAF V600E CRC's microbiota profile, is likely found in the intriguing demonstration that several correlations do exist between bacterial genera abundance ad genes typically involved in the BRAF V600E pathway (Fig.  5a). As expected, a positive trend of correlation for genes driving inflammation, innate immunity and invasion processes, mostly up-regulated in the BRAF-mutated pathway, was shown, whereas the typically downregulated CDX2 and Wnt target genes, displayed a negative trend [6,11,38,39]. Furthermore, when comparing microbiota functional composition of BRAF-mutated versus wild-type counterpart a decrease in microbiotaassociated metabolic functions was predicted (Figs. 4 and 7), thus confirming the capability of gut microbes to affect tumor through their metabolic functions, beside their impact on host immune and inflammatory responses [44]. These data not only suggested the existence of a bidirectional crosstalk whereby tumor impinge on gut microbiota and the microbiota influence tumor progression but also that a distinctive microbial fingerprint might be sustained by BRAF V600E mutation itself. It is increasingly evident that microbial equilibrium plays a fundamental role in human health, since dysbiosis can contribute to or even initiate CRC development, through several mechanisms including triggering of a chronic inflammatory state, production of reactive oxygen species, genotoxins and carcinogenic compounds, interference with host immunity and metabolism [47,48]. It should be noted, however, that the vast majority of investigation, addressed so far, concerned the role of gut microbiota in conventional CRC, whereas the few reports about the serrated-CRC mainly focused on the presence of F. nucleatum.
We demonstrated that, both in mice and in humans, BRAF-mutated CRC is characterized by a gut microbiota which, compared to conventional CRC, is more resembling to but still remains different from that of healthy subjects. This result was supported by previous findings by Peters et al., [26] in which conventional adenomas (precursor lesions of conventional CRC) but not SSAs (precursor lesions of the serrated pathway) were reported to display lower bacterial communities' richness and diversity and a drop in butyrate-producing bacteria as compared to controls [26]. The reason why serrated BRAF mut CRC is associated to a more eubiotic condition is still to be clarified. One hypothesis is that gut dysbiosis may play a role only in the development of conventional CRC [25,26], but not in the BRAFmutated serrated one), which is also genetically, epigenetically, and molecularly different from the former. Nevertheless, we observed a positive correlation between bacterial genera found in BRAF mut microbiota and genes involved in inflammation, immunity and invasion processes typically expressed in the BRAF-mutated pathway. Another possible speculation is that, despite a microbiota composition generally resembling to that of healthy status, few or even single microorganisms (e.g. F. nucleatum) in the gut of BRAF mut CRC carriers might be sufficient to drive that specific carcinogenetic pathway.
Focusing on the microbiota composition of CRC patients enrolled in our study, an increased abundance of Fusobacteria in either BRAF-mutated or wild-type cases compared with controls emerged [27,49]. Likewise, among the bacteria distinctive of each CRC group down at the species level, Prevotella enoeca and Prevotella dentalis were significantly enriched in patients harboring BRAF V600E mutation (Fig. 7c-d), likely reflecting their massive presence in the biofilm lining the colonic mucosa [50]. Strikingly, as single factor, Prevotella enoeca together with Ruthenibacterium lactatiforman, overrepresented in BRAF wt cases, emerged as the species best discriminating BRAF status in CRC patients, therefore putative candidate non-invasive biomarkers (Fig. 7f). Furthermore, when considering the combination of all the 10 bacterial species differentially represented between the two CRC groups a best performance as a biomarker signature distinguishing BRAF V600E from BRAF wt cases was reached, thus identifying potential diagnostic fecal biomarkers for CRC patients (Fig. 7g). Thus, our work opens new and exciting possibilities for studying the biological underpinnings of the serrated BRAF V600E CRCs and, most important, for the development of innovative diagnostic and therapeutic strategies for the cure of this deadly tumor.

Conclusions
In the present study, we provide the unprecedented findings, observed in xenogeneic BRAF V600E CRC and, most important, confirmed in patients harbouring BRAF mutation, that a distinctive microbiota profile could distinguish BRAF-mutated cases among CRCs. BRAF V600E mutation drives itself a distinctive gut microbiota fingerprint in CRC, suggesting the existence of a bidirectional Tumor-Microbiota-Tumor connection. We identify a bacterial marker signature discriminating BRAF status in CRC patients, thus acting as reliable novel non-invasive clinical biomarkers for patient-tailored diagnostic and therapeutic applications.