The primary focus of the current study pertains to bacterial putrefaction pathways which lead to production of compounds that have been reported to be detrimental to gut. Several in vivo/in vitro studies have demonstrated the harmful effects of some of the products of amino acid fermentation (putrefaction) on gut. These include ammonia, H₂S, amine (like putrescine, spermidine, spermine, and cadaverine), cresol, indole, and phenol (Louis et al., 2014; Neis et al., 2015; Yao et al., 2016). For example, such amines have been implicated in DNA damage and tumorigenesis of gastric and colon cancer (Gerner, 2007; Minois et al., 2011; Pegg and Casero, 2011). In addition, N-nitroso compounds (NOCs), which can be formed through nitrosation of amines, have been suggested to exhibit carcinogenic effects (Hughes et al., 2000). The metabolic pathways for NOCs have widely been studied in the context of nitrate assimilation by gut bacteria. Thus, the bacterial putrefaction pathways leading to production of the above mentioned nine compounds have been considered in the present study. It may be noted that ammonia is released during the catabolism of any amino acid in the initial deamination step. Consequently, bacteria capable of fermenting any amino acid would release ammonia during the process. This ammonia can enter the urea cycle and subsequently be excreted under normal condition. However, certain conditions (like high protein and low fiber diet) may lead to production of excess ammonia that can damage colonic cells (Fung et al., 2013). Therefore, mere presence of any ammonia releasing pathways, in most cases, is probably not predictive of the harmful effects of the corresponding bacteria. Thus, in the current study, a set of three ammonia releasing pathways [histidine → glutamate, histidine → tetrahydrofolate (THF) and glutamate → acetate + pyruvate], previously reported to be functional in one of the pathogenic gut bacteria (Fusobacterium), has been selected for further analysis. Although phenylacetic acid and branched chain fatty acid (BCFA) have been suggested to be efficient indicators of amino acid fermentation in the colon (Louis et al., 2014; Yao et al., 2016), we have not considered them since to the best of our knowledge, these compounds have not been reported to have any harmful effect on the gut. The available literature data on the pathways corresponding to the selected above mentioned nine compounds was further surveyed and utilized for prediction of these pathways across bacterial genomes.

Prediction of a pathway in an organism can be performed by mapping the homologs of its constituent enzymes. However, some pathways may contain one or more enzymes that have generic function (like dehydrogenase) which participate in multiple functional pathways. Thus, identification of a particular pathway based on only presence of homologs may lead to false predictions. Therefore, in the current study, in addition to enzyme homology, the genomic proximity of the constituent genes has been utilized for prediction of a pathway. However, it may be noted that the genomic proximity (of the constituent genes) may not be true for all pathways. Therefore, in the current study, the putrefaction pathways leading to production of the selected nine metabolites (as discussed in the previous section) were further filtered to shortlist the following three categories of pathways-

For each of the harmful metabolites, the experimentally validated putrefaction pathways (Supplementary Table S1) that belong to the above three categories are described below.

The methodology used for prediction of the selected 10 putrefaction pathways across bacterial genomes has been described below.

One of the primary objectives of the current study pertains to analysis of gut microbiome for delineating any probable association of the aforementioned microbial (putrefaction) pathways with CRC. Since, the currently available technologies do not allow the identification of all bacteria in a given metagenome at strain level, it was necessary to obtain measures of putrefaction capabilities for bacterial groups at higher taxonomic levels. Thus, based on the proportion of constituent putrefying strains and a confidence value of the corresponding group, a score was evaluated at the levels of phylum and genus. The confidence value was incorporated to assign a higher weightage to the bacterial group having relatively higher representative organisms in the database. For evaluating the confidence score, the number of strains under all bacterial groups was first noted. Based on these counts a percentile value was computed for each of the groups. Subsequently, these percentile values were used to assign ranks (between one and five) to the groups. The final putrefaction score (Pfacs), for a particular putrefaction pathway ‘i’ corresponding to a bacterial group ‘j,’ was calculated using the following equation-

where, S represents the proportion of putrefying organisms of the particular bacterial group and α denotes the confidence value of the corresponding bacterial group.

Thus, the values of the computed ‘Pfacs’ scores ranged between ‘zero’ and ‘five.’ For a particular putrefaction pathway, a bacterial taxon having a higher ‘Pfacs’ would indicate a greater probability of presence of that pathway in the corresponding strains as opposed to a taxon with a lower ‘Pfacs.’

In order to investigate the association between bacterial pathogenicity and putrefaction capability, comprehensive list of known gut commensals and pathogens was collated from ‘Integrated Microbial Genomes and Microbiome’ (IMG/M) (Chen et al., 2017) and literature (Supplementary Table S3). The gut strains affiliated as pathogens and commensals were then mapped to the catalog of putrefactors identified in the current study and the respective distributions of putrefaction pathways were obtained.

In order to compare the putrefaction capabilities of the microbes residing in the gut of CRC patients with that of healthy individuals, the gut microbiome data (16S rRNA sequences) provided in five published studies on CRC was analyzed (Kostic et al., 2012; Wang et al., 2012; Zackular et al., 2014; Burns et al., 2015; Nakatsu et al., 2015). Details on the analyzed datasets are provided in Supplementary Table S4. The ‘sra’ files obtained from the datasets under study were extracted using SRA toolkit 2.3.4 (Leinonen et al., 2011) and the retrieved fastq sequences were subjected to quality filtration. Prinseq-lite (Schmieder and Edwards, 2011) was utilized to obtain the sequences (in ‘FASTA’ format) having an average phred quality score of more than or equal to 25 for further downstream analyses. Subsequently, Naive Bayesian classifier of the Ribosomal Database Project (RDP classifier 2.10) (Wang et al., 2007) was utilized for taxonomic assignment of the sequences at a bootstrap confidence threshold of 80% for phylum, class, order, family and genus levels. In-house scripts were used to generate abundance of each taxon for healthy, adenoma and carcinoma samples. The obtained taxonomic abundances were normalized to estimate the relative abundances of taxa in each sample.

For identification of differentially abundant taxa in healthy, adenoma and carcinoma samples, multivariate analysis was performed on the normalized abundance for each taxon. Studies containing two distinct classes or groups (healthy and carcinoma) were subjected to Welch t-test (Welch, 1938). Kruskal–Wallis test (Kruskal and Wallis, 1952) was performed to identify differential bacterial groups in studies comprising three sets of cohorts (healthy, adenoma, and carcinoma). The genera that appeared as significantly different (in terms of abundance with a p-value ≤ 0.05) were identified as differentially abundant genera in the corresponding cohort. Subsequently, the differentially abundant genera of CRC cohort (corresponding to carcinoma as well as adenoma stages) were compared with those of healthy population, in terms of – (i) abundances of putrefying bacteria and (ii) distribution of putrefaction pathways.

The advanced stage of CRC has been associated with hypoxic (oxygen deficit) tumor micro-environment and subsequent growth of anaerobic pathogenic bacteria (Cummins and Tangney, 2013; Scanlon and Glazer, 2015). In order to gain insights into the proportion of putrefactors that may promote tumor aggravation by contributing to the virulence of the anaerobic pathogens, the differentially abundant taxa in healthy, adenoma and carcinoma were further analyzed from the perspective of their oxygen requirement. For each of the differentially abundant genera in healthy, adenoma and carcinoma cohorts of different studies, data on their oxygen requirement were collected from literature. The differential genera were then classified into five categories, namely, aerobe, anaerobe, microaerophile, facultative anaerobe, and obligate anaerobe. The strains belonging to genera like Streptococcus have been reported to be either obligate or facultative anaerobes. Thus, for such cases, a generic classification of ‘anaerobe’ was assigned. Further, the proportion of putrefactors in each of the above mentioned categories were noted.

The results indicate presence of putrefaction pathways in 1368 (50%) bacteria out of the analyzed 2,738 completely sequenced genomes (Supplementary Data Sheet S1). One or more of the selected putrefaction pathways were observed in bacteria belonging to 14 phyla (Figure 2). The distributions of the selected 10 putrefaction pathways which lead to release of nine compounds (ammonia, putrescine, spermidine, spermine, cresol, indole, phenol, cadaverine, and H₂S) indicate presence of all the 10 pathways in Firmicutes. Considering the cumulative contribution of putrefying strains belonging to a phylum, while Proteobacteria was predicted to have nine pathways (except tyrosine → cresol), Fusobacteria and Bacteroidetes, were predicted to contain eight pathways (except tyrosine → cresol and lysine → cadaverine). Further, it is noteworthy that, although the strains of Firmicutes cumulatively represented all the 10 pathways, the corresponding ‘Pfacs’ were relatively lower (≤1.4). This suggests that, the putrefaction capability of the phylum Firmicutes is limited to selected members. In contrast, the phyla Acidobacteria, Actinobacteria, Bacteroidetes, Fusobacteria, and Proteobacteria were observed to have comparatively higher ‘Pfacs’ of at least half of the maximum value (2.5) in one or more of the pathways. Interestingly, bacteria belonging to five (Firmicutes, Actinobacteria, Bacteroidetes, Fusobacteria, and Proteobacteria) out of six above mentioned phyla have been reported to be common inhabitants of the human gut (Rajilić-Stojanović et al., 2007; Zhang et al., 2015). Organisms belonging to the sixth phylum (Acidobacteria) have mostly been found in soil (Jones et al., 2009; Zhang et al., 2014). However, earlier studies have reported presence of some of its strains in human gut (Stearns et al., 2011). This suggests the possible role of bacterial putrefaction in the gut, where availability of undigested proteins can trigger activation of certain putrefaction pathways in the above mentioned bacterial groups.

Given the observed association of the selected bacterial putrefaction pathways and gut environment, further analysis was performed on the bacterial genera previously known to be associated with gut (Figure 3). The results indicate that, the strains belonging to some of the genera contain diverse combinations of analyzed putrefaction pathways, similar to the trend observed in phyla level analysis. This suggests that these pathways may not have been evolutionary retained, but rather have been acquired by the respective organisms for adaptation to the environment they survive in. Overall, the putrescine production pathway (arginine → putrescine) was observed to have highest occurrence (87.5%) among all other pathways (Figure 3). Putrescine like polyamines have been reported to be involved in functions like growth, cell wall synthesis, cell signaling, biofilm formation in bacteria (Wortham et al., 2007). Thus, the present results indicate the probable role of this pathway in growth and survival of putrescine producing bacteria. The next highest occurring putrefaction pathways (following the putrescine production pathway) corresponded to histidine degradation (histidine → glutamate) and H₂S production (cysteine → H₂S) with 50% and 56% occurrence, respectively. Among others, the occurrence of five pathways for production of THF (from histidine), glutamate (from histidine), indole (from tryptophan), spermidine/spermine (from methionine) and cadaverine (from lysine) were observed to vary between 25% and 35%. The remaining two pathways corresponding to phenol (from tyrosine) and cresol (from tyrosine) production were predicted to be present in limited number of genera (12.5% and 3%, respectively). The genera observed to have higher ‘Pfacs’ (≥2.5) for at least one pathway include Bacillus, Bacteroides, Enterobacter, Escherichia, Klebsiella, Prevotella, Salmonella, Shigella, Vibrio, and Yersinia. In addition, the group of representative strains of Firmicutes phylum, having as much as nine pathways including exclusive presence of cresol production pathway (as mentioned in previous paragraph) was observed to belong to the genus Clostridium. Earlier studies have also suggested the ability of production of cresol through tyrosine fermentation to be unique to Clostridium genus (Selmer and Andrei, 2001). Further, the genera Escherichia and Klebsiella were predicted to harbor seven of the putrefaction pathways. This was followed by Fusobacterium, Enterobacter, Salmonella, and Shigella, which were observed to possess six out of the 10 putrefaction pathways. Thus, the current results indicate that the above mentioned genera have the capability to ferment various amino acids (like histidine, glutamate, arginine, lysine, cysteine, tyrosine, and tryptophan), acquired from undigested proteins that escape energy metabolism and consequently are likely to affect the gut functioning. In addition, since some of the strains of the above mentioned genera have been reported to exert pathogenicity in gut, further analysis of putrefaction pathways in pathogenic and commensal bacteria were carried out.

The analysis of pathogenic and commensal gut bacteria revealed that, all the 21 strains affiliated as pathogens contain one or more putrefaction pathways (Figure 4). The three pathways occurring in relatively higher number of pathogens include putrescine, spermidine/spermine production and histidine degradation (predicted in ∼90%, ∼60% and ∼50%, respectively) (Figure 4A). In contrast, only the pathways for H₂S and putrescine production were predicted in around 30% of the commensals (Figure 4B). Apart from the toxic effect on colonic cells, H₂S has been proposed to aid bacteria in the formation of biofilm as well as benefit the host cell through mediating production of colonic mucus layer (Motta et al., 2015). Thus, the occurrence of H₂S production pathway in commensal strains of Bifidobacterium and Coprococcus may reflect their own survival strategies and/or their beneficial effects on host cell. Most of the pathogenic putrescine producers (61%), belonging to the family Enterobacteriaceae, were found to have the pathway involving agmatinase (arginine → agmatine → putrescine) (Supplementary Figure S1). Earlier studies have also reported the role of this pathway in contamination of food by these groups of organisms (Wunderlichová et al., 2014). On the other hand, all the four commensal putrescine producers were predicted to utilize either ornithine decarboxylase (ODC) (arginine → ornithine → putrescine) or carbamoylputrescine hydrolase (arginine → agmatine →N-carbamoylputrescine → putrescine) pathway (Supplementary Figure S1). These observations suggest that, pathogen and commensal gut bacteria probably prefer different pathways for production of putrescine through arginine fermentation. Another interesting observation pertains to the organization of the genes corresponding to ODC pathway. Among the bacteria producing putrescine through the utilization of this pathway, only the pathogenic strains under Fusobacterium genus were observed to possess a single fused gene (encoding a single protein containing two functional domains, ‘Arginase’ and ‘OKR_DC_1’) (Supplementary Figure S2). In contrast, other organisms (pathogenic strains under Clostridium genus and commensal strains under Faecalibacterium genus) were identified to have these two domains as part of two different proteins encoded by separate genes (Supplementary Figure S2). Gene fusion has been reported earlier to confer a transcriptional advantage for synthesis of the corresponding proteins (Long, 2000). Thus, the functional potential of ODC pathway in Fusobacterium may be higher than others. Another pathway, prevalent in the pathogens, pertains to histidine degradation (leading to release of ammonia), which was predicted in 50% of the pathogenic strains (Figures 4A,C). The results suggest that, organisms under Enterobacteriaceae family (corresponding to the genera Citrobacter, Enterobacter, Klebsiella, Salmonella, and Vibrio) and the genus Bacillus probably utilize formiminoglutamase (EC 3.5.3.8) for catalyzing one of the reactions (N-formimino-L-glutamate →L-glutamate) of the pathway. The remaining pathogenic strains belonging to the genera Fusobacterium and Bacteroides were predicted to utilize an alternate enzyme, namely glutamate formyltransferase (EC 2.1.2.5), for catalyzing the same reaction. These observations indicate that, during evolution different bacterial groups may have acquired different genetic elements for fermenting same amino acid (like arginine and histidine). The remaining seven putrefaction pathways, namely glutamate degradation and production of THF, cadaverine, cresol, indole, phenol and H₂S, were observed to be present in relatively lesser proportion (≤24%) of the pathogens (Figures 4A,C). Earlier studies have implicated the involvement of some of the pathogens like Salmonella Typhimurium, Bacillus cereus, Escherichia coli O157:H7 in contamination and fermentation of meat leading to release of harmful toxins, which has been reported to cause enteric inflammation (Lawrie and Ledward, 2006). Thus, the insights obtained from the current analysis suggest possible role of two of the putrefaction pathways, namely, histidine degradation and spermidine/spermine production (occurring in ∼50% and 60%, respectively, in the analyzed pathogenic strains), in the pathogenicity of gut bacteria.

Given the putrefying capabilities of gut associated pathogenic organisms (as discussed in section “Gut Bacteria and Putrefaction Pathways Associated with Release of Harmful Compounds”) and the potential role of some of the putrefaction products in disease etiology of CRC, a comparative analysis was performed on publicly available gut microbiome data of CRC patients and corresponding healthy controls. The differentially abundant bacterial genera in these cohorts were considered for capturing possible distinctness in the putrefaction capabilities of their gut microbiomes. A relative enrichment of putrefying bacteria was observed in carcinoma microbiome as compared to that in healthy samples (Figure 5). Out of the 10 analyzed putrefaction pathways, the differentially abundant genera in carcinoma and healthy cohorts were observed to represent nine and seven pathways, respectively. However, the percentage of occurrence of these pathways in the carcinoma cohort was relatively higher compared to that in the healthy cohort. For example, although putrescine production pathway was observed to have the highest occurrence in both the cohorts, it was found to be present in 58% (occurring in 11 out of the 19 differentially abundant genera) and 26% (occurring in 4 out of the 15 differentially abundant genera) in carcinoma and healthy cohorts, respectively. The next highest occurring pathway in carcinoma cohort was found to be histidine degradation (31.5%), occurrence of which was seen to be only 13.3% in healthy cohort. Considering the differentially abundant genera in carcinoma samples, the genus Fusobacterium, containing six of the analyzed putrefaction pathways (distributed among its strains), was observed to be differentially abundant in four out the five datasets. The putrefaction capability of Fusobacterium has also been implicated in promoting periodontitis in oral cavity (Flynn et al., 2016). Apart from Fusobacterium, some of the other genera (such as Streptococcus, Candidatus, Escherichia, Shigella, Prevotella, Selenomonas, and Clostridium_XIX), predicted to possess at least three putrefaction pathways, were observed to be significantly higher in carcinoma microbiome. It may be noted that, the completely sequenced strains belonging to the ‘Clostridium_XIX’ group corresponded to those under Fusobacterium nucleatum species (Xing et al., 2011). Consequently, in this analysis, ‘Clostridium_XIX’ has been considered to represent a subset of Fusobacterium genus. In contrast to the carcinoma cohort, only three differentially abundant genera in healthy cohort, namely, Alistipes, Parabacteroides, and Ruminococcus, were predicted to have at least three of the analyzed putrefaction pathways. Among these, Alistipes was observed to have highest number of pathways which included histidine degradation and productions of THF, indole, and phenol. This genus, although generally considered as a commensal, has been implicated in inflammation and bacteremia in patients of gastrointestinal disorders (Fenner et al., 2007; Jiang et al., 2015). Thus, the identified putrefaction pathways in Alistipes may contribute in deleterious consequences under dysbiotic conditions. Apart from putrescine production pathway which was found to be common in Parabacteroides and Ruminococcus, the former was predicted to harbor histidine degradation as well as THF production pathways and the latter was seen to have cadaverine and H₂S production pathways. Organisms belonging to Parabacteroides have been reported to carry resistance genes against antimicrobial drugs and consequently behave as opportunistic pathogens (Boente et al., 2010). Thus, it is possible that the identified putrefaction pathways in Parabacteroides may function only under certain conditions, when the bacteria exhibit pathogenicity. The other two genera (in addition to the above mentioned three genera) which were found to be differentially abundant in healthy cohort, namely, Oscillibacter (containing two pathways) and Faecalibacterium (containing one pathway), were predicted to have putrescine production pathway in common and spermidine/spermine production as an additional pathway in the former. Earlier studies have indicated importance of metabolites like putrescine and H₂S in functions like growth and biofilm formation in bacteria (Wortham et al., 2007; Shatalin et al., 2011). Thus, it is likely that the putrefaction pathways producing these metabolites by the above mentioned four genera (Faecalibacterium, Oscillibacter, Ruminococcus, and Parabacteroides) may be utilized by the bacteria for their own benefit, rather than exerting harmful effects in the healthy cohort. Additionally, three of the above mentioned genera from healthy cohort (Ruminococcus, Faecalibacterium, and Oscillibacter) have been reported to produce butyrate, which is known to participate in multiple functions (including anti-inflammatory role) that are beneficial to gut health (Anand et al., 2016; Rivière et al., 2016). Overall, the observation of relatively higher percentage of histidine degradation and putrescine production pathways in carcinoma cohort, combined with the results of the analysis on gut pathogens (as discussed in the previous section) indicate possible role of these pathways in disease pathogenesis of CRC.

The results of the microbiome analysis indicated a relative enrichment of putrefying bacteria in gut microbiome of CRC cohort at later stage of the disease as compared to that in the early adenoma stage (Figure 5). Advanced stage of the disease has been shown earlier to be correlated with hypoxic (oxygen deficit) microenvironment and consequent growth of anaerobic pathogenic bacteria (Scanlon and Glazer, 2015). The current results suggest that, the anaerobic bacterial genera (especially obligate anaerobes and microaerophiles) which are differentially abundant in carcinoma microbiome are mostly putrefactors, as opposed to that of healthy population in all datasets examined in this study (Figure 6). Interestingly, while the differentially abundant anaerobic genera corresponding to one of the adenoma cohort were predicted to have only histidine degradation and production of putrescine, cadaverine, and H₂S, the other dataset lacked any putrefaction pathway. The decreased number of putrefaction pathways in adenoma (as compared to the healthy samples), as observed from the analysis based on the limited number of available datasets, would require further validation on a larger cohort. Such studies are likely to provide better insights on the cause-consequence relationship between hypoxic tumor micro-environment and putrefaction by anaerobic bacteria.