Peptidomic and proteomic analysis of stool for diagnosing IBD and deciphering disease pathogenesis

Introduction

The incidence of inflammatory bowel diseases (IBD) is increasing worldwide [1], [2], [3]. At onset, the symptoms of the two main clinical IBD subtypes, Crohn’s disease (CD) and ulcerative colitis (UC), are often common to functional disorders such as irritable bowel syndrome (IBS), thus causing delay in diagnosis. Intestinal inflammation with flares and remissions is a hallmark of IBD, with different pro-inflammatory cells that release inflammatory cytokines and chemokines abundantly present within the intestinal mucosa [4], [5], [6]. The sensitivities and specificities of blood inflammatory biomarkers, mainly C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), white blood cells and platelet counts are very limited [7]. Faecal inflammatory biomarkers, mainly calprotectin (fCal), are more reliable indices of the intestinal inflammation typical of IBD [8], [9], [10], [11]. fCal is currently recommended for distinguishing between IBD from IBS and also for IBD monitoring [12], [13], [14], [15], [16]. However, neither the sensitivity nor specificity of fCal exceed 80% for IBD diagnosis [17], with further reductions in reliability in the presence of mild forms of the disease [18]. Moreover, none of the blood and faecal inflammatory biomarkers studied so far allow CD to be distinguished from UC. In this setting, ASCA and pANCA determinations in blood have sensitivities that do not exceed 65% and specificities of only around 80% [19].

The approaches reported in the literature in the search for new biomarkers for IBD diagnosis include microRNA expression profiling of circulating blood mononuclear cells, metabolomic profiling, microbiota characterisation as well as proteomic studies focusing on the search for differentially expressed proteins in tissue samples [20], [21], [22]. Among the novel biomarkers proposed, however, none has proven to outperform the established inflammatory indices in distinguishing between IBD and IBS nor in distinguishing between CD and UC [23].

A complex inflammatory condition, such as IBD, is expected to be associated with an increased mucosal accumulation and release of inflammatory biomarkers into the faecal stream. This includes, but is not limited to, fCal. In agreement with this, Lehmann et al. demonstrated that the most significant alterations of the human faecal proteome between IBD patients and healthy individuals included not only inflammatory proteins but also immunoglobulins, antimicrobial and cell integrity proteins [24]. Proteases are also involved in the inflammatory process, and, in line with this, increased matrix metalloproteases, such as MMP9, have been found in IBD diseased mucosa and in stool [25], [26]. Therefore, in IBD it might be expected that an increased mucosal accumulation of inflammatory proteins and proteases that enhance protein fragmentation occurs, with a consequent accumulation of peptides in stool. An integrated approach aimed to evaluate both peptides and proteins in stool might enhance the chance to discover not only new biomarkers for IBD diagnosis but also to identify new molecular pathways involved in disease pathogenesis and outcome. Therefore, the aim of the present prospective study was to analyse the faecal peptidome and proteome in IBD patients, comparing them with values in healthy subjects by using high throughput techniques, namely, MALDI-TOF/MS and LTQ-Orbitrap XL MS.

Materials and methods

Results

The clinical characteristics of patients and controls are detailed in Tables 1 and 2. IBD patients comprised 99 CD, 71 UC and five patients with unclassified IBD. The exploratory cohort comprised patients and controls enrolled from September 2016 to September 2017 (34 controls, 72 CD, 56 UC and 5 unclassified IBD). The validation cohort comprised patients enrolled from October 2017 to February 2019 (28 controls, 27 CD and 15 UC). Exploratory cohort MALDI-TOF/MS analysis of faecal peptidomic profiles in the m/z range 1000–4000 identified a total of 438 features. Some features were common to patients and controls, while others were almost exclusively measured in either patients or controls (Supplementary Table 1). The spectra of control subjects were devoid of features (flat spectra) in 13/34 (38%) cases, a representative result being shown in Figure 1A. Similar flat spectra were found in a lower percentage (17/133, 13%) of patient samples (Pearson X²=11.904, p=0.001). In 21/34 control samples, MALDI-TOF/MS spectra were characterised by the presence of one or more features (representative example in Figure 1B), that belonged to the 79 control features or shared features list reported in Supplementary Table 1. The spectra of patient samples appeared much more numerous in features (representative examples in Figure 1C–F). On the basis of the high enrichment in features found in stool samples from IBD patients, we verified whether parallel variations could be observed in stool proteins by performing SDS-PAGE electrophoresis in a subset of 109 IBD patients (65 males, 44 females; mean age±DS=49±16 years) and 81 controls (48 males, 33 females; mean age±SD=50±8 years). Supplementary Figure 2 shows the results of a representative SDS-PAGE experiment.

Figure 1:

Faecal peptidomic profiling.

Representative MALDI-TOF/MS spectra in the m/z 1000–4000 range. Control spectra without features (A) or with control-associated and shared features (B). (C and D) Crohn’s disease spectra. (E and F) Ulcerative colitis spectra.

The MALDI-TOF/MS features collected from the exploratory cohort included 67/438 (15.3%) features shared between IBD patients and controls (shared features listed in Supplementary Table 1) and 359/438 (81.9%) disease-associated features (IBD features listed in Supplementary Table 1). Twelve out of 438 features (2.8%) were found only in control samples (control features listed in Supplementary Table 1). MALDI-TOF/MS spectra were evaluated and classified as control or IBD spectra on the basis of the following criteria: (1) control spectra if flat or showing only control or shared features; (2) IBD spectra if showing at least one IBD associated feature. The same criteria were then adopted to classify in blind MALDI-TOF/MS spectra obtained from samples of the validation cohort. Table 3 reports sensitivity, specificity and positive and negative predictive values of the above described MALDI-TOF/MS spectra classification in distinguishing controls from IBD patients in the exploratory and in the validation cohorts. For comparison fCal results at 50 μg/g cutoff are reported in the two cohorts. IBD MALDI-TOF/MS spectra were correlated with the absence of SDS-PAGE protein bands at a MW higher than 45 kDa, and with the presence of protein bands at a MW lower than 43 kDa (Supplementary Table 2).

IBD spectra were compared with clinical and biochemical parameters considering all patients and controls of the exploratory and validation cohorts. IBD spectra were correlated with positive fCal (>50 μg/g) when considering patient and control samples overall (Pearson X²=16.527, p<0.0001) but not when controls (Fisher’s exact test, p=0.407), CD (Fisher’s exact test, p=0.791) or UC (Fisher’s exact test, p=0.079) patients were considered singly. Neither in CD nor UC were IBD spectra and fCal correlated with clinical score (HBI or PMS), disease location, previous surgery or therapy. IBD spectra correlated with disease activity both in CD (Fisher’s exact test, p<0.0001) and UC (Fisher’s exact test, p=0.010). Positive fCal results were also correlated with disease activity (Fisher’s exact test, p<0.0001 for CD and UC), but this finding was expected since fCal was included in the classification of DR.

Binary logistic regression analysis was performed to ascertain which of the parameters associated with the presence of IBD is more effective in making a diagnosis. The results, shown in Table 4, demonstrate that IBD spectra were independently and highly significantly correlated with IBD. Of the 359 IBD-associated features, 34 were almost exclusively observed in CD, while 25 were almost exclusively observed in UC stool samples. With the aim of ascertaining whether these features might aid the classification of disease type, IBD spectra were classified as IBD-CD or IBD-UC when only CD or only UC associated features were found. IBD-CD spectra were significantly correlated with CD in the exploratory (Fisher’s exact, p<0.0001) and the validation cohorts (Fisher’s exact, p=0.028), as IBD-UC spectra were correlated with UC in the exploratory (Fisher’s exact, p<0.0001) and validation cohorts (Fisher’s exact, p=0.008).

MALDI-TOF/MS stool features were further characterised by MALDI-TOF/MS–MS, and fragmentation results were evaluated using the MS-Tag bioinformatics tool of ProteinProspector v 5.22.1 (available at http://prospector.ucsf.edu/prospector/mshome.htm) in the attempt to identify the candidate proteins (full molecule). A significant pattern was obtained for 14 features (Supplementary Table 3). We focused on the feature at m/z 1810.8, the most frequently present in IBD patients’ stool, that matched with the adenomatous polyposis coli gene 2, a homologue of the adenomatous polyposis coli (APC) tumour suppressor. In no cases was the entire APC protein detected in stool samples when western blot analyses were performed (data not shown). On the contrary, positive staining was obtained at immunohistochemistry of tissue samples. Figure 2 shows representative results obtained at immunohistochemistry, performed using intestinal tissue samples from CD and UC patients and from the healthy mucosa of one patient with colorectal cancer. In colorectal cancer, a few sparse interstitial cells were positively stained; in UC and CD an increasing number of infiltrating positive cells were observed in the colon mucosa, these cells being polarised towards the lumen in UC or towards the muscularis mucosae in CD. In CD patients, a diffuse infiltrate of APC positive cells was also detected in the ileum mucosa.

Figure 2:

Adenomatous polyposis coli (APC) immunohistochemistry.

In the left side the colon mucosa of a patient with colorectal cancer (A), a patient with UC (C) and a patient with CD (E) are shown. In the right side of the figure the ileum mucosa of CD patients are shown. In CD, APC positively stained cells (in brown) are highly represented in the stroma of the inflamed ileum (B, D and F), but also in the non-inflamed colon (C), and they are preferentially localised towards the muscolaris mucosae. In UC the same APC positive cells are mainly localised towards the surface epithelium.

To detect new IBD diagnostic protein biomarkers, we analysed three stool samples pools, made up of six control, five CD and five UC samples. After SDS-PAGE electrophoresis, proteins were gel extracted and analysed by LC-MS using an Orbitrap mass analyser. A total of 193 proteins were identified. Proteins were considered differentially expressed between patients and controls when the ratio between absolute intensity of the quantification values were above 2.0 (over-expressed proteins in patients, ↑) or below 0.5 (under-expressed proteins in patients, ↓). This left 161 differentially expressed proteins that are reported in Supplementary Table 4. A comprehensive analysis was made of the differentially expressed proteins using David software. GO terms for biological processes were significant (Benjamini and Hochberg adjustment for multiple comparisons, p<0.05) in the following comparisons: (1) increased in IBD with respect to controls, with no difference between CD and UC; (2) increased in IBD with respect to controls, with higher levels in UC than in CD; (3) reduced in IBD with respect to controls. Supplementary Table 5 reports the complete list of the significant GO terms for proteins found to be increased in faeces of IBD patients with respect to controls. Figure 3 shows the significant GO biological processes associated with over-represented and under-represented stool proteins in IBD patients with respect to controls.

Figure 3:

Significant GO terms for biological processes.

The pie chart shows the relative fold enrichment of GO terms found to be significant for stool proteins increased in IBD patients with respect to controls. The sorted slices show the GO terms for proteins with increased expression in IBD patients with respect to controls, with higher levels in UC than in CD. In the bottom, fold enrichment of significant GO terms for proteins with a reduced expression in IBD patients with respect to controls.

Discussion

Serum and faecal inflammatory biomarkers supporting a diagnosis of IBD have poor sensitivities and specificities [7], [8], [9], [10], [11], [13], [17], [18], [29]. The need to discover new biomarkers for diagnosing IBD and stratifying risk for tailored individualised treatment has yet to be met. Stool was chosen as the sample material because it is likely to contain more intestinal biomarkers than blood. Further, we also analysed MALDI-TOF low molecular weight features and proteins on the premise that they might directly reflect disease-associated cellular dysfunction [30]. The MALDI-TOF/MS analysis of stool from the exploratory cohort revealed very few, or no, features in healthy subjects, while stool of IBD patients were highly enriched in features. This finding, suggesting that IBD is associated with increased proteolysis, is in line with data in the literature reporting an increased expression of proteases in the diseased IBD mucosa, including MMP9, which was confirmed to be over-represented in IBD patients’ stool (see Supplementary Table 4) [26]. A higher frequency of proteolysis in IBD than in control samples was further supported by the SDS-PAGE findings exploring faecal proteins in a MW ranging from 200 to 3 kDa, with a MW higher than 45 kDa observed less frequently in IBD than in controls. The overall number of features commonly detected in both controls and patients was limited, while IBD-associated features were very numerous (>350), and each was variably associated with the other features in a very complex manner. In order to decipher and interpret the MALDI-TOF/MS results, each spectrum was classified as control spectrum if flat (no features) or with features (one or more than one) found in controls; MALDI-TOF/MS spectra were classified as IBD spectra if they presented one (or more than one) features found in IBD stool samples only. On applying this classification to the exploratory cohort, 58 control spectra and 109 IBD spectra were defined, this classification allowing a distinction between IBD patients and controls with a sensitivity of 81% and a specificity of 97%. This showed an improvement on fCal, which had a specificity of only 65%. To empower our findings, MALDI-TOF/MS blind analyses and spectra classification were performed using samples from a validation cohort. The obtained results confirmed that the specificity of MALDI-TOF/MS is high (97%), the sensitivity being 55%. In this validation cohort, fCal specificity was even lower (42%) than in the exploratory cohort. To explain why in the validation cohort IBD spectra had a lower sensitivity than in the exploratory cohort, it is important to bear in mind that, in general, validation of new biomarkers in patient cohorts different from those initially studied is almost always associated with reduction in clinical performance. More specifically, a different frequency of active IBD between the two cohorts (37% in the exploratory and 24% in the validation) might have had impact on the results, in view of the fact that positive IBD spectra were correlated with disease activity.

The numerous IBD associated features found prompted investigation of their parent proteins. Peptides were fragmented and analysed by MALDI-TOF/MS-MS, which was successful for 14 features. This small number of identified peptides with respect to the whole IDB associated features is a limitation that depends on the following: (1) MALDI-TOF/MS-MS can be performed on isolated features; (2) for a successful analysis, the features should have a high signal-to-noise ratio; and (3) an uncomplete and unpredictable feature fragmentation prevents the analysis. However, despite these limitations, our results indicated accumulation of immunoglobulins in IBD patients’ stool with respect to controls. This supports the relevance of the adaptive immune response in these inflammatory diseases, in agreement with Lehmann et al. [24], who found increased IgG and decreased IgA in IBD stool samples with respect to healthy controls. Pathways involved in the regulation of gene transcription, in the control of cell cycle and differentiation, and the Wnt pathway also emerged, confirming their involvement in IBD [31], [32]. Of particular interest are our results obtained with the m/z 1810.8 feature being observed frequently in patient stool but rarely in control stool. This matched significantly with the APC2 protein, homologous with the APC protein [33], a polarity regulator and tumour suppressor associated with development of familial adenomatous polyposis and colorectal cancer. To ascertain whether APC expression is altered in IBD mucosa, we performed immunohistochemistry that documented APC overexpression by stromal inflammatory cells which were not limited to the ill mucosa but diffuse throughout all the examined areas of the intestinal mucosa. APC is not only a tumour suppressor but is also involved in tissue damage and repair, especially in IBD, this function being correlated with its expression by vascular endothelial cells as recently demonstrated by Yoshimi et al. in a colitis animal model [34]. Moreover, a close relationship between lymphocyte function and APC expression has been recently reported by Aguera-Gonzalez et al. [35], who demonstrated that APC deficiency reduced NFAT nuclear localisation that determines a reduced Treg differentiation, favouring uncontrolled intestinal inflammation. The APC overexpressing cells infiltrating the IBD intestinal mucosa also showed a different polarisation between CD and UC: while in UC they were mainly polarised towards the intestinal lumen, in CD they were mainly polarised towards the muscularis mucosae. This different polarisation might be one of the mechanisms underlying the differences between the CD and UC intestinal inflammation patterns (transmural in CD, limited to the mucosa in UC).

In order to achieve a comprehensive proteomic analysis to shed further light on the diagnosis of IBD, we also evaluated differently expressed stool proteins by LC-MS. Although most of the identified stool proteins are plasma proteins and their increase in IBD stool might be a consequence of intestinal bleeding, we found also a large series of these plasma proteins that were reduced in IBD patients’ stool with respect to controls. Therefore, IBD associated variations of stool proteins might be the direct expression of disease alterations in the intestinal mucosa and not only of intestinal blood loss. Immunoglobulins were confirmed to be over-expressed in IBD, as were polymorphonuclear derived inflammatory proteins, such as S100A8, partner of S100A9 in the calprotectin heterodimer, neutrophil elastase and defensin. These proteins were found overexpressed in IBD stool also by Lehmann et al. [24]. Overall neutrophil mediated cytotoxicity and neutrophil mediated killing of symbiont cells represented the two main GO biological pathways identified. On the other hand, IBD-associated down expressed proteins were mainly involved in nucleic acid assembly, organisation and conformation, but also proteins related to the adherens junction organisation. Two identified proteins deserve further studies, since their differential expressions indicate not only a potential role for distinguishing between CD and UC but also in understanding why IBD patients are at a higher risk of developing cancer [36], [37]: olfactomedin-4 (OLFM4), absent in CD, and ectonucleotide pyrophosphatase/phosphodiesterase family member 7 (ENPP7), absent in UC. Both proteins, if reduced, might be related to an increased risk of primary cancer and metastases: OLFM4, normally expressed by intestinal crypt base stem cells, was found to be reduced in the subset of poorly differentiated and metastatic colorectal cancer [38], [39], while the lack of ENPP7, involved in sphingomyelin metabolism, significantly enhances colon cancer susceptibility [40].

In conclusion, our study provides evidence of the clinical utility of a new proteomic method for diagnosing IBD and offers insights on the pathogenic role of APC and other newly described proteins that, after validation, might become new tools in the diagnosis of IBD and the assessment of cancer risk in patients with the disease.