SEARCH

SEARCH BY CITATION

Keywords:

  • electronic nose;
  • colorectal cancer;
  • adenoma;
  • screening;
  • fecal immunochemical test;
  • volatile organic compounds;
  • smellprint;
  • flatography

Abstract

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. References

In the course and prognosis of colorectal cancer (CRC), early detection and treatment are essential factors. Fecal immunochemical tests (FITs) are currently the most commonly used non-invasive screening tests for CRC and premalignant (advanced) adenomas, however, with restricted sensitivity. We hypothesized that fecal volatile organic compounds (VOCs) may serve as a diagnostic biomarker of CRC and adenomas. In this proof of concept study, we aimed to assess disease-specific VOC smellprints in fecal gas to distinguish patients with CRC and advanced adenomas from healthy controls. Fecal samples of patients who were scheduled to undergo an elective colonoscopy were collected. An electronic nose (Cyranose 320®) was used to measure VOC patterns in fecal gas from patients with histopathologically proven CRC, with advanced adenomas and from controls (no abnormalities seen at colonoscopy). Receiver operator characteristic curves and corresponding sensitivity and specificity for detection of CRC and advanced adenomas were calculated. A total of 157 stool samples (40 patients with CRC, 60 patients with advanced adenomas, and 57 healthy controls) were analyzed by electronic nose. Fecal VOC profiles of patients with CRC differed significantly from controls (area under curve ± 95%CI, p-value, sensitivity, specificity; 0.92 ± 0.03, <0.001, 85%, 87%). Also VOC profiles of patients with advanced adenomas could be discriminated from controls (0.79 ± 0.04, <0.001, 62%, 86%). The results of this proof of concept study suggest that fecal gas analysis by an electronic nose seems to hold promise as a novel screening tool for the (early) detection of advanced neoplasia and CRC.

Colorectal cancer (CRC) is one of the predominant cancers, contributing to a high burden of morbidity and mortality in the United States of America and Europe.[1, 2] Early detection and treatment are critical factors in the course and prognosis of CRC, and screening programs have proven to be an important means to reduce both mortality and secondary economic burden.[3-5] Colonoscopy is considered the gold standard for CRC and advanced adenoma screening. Fecal immunochemical tests (FIT) are currently the most commonly used non-invasive fecal screening tests. However, sensitivity of FIT for CRC is between 66–88%[6-10] depending on the cut-off values used, whereas sensitivity for advanced adenomas is disturbingly low (27–41%).[6, 8, 11, 12] As CRC prevention programs should primarily focus on early detection of premalignant advanced adenomas, the search for novel, more accurate non-invasive screening methods remains warranted.

Analysis of volatile organic compounds (VOCs) in exhaled breath has been reported as a potential non-invasive diagnostic biomarker test for lung cancer, breast cancer, malignant melanomas and CRC.[13-15] VOCs are gaseous carbon-based chemicals resulting from biochemical processes in the body, which are discharged by exhaled air, sweat, urine and feces.[16] VOCs in fecal gases are mainly produced by the intestinal microbiota in the colon during fermentation processes and also derive from metabolic processes within microorganisms.[17-20] Molecular compounds of VOCs in exhaled air or from feces can be individually detected by gas chromatography and mass spectometry (GC-MS). However, this technique is expensive and requires time-consuming off-line analysis, making its widespread application in clinical practice unfeasible.

Electronic noses enable real-time, high-throughput analysis of the complete spectrum of VOCs in complex gas mixtures. The electronic nose (e-nose) technology is based on an array of nanosensors, each reacting to different fractions of the VOC mixture by a sensor-specific change in resistance. Combination of individual sensor measurements create a specific smellprint (fingerprint), which can subsequently be analyzed by means of pattern recognition algorithms.[21, 22] No studies on the usage of the e-nose in fecal gas analysis are currently available in literature.

We hypothesized that e-nose may discriminate fecal samples from patients with CRC from those with advanced adenomas and from (healthy) controls, by disease-specific pattern recognition of VOC smellprints. This hypothesis was tested in a cross-sectional proof of principle study analyzing VOC in the headspace of fecal samples, obtained from patients with CRC, patients with advanced adenomas, and controls. As secondary aim, accuracy of fecal VOC-analysis with the FIT was compared by performing a FIT on fecal samples from patients with CRC, patients with advanced adenomas, and controls.

Material and methods

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. References

Subjects

For this proof of principle study, we used fecal samples of patients who were scheduled to undergo an elective colonoscopy at the endoscopy unit of the VU University medical centre and Sint Lucas Andreas Hospital (both located in Amsterdam, the Netherlands). The samples were primarily obtained for an explorative pilot-study on FIT and fecal DNA characteristics.[23, 24] The indication for colonoscopy was based on the judgment of the treating physician and included both symptomatic and screening/surveillance patients. Patients with a history of inflammatory bowel disease were excluded, as well as patients with inadequate bowel cleansing or in case of failure to perform a complete colonoscopy. Patients were requested to collect a fecal sample before colonoscopy and before bowel preparation was initiated and to store the sample in their refrigerator at home. After transport to the hospital, the samples were stored at −20°C until further handling. No clinical or demographic data other than procedure outcome, age and gender were obtained from the included patients. The samples were labeled based only on endoscopic and histopathological outcomes, in case biopsies were harvested or polypectomies were performed. For the purpose of this study, we only selected patients without abnormalities at colonoscopy and absence of histopathological abnormalities of the harvested biopsies (controls), and patients with histopathologically proven advanced adenomas or CRC.

The study was approved by the Medical Ethics Review Board of the VU University medical centre and all subjects gave consent before inclusion.

Endoscopic evaluation

Colonoscopy was the standard of reference for the presence and size of colorectal neoplasia. The colonoscopies were performed or supervised by experienced gastroenterologists. Endoscopists were unaware of the FIT and e-nose results. A complete colonoscopy was defined as intubation of the cecum with identification of the terminal ileum or ileocecal valve in combination with the appendiceal orifice, or intubation up to an obstructing CRC. The results of histopathological analysis of tissue samples obtained during colonoscopy were the standard of reference for the diagnosis of cancer or advanced adenoma. Adenomas ≥1.0 cm, with any villous features (i.e., tubulovillous or villous adenoma) or high-grade dysplasia, were considered advanced adenomas.[25] If multiple lesions were present, classification was based on the most advanced lesion found.

Fecal volatile biomarker analysis

Fecal gas analysis was performed with a Cyranose 320 e-nose® (Smiths Detections, Pasadena, CA), a portable chemical vapor analyzer, containing a nanocomposite array consisting of 32 polymer sensors. The VOCs interact competitively with these sensors, depending on the chemical characteristics of both sensor material and present VOCs. Multiple biomarkers interact with each individual sensor and individual biomarkers interact with multiple sensors. The polymers swell in response to this interaction, inducing a change in electrical resistance. Modification of resistance of each sensor is combined into a so-called smellprint, representing composition of the total VOC mixture. This smellprint can be used to discriminate different clinical groups by pattern recognition analysis.

Before fecal gas measurements with usage of the e-nose were performed, ∼2 grams of frozen feces was transferred from the stored fecal samples into a sealed vacutainer (BD vacutainer, Franklin Lakes). This amount of feces was chosen to provide an optimum ratio of VOCs to headspace, as validated by GC-MS analyses. The vacutainers were resealed and gradually heated to 37°C for 1 hr in an incubator, to enhance vapor release from the stools. The heated vacutainers were subsequently connected to the e-nose in an air-tight closed loop system. This system was created by piercing a needle through the cover top of the container (Terumo Europe N.V., Leuven, Belgium) and connecting this with the e-nose by a tube (Argyle Kendall tube 3 mm, Mansfield, Massachusetts). The outlet of the e-nose was subsequently connected with the headspace with an identical needle and tube to prevent dilution of the headspace with ambient air. A polyethersulfone syringe waterfilter (VWR international B.V., Arlington Heights, IL) and a 3-way stopcock system (BD Connecta, Helsinborg, Sweden) were included in the system to control airflow direction and prevent contamination of e-nose by condensation.

A stable baseline reference signal was created by connecting a VOC-filter (A1, North Safety, NL) to the e-nose. Subsequently the headspace was sampled for 60 sec in order to reach a stable reference sensor response. After sample analysis all sensors were purged with VOC-filtered air to wash off the VOCs and create a stable baseline for the subsequent samples (control value). All measurements were performed under a fume hood. Needles, hoses and 3-way stopcocks were replaced after each sample measurement.

FIT methods

The FIT used in this study is an automated OC-sensor test (Eiken Chemical Co., Tokyo, Japan), which provides a quantitative outcome. Patients were instructed to discontinue anticoagulants 5 days before colonoscopy. Tests were analyzed using the OC sensor MICRO desktop analyzer (Eiken Chemical Co., Tokyo, Japan) according to instructions of the manufacturer.[26] A cut-off level of 75 ng/ml (or 15 µg Hb/g feces) was used.[27] Tests were analyzed by one of two experienced technicians who were unaware of the clinical data. Both technicians received special training for analyzing the tests.

Data analysis

The available demographic data are given descriptively. VOC smellprints as measured with Cyranose 320 e-nose® were analyzed with Statistical Package for the Social Sciences software version 20 (SPSS, Chicago, IL). Principal component analysis (PCA) was used to recombine the variance of the original dataset into a set of principal components. The potential of principal components to discriminate our clinical groups was tested by independent t-test. Selected principle components were used in a canonical discriminant analysis (CDA) to calculate a probability of belonging to either of the diagnostic groups for two out of three cases (training set). The resulting classification algorithm based on this training set was externally validated in the remaining cases (validation set).[28] This analysis was repeated with a 1000 random distributions of cases into either the training or validation set. The resulting cross-validated probabilities were used to construct a mean receiver operator characteristic (ROC) curve for these 1000 simulations providing the sensitivity, specificity, area under the curve and positive and negative predictive values for the algorithm. Because of the limited availability of demographic data, we sought to minimize the risk of a Type 1 error. To do this we generated 1000 random classifications of the subjects into case or control. Subsequently, we used the methodology described above to construct a diagnostic algorithm for each of these distributions. The percentage of these random classifications that led to a cross-validated accuracy identical or greater than our primary distribution (CRC, advanced adenoma, healthy control) can be interpreted as the risk of a Type 1 error and was therefore used as our p-value.[28] The repeatability of fecal volatile analysis by e-nose was assessed by calculating intraclass correlation coefficient for five duplicate measurements of three distinct stool samples.

Results

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. References

Characteristics of the study population

Stool samples from 40 patients with CRC, 60 patients with advanced adenomas (together 100 samples with advanced neoplasia) and 57 controls were analyzed with both e-nose and FIT. Patient characteristics are depicted in Table 1. The median age of all patients was 60 years (IQR 42–71) and 54% of these were female. In all patients a complete colonoscopy was performed and bowel preparation was judged adequate by the individual endoscopists.

Table 1.  Subject characteristics
 ControlsAACRCTotal
  1. AA = advanced adenomas, CRC = colorectal carcinoma

Number of samples576040157
Age at diagnosis (years)(median, (IQR))38 (30–44)66 (58–75)69 (61–78)60 (42–71)
Missing (n)11  11
Gender    
Male (n) (%)16 (28)35 (58)15 (3866 (42)
Female (n) (%)28 (49)25 (42)25 (62)78 (50)
Insufficiently documented (n) (%)13 (23)0013 (8)
Lesion size    
>10 mm 59  
<10 mm 0  
Insufficiently documented (n) 1  
Location of lesion    
Proximal colon (n)  14 
Distal colon (n)  20 
Insufficiently documented (n)  6 

Fecal volatile organic compound analysis

Fecal VOC profiles of patients with CRC differed significantly from healthy controls (mean area under the curve (μAUC) ± 95% CI for the 1000 random distributions of training and validation sets 0.92 ± 0.03, p-value < 0.001, sensitivity 85%, specificity 87%). To limit the number of false negative results (Type 2 errors), a different cut-off point in the ROC curve could be selected with corresponding sensitivity of e-nose for CRC of 93%, at the expense of specificity, as a consequence decreasing to 76% (Fig. 1). In addition, VOC profiles of patients with advanced adenomas could be discriminated from healthy controls (μAUC ± 95% CI 0.79 ± 0.04, p < 0.001, sens. 62%, spec. 86%). By ROC curve analysis, sensitivity of e-nose for advanced adenomas could be improved to 81%, with corresponding specificity of 58% (Fig. 2). Furthermore, patients with CRC differed significantly from patients with advanced adenomas (μAUC ± 95% CI 0.82 ± 0.04, p < 0.001, sens. 75%, spec. 73 %) (Fig. 3) and advanced neoplasia (advanced adenomas and CRC together) could be distinguished from healthy controls (μAUC ± 95% CI 0.83 ± 0.04, p < 0.001, sens. 85%, spec. 68%). Results are summarized in Table 2.

image

Figure 1.  ROC-curve with 95% confidence interval for diagnosis of CRC compared with the controls. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

image

Figure 2.  ROC-curve with 95% confidence interval for diagnosis of advanced adenomas compared with the controls. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

image

Figure 3.  ROC-curve with 95% confidence interval for diagnosis of CRC compared with advanced adenomas. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

Table 2.  Test characteristics for 1000-fold cross validated VOC-analysis of fecal samples of colorectal carcinoma, advanced adenoma and controls
 μAUC ± 95% CIp-valueSensitivitySpecificityLR+LR−
  1. Sensitivities, specificities, positive and negative likelihood ratios are reported for the respective optimum cut-points. Advanced Neoplasia is the combination of advanced adenomas and colorectal carcinoma.

  2. μAUC ± 95% CI = mean area under the curve with 95% confidence interval for the 1000 random distributions of training and validation sets; LR+ = positive likelihood ratio; LR- = negative likelihood ratio.

Colorectal carcinoma vs. control0.92 ± 0.03<0.00185%87%6,50,2
Advanced adenoma vs control0.79 ± 0.04<0.00162%86%4,40,4
Advanced adenomas vs. colorectal carcinoma0.82 ± 0.04<0.00175%73%2,80,3
Advanced neoplasia vs. control0.83 ± 0.04<0.00185%68%2,60,2

Repeatability

The intraclass correlation coefficients (± 95% CI) for five duplicate measurements of three distinct fecal sample were 0.997 (±0.002), 0.997 (±0.001) and 0.997 (±0.001), indicating extremely high levels of measurement repeatability.

Results FIT

The FIT was performed on 133 fecal samples (35 CRC, 46 advanced adenomas, 52 controls). The FIT positivity rate was 19% (25/133; 3 advanced adenomas and 22 CRC samples, negative in all control samples). Sensitivity and specificity of FIT for CRC, advanced adenomas and advanced neoplasia (advanced adenomas and CRC together) are given in Table 3.

Table 3.  Results FIT
 SpecificitySensitivitycut-off (ng/ml)
  1. AA, advanced adenomas, CRC, colorectal carcinoma, AN, advanced neoplasia (advanced adenomas and colorectal carcinoma combined).

AA100%7% (3/46)75
CRC100%63% (22/35)75
AN100%31% (25/81)75

Discussion

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. References

In this proof of concept study, fecal VOC patterns of patients with CRC and advanced adenomas were compared with controls by means of e-nose. We have shown that fecal VOC analysis by e-nose is a feasible technique with high-throughput capability. Fecal samples from patients with colorectal carcinoma and advanced adenomas could firmly be distinguished from controls. Moreover, an e-nose could differentiate between advanced adenomas and CRC by fecal gas analysis.

Information on fecal VOC and their potential as diagnostic biomarkers in gastrointestinal diseases is limited, since most studies were performed with exhaled breath analysis. In few studies, differences in VOC profiles from patients with various gastrointestinal diseases, compared with controls have been reported, all of these measured by GC-MS.[18, 19, 29, 30] Since fecal VOC are mainly produced by the intestinal microbiome, it is not surprising that specific differences were detected in that set of diseases that are linked to alterations in the intestinal microbiota composition (e.g., inflammatory bowel disease (IBD), necrotizing enterocolitis (NEC), diarrhea-predominant IBS and bacterial gastro-enteritis.[18, 19, 29, 30] Although these studies were all pilot-studies with limited numbers of patients, fecal VOC seemed to contain biomarkers with clear diagnostic and disease activity monitoring potential, a finding we reproduced in our study aiming at (early) detection of advanced adenomas and CRC.

In the pathogenesis of CRC, changes in intestinal microbiome composition have been reported to be present, but a cancer-related microbiotal signature has not yet been identified.[31, 32] Recently, Altomare and colleagues identified disease-specific VOC in exhaled breath from 37 patients with CRC by means of GC-MS analysis, compared with 41 controls.[33] On the basis of the differences in chemical compound composition, CRC could be distinguished from controls with an accuracy of 76%. Furthermore, it was shown that a specially trained Labrador dog was able to distinguish stool samples from CRC patients from controls by detection of canine scent (sensitivity 97%, specificity 99%).[15] However, patients with advanced adenomas were not included in both studies. Whether the observed differences in VOCs or smell were because of specific changes in the intestinal microbiotal composition or due to alterations in (local) metabolism secondary to cancer remain to be clarified.

The e-nose technique is a non-invasive, high-throughput method based on pattern recognition of response to vapors. Real-time analysis of the complete spectrum of VOCs is possible, reinforcing its potential as clinical tool. Previous studies in exhaled breath have shown adequate repeatability and reproducibility.[21, 22, 34, 35] In this study, reproducibility of the e-nose technique in fecal gas analysis was reassuringly high. Because no previous studies on VOC analysis in fecal gases with this e-nose technique have been described, information on the optimal circumstances, such as temperature, amount of substrate, pressure and water content of the fecal material, was not available. To test our hypothesis, we used fecal samples instead of exhaled breath samples, because disease-specific VOCs are produced presumably by an altered (specific) intestinal microbiota composition due to the presence of premalignant or colorectal tumor cells. In addition, collection, storage and analysis of fecal samples can be controlled more vigorously compared with exhaled breath samples. Confirmation of our hypothesis as measured in fecal samples at body temperature would reinforce e-nose as a potential bedside CRC screening instrument.

This study has shown that e-nose could distinguish CRC from controls by detection of disease-specific fecal VOC profiles with a sensitivity of 85% and a specificity of 87%. We selected this cut-off point in the ROC curve to optimize corresponding specificity for CRC. To limit Type 1 errors, a different cut-off point in the ROC curve may be selected improving sensitivity for CRC to 93%, but at the expense of specificity. Reported historical FIT data show a sensitivity for CRC of 66–88% and a specificity of 87–96%.[6-10] More importantly, e-nose enabled differentiation between patients with advanced adenomas and controls, with a clinically relevant sensitivity (sensitivity 62%, specificity 86%) compared with reported FIT characteristics (sensitivity FIT for advanced adenomas 27–41%, specificity 91–97%).[6-10] Again, by selecting a different cut-off point on the ROC curve, sensitivity of e-nose for advanced adenomas may even be improved to 81% (with a corresponding decrease of specificity to 58%)

In this study, we compared the e-nose characteristics with FIT results obtained from the same fecal samples. Sensitivity and specificity of FIT for both CRC and advanced adenomas differed largely from historical FIT data. A potential explanation for these differences could be that all fecal samples had been stored at −20°C for a period of time, before FIT was performed. In the setting of this study, these circumstances may be detrimental for the condition of the stored samples, for example by a decrease in haemoglobin concentration. Another limitation of this study is the limited availability of demographic data of the included individuals. Information on use of medication potentially influencing VOC composition (like antibiotics), and dietary intake during collection of the fecal samples was not obtained. However, in a previous landmark study on fecal VOCs in healthy subjects, it was shown that the majority of fecal VOCs remain relatively constant in health, with only limited changes because of day-to-day dietary habits.[17] To assess whether the limited demographic dataset might have influenced our e-nose results, we performed a stringent additional statistical analysis, revealing that the probability of a Type 1 error (false positive outcome) was very low for all of our diagnostic algorithms (p < 0,001). Future studies comparing and combining e-nose and GC-MS with FIT analysis, performed on fresh fecal samples, should include complete demographic data to allow clinical delineation of these non-invasive tests in CRC screening. Such an approach may help to identify biomarkers both specific and sensitive for CRC. Future tailoring of sensors toward such biomarkers may further increase the accuracy of the e-nose technique in the detection of CRC and advanced adenomas, enhancing its potential value as screening tool. In summary, the reported results showed that an e-nose can be used to distinguish CRC from healthy controls by analysis of fecal VOC profiles in the headspace of stool samples. Interestingly, advanced adenomas could be distinguished from controls with a higher sensitivity compared with historical FIT results. Therefore, analysis of fecal gas by e-nose seems to hold promise as a novel screening tool for the (early) detection of advanced neoplasia and CRC.

References

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. References