UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)
UvA-DARE (Digital Academic Repository)
The volatile metabolome and microbiome in pulmonary and gastro-intestinal
disease
van der Schee, M.P.C.
Publication date
2015
Document Version
Final published version
Link to publication
Citation for published version (APA):
van der Schee, M. P. C. (2015). The volatile metabolome and microbiome in pulmonary and
gastro-intestinal disease.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.
Chapter 13 General Discussion
Omics in Medicine: Pitfalls and Potential
Marc P. van der Schee Departments of Respiratory Medicine, Pediatric Respiratory Medicine, Academic
Medical Center, University of Amsterdam, Amsterdam, The Netherlands Departments of Pediatric Pulmonology, Pediatric Gastroenterology, Gastroenterology
and Hepatology, VU University Medical Centre, Amsterdam, The Netherlands Unpublished
Introduction
From genotype to phenotype in health and disease
The foundations for the way our body develops and interacts with our environment is detailed in our genetic material DNA1, as our genotype. Our genotype interacts
with our environment from the beginning of life to shape the way DNA is transcribed into RNA and subsequently translated into proteins. This ultimately determines our phenotype, the total of our observable traits, including health and disease. The human body strives to continuously maintain homeostasis; a state of balance within which our biological systems function optimally. The late respiratory physiologist Peter Macklem has described this more accurately as homeokinesis2, which reflects the dynamic state living organisms maintain
in interaction with their environment. This balance of stable adaptability is essential as it allows organisms to be stable but evolve in response to their environments. Disease can be considered as a failure of our compensatory mechanisms to maintain an adequate level of homeokinesis in response to environmental challenges. Understanding health and disease therefore means to understand the interaction between environmental exposures such as infection, pollution, diet and lifestyle on our bodies at the DNA, RNA, protein and metabolite level in an integrated way2.
Omics techniques; a novel approach to biomedical research
Despite the impressive amount of biomedical knowledge medicine is far from understanding this enormously complex system of the human organism as there are hundred thousands of variables involved. Science is however advancing towards the use of so-called ‘omics’-techniques that might expedite this process by enabling detailed profiling of the entirety of DNA (Genomics), RNA (Transcriptomics), Protein (Proteomics) or Metabolites (Metabolomics) in a single sample3. These omics techniques work from the basic concept that a molecular
signature differentiates between two phenotypes of interest. This is essentially different from ‘classical’ hypothesis driven research which would aim to test a predetermined assumption regarding a fixed set of parameters. Since omics techniques allow in depth qualification and quantification of biomarkers they may more accurately reflect the highly complex nature of health and disease thereby helping to identify novel biomarkers, pathways and treatment targets4,5.
In this thesis these concepts have been explored by studying the value of the volatile metabolome and the microbiome for disease diagnosis, phenotyping, monitoring and prognosis.
Volatile Metabolome
As discussed in greater detail in chapter 1, 11 and 12 metabolites are the end products of biochemical processes inside our bodies involving proteins, lipids,
amino acids, drugs, food, etcetera. Because a disease is by definition accompanied by a change in metabolism these are obvious candidate biomarkers. With respect to accessibility the volatile metabolome is especially appealing as it can be studied fully non-invasively through analysis of exhaled breath, skin emanations, urine, saliva, human breast milk, blood and feces6. The target molecules of interest are
so-called Volatile Organic Compounds, or VOCs. Specific VOCs can be identified through the use of chemical analytical techniques such as Gas-Chromatography coupled to Mass-Spectrometry. Alternatively, patterns of VOCs can be analyzed by promiscuous and cross-reactive gas-sensors that provide a pattern of sensor responses that reflects the composition of the entire VOC-mixture7. Since such an
approach closely mimics mammalian olfaction such techniques have been dubbed an electronic nose (eNose)8. The potential of such an approach is illustrated by
the impressive resolution of the human sense of smell discriminating at least 1 trillion separate scents9.
Microbiome
Microbiomics is a specific kind of genomics that focuses on profiling our microbiome, the genome of all microbial species we carry on and in our bodies10.
Microbiome analysis is generally performed by analyzing the encoding DNA of the s16ribosomal RNA component of the small subunit of prokaryotic ribosomes11
or, as in this thesis, by determining the length of the interspace (IS) region12. Such
analysis allows detailed characterization of the multitude of microbial strains present in a sample. In this thesis, as in general, microbiome analysis focuses on identifying bacterial species. Analysis of the microbiome is relevant because it constitutes one of the most extensive and omnipresent environmental exposures we face. This interaction is so close that we live in mutualism with these bacterial species; they inhabit us as their ecosystem and in turn can provide us with metabolic functions we lack13. Furthermore they play an important role in the
development of our immune system14–18. Early life disturbances in the buildup
of our microbiome may affect this balance and predispose to the development of a broad variety of diseases such as allergy14,19,20, asthma21–23, inflammatory
bowel disease24 and diabetes25. As such analysis of the microbiome holds a unique
position within omics techniques: Despite it being of extracorporeal origins its constitution may hold valuable information with respect to our bodily functions.
In this thesis we set out to identify the potential merits and pitfalls of omics techniques for disease diagnosis, phenotyping, monitoring and prognosis through
analysis of the volatile metabolome and the microbiome. In this chapter I will evaluate whether these objectives have been met and what the implications are of
the evidence generated in these studies for omics studies in general.
Disease diagnosis
Diagnosis; Infectious diseases affect the volatile metabolome
Infectious diseases are likely to affect the volatile metabolome in two ways; firstly the pathogen itself may produce volatile metabolites. Secondly a potential host response to this infection affects the exhaled volatiles26. In this thesis we
present evidence that VOCs might help to diagnose fungal, bacterial and viral infections. This is clinically relevant for it may enable more rapid selection of the appropriate therapy than classic culture based techniques.
In chapter 2 we have shown that exhaled breath analysis can discriminate between patients with and without invasive pulmonary aspergillosis. Invasive (Pulmonary) Aspergillosis (IA) is a life threatening fungal infection that can occur in patients who receive treatment for a hematological malignancy, resulting in prolonged neutropenia27. Because these patients are severely
immunocompromised the infection is fatal in 30-50% of patients28. Effectiveness
of anti-fungal therapy is highly dependent on early detection of the disease29.
Unfortunately symptoms are not specific nor sensitive27, resulting in the relatively
complex diagnostic criteria used to date30. This creates an obvious need for an
early diagnostic tool. In this study we provide proof of principle that distinct volatile metabolites can be associated with the presence of invasive aspergillosis. This may be related to VOCs directly produced by aspergillosis31 or the profound
host response initiated by this infection32. The potential impact of early diagnosis
of aspergillosis is profound: it may reduce the need for invasive bronchoscopy and reduce mortality by enabling timely installment of therapy. Because the current proof of concept study only discriminated patients with established invasive aspergillosis from controls it is not clear whether volatile metabolites would allow diagnosis prior to current diagnostic techniques. One subject did however appear to exhibit a change in exhaled volatiles two weeks prior to the clinical diagnosis of invasive aspergillosis was made. A currently ongoing prospective follow-up study will frequently monitor patients at risk of the development of invasive aspergillosis to confirm these results and assess the potential position of VOCs in the diagnostic workup.
The potential of volatile metabolites to diagnose infectious diseases was further explored in chapter 6. In this study we were able to associate the presence of an exacerbation with a distinct VOC-profile in children with cystic fibrosis (CF) or primary ciliary dyskinesia (PCD) with reasonable accuracy. This is clinically relevant because these, mostly bacterial, exacerbations are associated with an increased decline in lung function and are a heavy burden on patients33. Because
the number of patients experiencing an exacerbation in this study was small we were unable to differentiate whether specific infections are characterized by
distinct metabolic profiles. Delineating the contributions of individual infections to the exhaled breath profile is relevant for it may help to guide antibiotic therapy. Besides analysis of specific compounds by GC-MS this would require a very large long-term follow-up study; many CF and PCD patients have multiple co-infections, which severely complicates such analyses. Again, a prospective follow-up study is currently running in CF aimed to capture and, if possible, to predict bacterial-driven exacerbations and their association with the microbiome.
The proof of concept studies in this thesis and by other researchers26,34 point
towards several challenges that need to be address in the coming years to move volatile biomarker based diagnosis of infections to the clinic:
The first step will be to analyze the chemical compounds exhaled by patients in detail by chemical analytical techniques. This will help to build a library of VOCs related to infections disease. Second, researchers will have to differentiate between host-derived and pathogen- derived VOCs. This is commonly done by studying volatile metabolites of in vitro cultures of the target pathogen. Subsequently researchers attempt the identification of the same compounds in vivo26,35. This
approach has been used with variable success as only part of the established in
vitro markers are generally reproduced in vivo31,36,37. This is understandable as
it’s unlikely that the in vitro and in vivo metabolism of pathogens are identical. Furthermore pathogen related metabolites may be altered by the host.
An alternative approach is the one used in chapter 8; We established significant differences in exhaled volatiles upon comparing symptomatic and asymptomatic
Rhinovirus infections in pre-school children (p <0.001). Conversely analysis
(outside that mentioned in the manuscript) revealed this difference was less profound upon comparing asymptomatic Rhinovirus infection to uninfected asymptomatic children (p = 0.068). This suggests that, in Rhinovirus infection, the host-response has a more profound influence on VOCs than viral-derived volatiles. Conversely it also suggest that pathogen derived VOCs can be detected. This approach however also has considerable drawbacks; Patients can have multiple concurrent infections, associated co-morbidities (atopy, eczema, etc) and differences in environmental exposures (smoke, food intake, etc). Whilst such a study accurately simulates the diagnostic challenges in the intention to diagnose population, this may negatively influence the signal to noise ratio, hampering the potential to reliable identify host response related VOCs.
In this respect the approach by Bean and co-workers is very appealing: They exposed mice to both live Pseudemonas Aeruginosa and Staphyloccocus Aureus and its cell lysates. This allowed them to simulate both active infection (host and
pathogen volatiles) and the host response without active infection (host-derived volatiles). Interestingly they found that host VOCs correlated with the host immune response and differentiated between the origins of the infection. Such findings may allow subsequent targeted studies validating these compounds in humans. As suggested by Bos et al34 these effort should focus on identifying
pathogen specific volatiles (not produced by other pathogens) and disease sensitive biomarkers (found in all active infections). This concept illustrates the potential of combining multiple biomarkers into a clinical decision rule allowing both optimal sensitivity and specificity in a single test.
In fact the best way to move these techniques to clinic may be to combine both in
vitro and in vivo measurements. In vitro analysis of microbial cultures provides
more flexibility because it allows continuous measurements and enables researcher to manipulate the culture conditions in order to elicit pathogen specific metabolic changes38. Identification of specific pathogens is therefore likely easier
in vitro, which suffices for most clinical applications. In vivo metabolomics on the
other hand are particularly useful to measure host response derived volatiles. These may enable monitoring of therapeutic response and more importantly could allow detection of infection prior to symptoms arising helping physicians to time additional testing and initiate therapy timely.
Diagnosis; Cancer and the origins of its VOCs
For most types of cancer early detection is key to improving survival. Recent years have therefore seen the initiation of many community based screening programs for the most prevalent types of cancer. In this thesis we provide evidence that omics based approaches may be a valuable addition to such screening programs. In Chapter 3 we showed that patients with malignant pleural mesothelioma had a distinct VOC profile compared to controls. Mesothelioma is a rare type of cancer generally arising from the lung pleura and related to asbestos exposure39. In our
study these patients were discriminated both from healthy controls and at-risk patients with occupational asbestos exposure, which was confirmed by others40.
Such an approach in which the intention to diagnose population is studied is particularly relevant in diagnostic omics studies. This is illustrated by figure 6 in this manuscript which shows a 3-way discrimination of mesothelioma patients, healthy controls and asbestos exposed patients. The separate clustering of all 3 groups in this graph is indicative of the fact that all groups are characterized by a separate volatile metabolic profile. This suggests that markers differentiating between mesothelioma and controls are likely different from those discriminating asbestos exposed patients and mesothelioma cases. On one hand this illustrates the potential of volatile metabolites to reflect environmental exposures but on
the other hand it clearly indicates the risk of comparing gold standard healthy controls with advanced disease cases. While such approaches are important as proof of concept studies they may identify biomarkers or biomarker profiles that have very little value in the intention to diagnose population. This may cause researchers to overestimate the diagnostic potential for a specific application or chase false positive findings41. This strongly underlines the need to carefully
consider the study population when performing an omics based study.
As discussed in Chapter 12, the biomarkers we (Chapter 3) and others42,43 found
to differentiate between cancer and controls are likely to have originated from both local and systemic effects of a developing tumor. Local effects include the altered metabolism of the tumor itself, angiogenesis, cell necrosis and shifts in the microbiome. Systemic effect of cancer likely include increased oxidative stress, catabolism and immune activation known to be associated with changes in VOCs44. In Chapter 4 we therefore set out to determine what th e relative
contributions of local and systemic VOCs are to lung cancer derived VOCs. In this study we showed that alveolar air sampled directly at the tumor site differed significantly from the contralateral control segment. Additional unpublished data revealed this was not the case in healthy non-smoking controls. Furthermore lung cancer patients and controls could be discriminated from one another irrespective of the site sampled. This suggests that the presence of lung cancer affects both local and systemic VOCs. In this study we were unfortunately not able to meet our aim of determining the relative contributions of these local and systemic volatiles to the discriminating VOCs. This was due to the fact that cases and controls were not well matched and did not represent the intention to diagnose population. As discussed in the previous paragraph this represents a potential source of bias and may inflate the diagnostic accuracy warranting us to cautiously interpret the results.
Bronchoscopy studies by Santonico45 and Darwiche46 confirmed the presence
of a specific tumor site volatile profile. The presence of systemic VOCs is furthermore supported by the potential of exhaled volatiles to diagnose non-pulmonary tumors47. An interesting example in this respect is the potential to
diagnose colorectal cancer through analysis of exhaled volatiles48,49 which was
subsequently expanded by our data showing that both advanced adenomas and colorectal cancer have a distinct fecal volatile pattern50. The latter example
illustrates an important advantage of metabolomics studies: Since metabolic changes and the resulting metabolites are generally less confined to the primary disease site than changes on the DNA, RNA and protein level metabolomics studies are less dependent on obtaining biopsies from the primary pathological process. Samples obtained at multiple sites may therefore be of value for diagnosis of one disease as illustrated in chapter 12. Nonetheless this still means it is important
to carefully select the sample studied depending on the diagnostic question to maximize the presence of either or both local and systemic biomarkers. The clinical relevance of discriminating these two groups is analogous to that of discriminating pathogen and host derived volatiles; tumor derived volatiles may allow disease specific diagnosis whereas systemic VOCs may allow sensitive detection of a neoplasm.
Diagnosis; Chronic inflammation affects VOCs
The observation that inflammation caused by infectious disease influences the volatile metabolites suggests that chronic inflammatory diseases are also characterized by a distinct volatile metabolome. In fact the chronic inflammatory condition asthma was amongst the first and best reproduced conditions in which VOCs were analyzed as a diagnostic tool51–54 (Chapter 11). In this thesis we
expand on this knowledge by showing VOCs have similar diagnostic accuracy as other biomarkers for asthma and have diagnostic potential in other chronic inflammatory conditions.
In chapter 5 we performed an interventional study in which we compared the accuracy of asthma diagnosis by exhaled nitric oxide (FENO), sputum eosinophils and VOC analysis in patients when steroid free and when on steroids. These diagnostic markers were selected because both FENO55 and sputum eosinophils56
are known to relate to disease activity in asthma. They therefore are more likely to correlate with volatile metabolite profiles. In this study we established the diagnostic accuracy of these markers was comparable to that of VOC analysis. These results are encouraging because analysis of exhaled VOCs may have some specific advantages. Firstly induction of sputum eosinophils is a relatively cumbersome and technically demanding procedure. By contrast exhaled breath analysis, especially when performed by electronic nose, is fast, easy to perform and provides immediate results.
Secondly FENO is known to increase in response to inflammation other than that induced by asthma57. As shown in chapter 2, 6 and 8 the same holds true for VOCs
which are also affected by inflammation induced by pulmonary infections. By contrast to FENO as a single marker, VOCs may allow differentiation of asthma induced inflammation from pulmonary infections. The potential of an omics approach for disease diagnosis is also illustrated in chapter 5 by our data showing a decrease in the diagnostic accuracy of FENO after steroid treatment but an increase of the accuracy for eNose analysis. For FENO this may be explained by a reduction of airway inflammation by steroids58,59. VOCs however also detect
metabolites unrelated to the FENO pathway or may detect VOCs induced by the treatment itself. From this study it is unclear whether either or both of these
factors play a role. This is important to delineate because ideally metabolites both unaffected and affected by steroid therapy would be available for disease diagnosis and evaluation of therapy response respectively.
The potential of volatile metabolites to reflect chronic inflammation is also reflected by our data in Chapter 6 and 7. In Chapter 6 we showed the potential of exhaled breath metabolites to discriminate children with cystic fibrosis and primary ciliary dyskinesia from controls. Both these inherited diseases are characterized by a buildup of mucus inside the airways. This makes patients prone to recurring and chronic infections resulting in chronic airway inflammation with superseding exacerbations60,61. As is apparent from our data and that of
others62 these processes affect exhaled metabolites.
Similarly we show in Chapter 7 that Inflammatory Bowel Disease (Crohn’s Disease and Ulcerative Colitis) affect the fecal volatile metabolome of children. The diagnostic accuracies in this study were particularly promising, although the sample size did not allow full external validation. Furthermore this study was not performed in the intention to diagnose population thereby possibly overestimating the diagnostic accuracy in the clinical setting. Nonetheless these data are encouraging because excluding Inflammatory Bowel Disease is particularly relevant in children. This may prevent unnecessary colonoscopy which is burdensome in children because it requires general anesthesia. This may become all the more relevant in the coming years as the incidence of inflammatory bowel disease appears to be rising whilst the age of onset falls63.
Phenotyping and monitoring
Therapeutic challenges for most chronic diseases involve determining what therapy is suited best for a patient to optimize disease control and subsequently monitor that control in order to reconsider therapy when symptoms change. Omics techniques downstream of genomics reflect cellular activity and are therefore potential tools to meet these challenges. The next section of this thesis therefore focused on studying the value of volatile metabolomics to identify phenotypes associated with treatment response and monitor disease activity.
Prediction of steroid responsiveness by the volatile metabolome
In Chapter 5 we have shown that exhaled metabolites of steroid free asthma patients can predict whether they will respond to steroids. This is relevant because the cornerstone of asthma treatment is the use of steroids64. Unfortunately not
all patients benefit from taking these steroids56. Therefore a trial of steroids is
generally prescribed to assess the therapeutic response empirically64. This
potentially results in unwanted side-effects and a delay in the onset of efficacious
therapy. We therefore emulated this diagnostic challenge by withdrawing steroids from patients with established asthma. After a steroid-free period of 28 days or upon losing asthma control patients were started on 30mg of daily oral prednisone for 2 weeks. After this trial we assessed the therapeutic response of patients by measuring potential improvement in lung function. Steroid responsiveness was predicted based on sputum eosinophil counts, FENO and VOCs, all assessed when subjects were steroid-free. Unfortunately due to safety reasons several patients did not complete the full protocol thereby considerably reducing the sample size. Nonetheless we established that analysis of exhaled VOCs could more accurately predict if a patient would be responsive to steroid therapy than sputum eosinophil counts and FENO.
A possible explanation for these findings lies in the mechanisms that determine steroid responsiveness which are strongly linked to the sputum inflammatory cell type65. This is supported by the notion that these cells and their activity affect
exhaled VOCs66,67. Furthermore we were able to correlate sputum eosinophil
cell counts to the exhaled metabolic profile. The higher accuracy for prediction of steroid responsiveness of exhaled volatiles compared to FENO and sputum eosinophils may be explained by two factors. Firstly this may be the omics nature of the technique combining a wide range of markers as discussed earlier. Secondly the VOCs produced by these cell lineages reflect their metabolic activity which is likely to be more relevant to the disease activity than the number of cells. This was corroborated by the study of Fens et al67 indicating that neutrophil
activity was associated with exhaled volatiles in COPD. The findings in our study are the first step towards using volatile metabolomics as a tool to tailor therapy in asthma patients. The next step would be to validate and extend these findings in a study focusing on steroid-naïve asthma patients prior to treatment with inhaled corticosteroids.
Phenotyping: Metabolomics discriminate diseases with similar clinical presentations
In chapters 6 en 7 we aimed to assess the ability of volatile metabolomics to discern clinically similar diseases that differ in their underlying pathophysiology. As described earlier we studied two mucociliary clearance diseases CF and PCD in chapter 6. Especially in childhood these diseases resemble each other in there clinical presentation: Frequent recurring pulmonary infections, cough, wheezing and failure to thrive; although PCD usually has a milder clinical course and less extra pulmonary manifestations60,61. The underlying cause of
the two diseases are however distinct: Cystic fibrosis is linked to a mutation in the CFTR gene resulting in abnormal chloride and sodium transport across the epithelium with hyper viscous secretions as a consequence68. In Primary Ciliary
the airways are dysfunctional69. Both diseases are therefore characterized by a
buildup of mucus in the lung which makes patients prone to recurring infections. This drives chronic inflammation that may progress to bronchiectasis and can potentially require lung transplantation68. The volatile biomarker profiles of
both diseases were discriminated from one another with a reasonable accuracy. These differences may be driven either by differences in the primary disease process or the resulting inflammatory processes that are known to be partially distinct70. The value of metabolomics for these diseases does however not lie in
discriminating these two phenotypes of mucociliary clearance disease because genetic markers are well established for CF60 and are becoming increasingly
available for PCD71. The primary benefit will lie in the monitoring of disease
activity (Section 3.3). These results however support that disease monitoring is possible because part of the diagnostic biomarkers are likely related to inflammatory activity.
In chapter 7 we performed a similar study in which we were able to discriminate two phenotypes of Inflammatory Bowel Disease (IBD) through analysis of fecal metabolites. Both phenotypes of IBD; Crohn’s Disease and Ulcerative Colitis are chronic, relapse-remitting, auto-immune conditions of the intestine that can present similarly with abdominal pain, vomiting, diarrhea, bloody stools, fever and weight loss72. They are however pathophysiologically distinct, although a
certain amount of overlap appears to exist73. Crohn can affect any part of the
gastrointestinal tract from mouth to anus but usually starts in the terminal ileum. Ulcerative Colitis is characterized by lesions that are confined to the large intestine and rectosigmoid72. Discrimination of these two phenotypes has
therapeutic implications and is generally done through colonoscopy.
In our study we were able to discriminate these two phenotypes non-invasively with very promising accuracies both at first presentation and after achieving remission. Besides host derived biomarkers a substantial part of the signal allowing discrimination of these phenotypes may originate from disturbances in the microbiome which are related both to the disease and its treatment74.
VOCs related to this aberrant microbiome may explain why we found only a slight decrease in diagnostic accuracy when patients achieved remission even though inflammation related VOCs likely decreased75. This would suggest a
relatively high contribution of the fecal microbiome to fecal VOCs which may be a consequence of the fact that microbes are more readily transported with the feces than the volatiles originating from the gut mucosa. Evaluating the relative contributions of microbial and host related VOCs would require a study similar to that in chapter 4 by analyzing VOCs during colonoscopy at the disease site. We are currently initiating such a study which will be very relevant to identify
VOCs that are directly linked to disease activity. Especially such biomarkers are prime candidates for disease monitoring (section 3.3) in IBD.
The studies in chapter 5, 6 and 7 illustrate the potential of volatile metabolomics to discriminate phenotypes that are not readily distinguishable based on their clinical presentation. In these studies all patients were classified as belonging to a certain phenotype based on other criteria such as therapy response or the underlying disease process. Whilst this is a valuable approach when phenotypes can clearly be defined, it may limit detection of clinically relevant biomarkers when phenotypes are not that clear cut. This has mainly to do with the fact that current definitions of disease are largely based on the clinical presentation of patients. Many clinical presentations however encompass multiple distinct disease processes having distinct biomarkers. Therefore, combining these distinct disease processes may obscure important biomarkers that are only present in a minority of the subjects. Especially in studies using biomarker profiles without identifying individual compounds this risk is relevant. A potential way to tackle this challenge is to apply unsupervised clustering on combined omics and clinical datasets. This statistical technique generates clusters by grouping patients together that resemble each other most closely with respect to both their symptoms and biomarkers. In asthma such an approach has proved valuable76.
These ‘unbiased’ approaches may identify novel phenotypes that are based on both biochemical and clinical characteristics and may have relevance for therapeutic response and disease prognosis.
Monitoring: Loss of disease control affects the volatile metabolome
A tool that allows continuous monitoring of disease activity is relevant to many chronic inflammatory conditions because symptoms often lag behind77, or can
be discordant to disease activity76. Metabolites are prime candidate biomarkers
for such applications as they change rapidly in response to changes in disease activity.
In chapter 5 we illustrate the potential of VOC analysis to monitor disease activity in asthma. In this study we observed that the exhaled metabolites of patients who lost asthma control after withdrawal of steroids differed significantly from those who did not. Sputum eosinophils had a similar accuracy but FENO could not significantly identify those patients that lost control. We attempted to elaborate on this through the study in chapter 6. We however did not fully meet our aim of assessing whether disease activity influences volatile metabolites in mucociliary clearance disease due to the low number of exacerbations in the dataset. Even though we did observe significant differences between patients with and without an exacerbation for both CF and PCD the accuracy for especially CF is too modest
to be relevant for clinical use. A better powered study could help to discern whether this is a power issue or if it represents the true accuracy. In CF and PCD this will likely remain challenging as many potentially interfering processes contribute to the exhaled metabolites: the primary disease process, chronic and acute infections, commensal microbes and chronic inflammation.
In chapter 7 we were not able to discern directly whether or not the disease activity affects the fecal volatile profile. Firstly we established that volatile metabolites differentiated between patients with IBD and controls, both during active disease and when achieving remission. Additional analysis revealed that fecal volatile profiles differed within patients between active and stable disease. This would suggest that disease activity affects the fecal metabolites. We however also established shifts in fecal volatiles in patients that did not achieve remission and in controls. As we have shown adequate stability of the technique, it is likely that normal daily variability contributes substantially to the fecal volatile profile. Unfortunately this prevents reliable assessment whether disease activity affects VOC-profiles in this study. This issue clearly illustrates a potential pitfall of metabolomics studies. Because a multitude of markers is assessed it is very likely that factors such as diet influence the fecal volatiles thereby explaining the significant shifts in fecal volatiles between different days.
In chapter 11 we discussed how such observations have spawned much research into potential confounding factors in VOC research such as smoking, exercise, diet, age, sex, pregnancy, medication, background air, etc78–85. All of these factors were
indeed shown to influence volatile metabolites. This has led some researchers to control all these aspects rigorously in their experiments. I however feel such an approach works under the false assumption that removing this influence will yield more reproducible results. To date there is no evidence to suggest that a fasting patient has a more reproducible VOC pattern than a subject that just ate. Even still the induced catabolism may in fact produce VOCs related to oxidative stress86 that hamper the detection of important volatiles of disease related
oxidative stress87. I feel this data therefore predominantly illustrates the power of
omics techniques to reflect a wide variety of conditions. This does however also mean researchers should be wary of any potential sources of systematic biases between cases and controls such as smoking when studying lung cancer.
Nonetheless these data do reveal that the ability to identify target volatiles may be adversely affected by the variability induced by these factors. Two complementary approaches are relevant to tackle this issue. Firstly statistical techniques that allow detection of relevant patterns in a background of considerable random variability are very rapidly improving88. The success of such an approach is
illustrated by our data showing that despite considerable day to day variability we were still able to discriminate patients with IBD from both control samples (Chapter 7). Secondly the signal to noise ratio could be significantly improved by developing a technique that is more sensitive and selective to the target compounds. After this proof of concept the first step will be to analyze these samples by chemical analytical techniques to identify the key compounds driving the signal; something we are currently pursuing. Subsequently knowledge of such compounds could be used to develop sensors tailored to a specific application possibly circumventing most of the confounding issues.
The observations in chapter 5, 6 and 7 support the potential of volatile metabolomics for disease monitoring. The required next steps are to initiate longitudinal studies that simultaneously assess known (invasive) measures of disease activity and the volatile metabolome by chemical analytical techniques. This will ultimately help to determine whether volatiles change prior to symptoms arising, allowing timely installment of therapy that could significantly reduce the severity of the oncoming exacerbations, or prevent it all together. In asthma first steps in this direction have been taken89.
Disease prognosis
The notion that metabolites change prior to clinical symptoms arise suggests that early prediction of disease could be facilitated by metabolomics. This is something we explored in the final section of this thesis. We did so by recruiting the 1200 infant EUROPA-study cohort which aims at studying early signs of asthma development. In this cohort children were prospectively monitored for the occurrence of respiratory symptoms. If children had respiratory symptoms warranting parents to contact their general practitioner, we performed a home visit to assess, amongst others, exhaled metabolites and the nasopharyngeal microbiome. The goal of the EUROPA-study is to assess the integrated value of such biomarkers for the early diagnosis of asthma. The studies in chapter 8 & 9 take the first step to explore this concept by assessing the potential of these biomarkers to discriminate between children at high-risk to develop asthma and controls. Children with Rhinovirus-induced wheezing were chosen in these studies as a high-risk phenotype as they have a tenfold increased risk to develop asthma90.
In chapter 8 we established that pre-school children with physician confirmed respiratory wheezing had an exhaled VOC-profile distinct from asymptomatic children. Strikingly, those children with RV-induced wheeze could still be discriminated from controls after full symptomatic recovery. This was contrasted by children with non-RV-induced wheeze whose exhaled volatiles more closely resembled those of asymptomatic controls after recovery. These observations
suggest that exhaled volatiles are associated with a pre-existing host-response predisposing children to develop symptomatic infections91, or a prolonged
inflammatory response induced by Rhinovirus in children with a symptomatic infection92. Irrespective of the underlying mechanism, these results indicate
metabolic profiling may have potential as an early diagnostic tool for asthma, as recently shown for GC-MS93.
In chapter 9 we discuss our findings regarding the nasopharyngeal microbiome in relation to children with Rhinovirus-induced wheezing. Firstly we established that the microbial abundance in the nasopharynx was lowest in asymptomatic children, was increased in children with non-wheezing respiratory illness and was highest in those children having confirmed expiratory wheeze. This abundance decreased in wheezy children after recovery from symptoms. This observation was not influenced by the presence or absence of Rhinovirus. Secondly we observed a decrease in the microbial diversity (Shannon-index) in the two bacterial families (phyla) of Proteobacteria and Firmicutes in children with Rhinovirus-induced wheeze relative to non-Rhinovirus-induced wheeze. Interestingly these differences disappeared after resolution of symptoms and were not found in non-wheezing respiratory illness nor in controls. These findings may be in keeping with earlier observations suggesting that decreased early life microbial exposures may predispose to the development of allergic disease and asthma14,94,95. Upon interpreting this data it is important to realize that the
Shannon diversity-index is a measure that increases when the number of strains in a sample increases and when the homogeneity of their abundance increases. As the diversity decreased and the abundance increased in children with Rhinovirus-induced wheeze this suggests that specific strains are up-regulated in children with Rhinovirus-induced wheeze. Upon closer examination we established this increase occurred across the phylum and couldn’t be isolated to specific strains. Bacteria from the Bacteroidetes phylum were only observed in a subset of children with symptomatic respiratory infections and none of the controls. The abundance of Bacteroidetes was higher in wheezing children as compared to non wheezing respiratory illness. In children with Rhinovirus-induced wheezing the abundance of bacteroidetes did not change upon recovery from infection whilst a marked decrease was noticeable for children with non-Rhinovirus-induced wheezing. Interestingly, this observation resembles that found by analysis of VOC-profiles in the same population in chapter 8.
The bacteria from the Bacteroidetes phylum normally reside within the gut. We can speculate that a disturbance in the mucosal immunity enables migration of bacterial species from the gut and/or oropharynx to colonize the nasopharynx
and predisposes to symptomatic Rhinovirus infection96–98. Another contributing
factor may be mechanical translocation from the oropharynx secondary to altered airway dynamics. Interestingly upregulation of gut Bacteroidetes has previously been associated with a positive asthma predictive index99. Furthermore, as
is apparent from the significantly higher bacterial load of Bacteroidetes in
Rhinovirus-negative wheezing children compared to Rhinovirus-negative non
wheezing children, the presence of Bacteroidetes is also associated with more severe symptoms in Rhinovirus-negative children.
Taken together these findings fit with a hypothesis considering the development of asthma20 which is graphically depicted in figure 1. Evidence is growing to suggest
the microbiome plays a key role in early life immune development by inducing a T-helper 1 response16, promoting allergen tolerance17,100 and shaping a beneficial
response to infections97,98,101. Early life exposures and a genetic predisposition
may predispose to airway inflammation and/or a disordered microbiome. Such a microbial disturbance, or dysbiosis, may prevent the microbiome from exerting its effects on the immune system14,23. These factors may reinforce one another in
a vicious circle based on impaired mucosal immunity96–98. The resulting chronic
inflammation may ultimately drive pathological changes inside the lung giving rise to the asthma phenotype19. There is convincing evidence for a window of
opportunity in early life during which microbial substitutions14,23,102 may prevent
development of asthma. The same interventions do not appear to have an effect at older ages, possible due to decreased plasticity of the immune response14,23. These
findings stress the importance of research in the pediatric population, because it may ultimately facilitate manipulation of the microbiome for treatment and prevention100. The testing of this hypothesis will be part of the follow-up of the
EUROPA-study.
Finally this hypothesis may also explain why both the volatile metabolites and the upper airway microbiome show prolonged changes in children at increased risk to develop asthma: the VOCs may reflect the underlying inflammation linked to a disorder microbiome and may originate from metabolites produced by the microbiota themselves103. The combination of microbiome analysis - as a measure
of bacterial presence - and the analysis of volatile metabolites - as a measure of activity - may therefore be highly complementary. The currently ongoing follow-up in the EUROPA-study will help to determine whether these biomarkers have complementary value for the early detection of asthma.
Fi gu re 1 . H yp ot he si s f or m ic ro bi om e d ri ve n a st hm a d ev el op m en t F ig ur e 1. G ra ph ic al de pi ct io n of m ic ro bi al hy po th es is . E ar ly lif e ex po su re s an d (e pi )g en et ic pr ed is po si tio ns ca n w or k in co ns or tiu m to re su lt in a di so rd er ed m ic ro bi om e an d a irw ay i nf la m m at io n. T he se t w o c an p ot en tia te e ac ho th er i n a v ic io us c irc le b as ed o n a n i m pa ire d m uc os al i m m un e r es po ns e. C hr on ic i nf la m m at io n e ve nt ua lly dr iv es pa th ol og ic al ch an ge s in th e lu ng th at pr es en t i ts el f i n cl in ic as hi gh -r is k ph en ot yp es fo r t he de ve lo pm en t o f a st hm a an d ev en tu al ly , a st hm a its el f. Th is fig ur e al so i llu st ra te d t he d iff er en t p oi nt o f a ct io n w he re p er so na liz ed m ic ro bi al s ub st itu tio n i nt er fe re s w ith t hi s v ic io us c irc le a s c om pa re d t o a nt i-in fla m m at or y t re at m en ts .
Impaired Mucosal Immune Response
Early Life Exposures - Maternal Smoking - Infection - C
-Section
- Prematurit
y
- Antibiotics - etc. (epi)Genetics
Disordered Microbiome Airw
ay infl
ammation
High-Risk asthma phenot
ypes - Wheezing to Rhino virus - A top y Chronic infl ammation - Altered airw ay mechanics - Hyper responsiv eness
- Altered systemic immunit
y
Asthma
Current ther
apies
Personalized microbial intervention
Technical validation VOC analysis
In this thesis we explored three novel applications of volatile metabolomics: Firstly, we showed adequate reproducibility of the discrimination of Rhinovirus induced wheeze and controls in chapter 8 between two electronic noses, consistent with findings in adults51. This study however does not allow us to
assess the repeatability of individual measurements in children, something that will need to be done in future studies. Furthermore the ‘normal’ development of VOC profiles over time should be investigated in a longitudinal study to establish age specific reference signals.
Secondly, we investigated the effects of stool sample handling on the fecal VOC profile (Chapter 7). In this first proof of concept study we found that multiple cycles of thawing and freezing stool samples did not affect the fecal VOC profile. Furthermore temperatures between 15 and 35 degrees did not significantly alter the fecal VOCs in this experimental setup. It is extremely relevant to expand and repeat these experiments with combined eNose and GC-MS analysis. Factors such as homogenizing, water content, sample volume, sampling speed, sampling volume, headspace volume and headspace saturation time may have profound influences on the obtained results.
Finally, we assessed the potential to store and transport exhaled breath samples on so-called Tenax tubes to allow centralized analysis (Chapter 10). This is relevant because multi-centre analysis can help prevent inter device variability. In this study we were able to show that storage of exhaled volatiles for a period of up to 14 days did not significantly affect core exhaled volatiles. Furthermore the potential of the eNose to discriminate asthma from controls was not significantly affected over time. This latter observation is particularly important because it suggests that the inevitable perturbations in VOC composition caused by storing, transporting and desorbing VOCs does not render these samples useless.
Nonetheless it is important to realize that whilst centralized analysis may be an essential first step in developing volatile biomarker based tests, it is not the future of such applications. Pattern recognition based VOC-analysis techniques such as the eNose are especially valuable as point of care tools (Chapter 11 & 12). The current inter device variability of many techniques, including the one used in this thesis (Cyranose 320), is however too high to allow current implementation in a clinical setting. It is important to realize this does by no means renders the results in this thesis useless; they should be regarded as an explorative analysis of the potential of volatile metabolomics in medicine, not a ready to implement clinical application. What these studies illustrate is that the potential for volatile
metabolomics is profound, warranting further studies and investments into such applications. During the past years the first public-private-partnerships have been initiated between technical companies and clinical partners helping to farther techniques, such as FAIMS, that are reproducible between devices, have increase sensitivity and specificity, lower costs and are more user friendly.
Outlook
Systems medicine integrates multiple omics techniques
Each omics technique has its own potential depending on the desired application. Besides genetic mutations in individual cells the genome is relatively stable during an individual’s lifetime irrespective of age, disease activity, potential co-morbidities, etc. This enables relatively easy sampling of large cohorts of patients without potential biases introduced by these factors. On the downside this means disease monitoring is not possible by genomics based techniques. Transcriptomics is the most downstream technique that allows profiling of all the molecules in that domain providing a complete overview of cellular activity. On the downside transcriptomics highly depend on acquiring a pure sample of target cells, which often requires invasive procedures and complex sample handling104,105. Transcriptomics
are therefore less suited to acquire a rapid overall view of the functioning of an entire individual. With proteomics sample collection procedures are generally less demanding as most body fluids are easily accessible. Furthermore, proteomics can help to determine the functioning of a specific organ or the body as a whole. The major downside of proteomics however is that there is no technique available that allows profiling of all proteins in a sample as is the case for RNA with RNAseq. This means proteomics are less unbiased compared to ‘true’ omics techniques that allow full profiling. Metabolomics have the same limitations regarding detection of compounds. The technique is however especially attractive because it allows rapid and fully non-invasive sampling. When based on pattern recognition techniques such as the eNose these are true point of care tools that allow snapshot profiling of the metabolic status of the entire body5.
Figure 2. Applications of omics techniques
Metabolites
Degradation
Genomics Transcriptomics Proteomics Metabolomics
Diagnosis ± + + +
Phenotyping ± + + +
Monitoring − ± ± +
Prediction ± ± ± ±
Figure 2. Depiction of the dominant flow of information through biological systems, the associated omics techniques and their theoretical value for biomedical applications.
As is apparent from figure 2 all omics techniques have their specific potentials and limitations. Their true potential therefore becomes apparent when information regarding the multiple levels at which biological systems function are integrated together with clinical, physiological, environmental (the exposome) and epidemiological parameters3. Such a holistic approach is ultimately necessary to
unravel the complex molecular pathways underlying disease and is the arena in which systems biology operates106. Ultimately, combining these techniques
can help to construct a so-called interactome; the whole network of molecular interactions that occur across the biochemical families of the central dogma. This provides a model to describe the way in which the genotype and environment interact to produce a specific phenotype which is in essence the way our bodies function107. The value of such an approach is supported by chapters 8 and 9 in
which novel insights might be generated by combining volatile metabolomics and microbiome analysis.
The challenge of translating omics applications to clinical practice
As is evident from this thesis high-throughput omics technologies have a wide range of potential applications: they may improve disease diagnosis, monitoring, phenotyping, could help to predict an individual patient’s clinical course and response to therapy. This relatively young research field is however still challenged by considerable pitfalls and limitations that prevent researchers from harnessing its full theoretical potential. This is reflected by the limited translation of the multitude of promising results to clinically useful applications that impact patient health. Several responsible issues can be identified some of which can be tackled through adequately designing experiments, others inherent to omics techniques.
Pitfalls and Limitations of Omics techniques High-dimensional
The vast array of analyzed molecules in omics studies usually generate high-dimensional datasets. Complex statistical approaches are applied to translate these patterns into clinically relevant outcomes. Whilst this data-processing is an essential step to generate interpretable outcomes, it also carries risks.
Firstly, with each processing step it becomes increasingly harder to interpret the individual contributions of the originally identified biomarkers. This is even apparent from the microbiome analysis in chapter 9; the statistics of overall abundance and diversity are relatively crude. This makes it challenging to identify individual microbes that drive these signals. Furthermore such summary statistics hold relevance on a population level but may not hold clinical relevance for individual patients. The resulting risk of over-interpretation is illustrated by
our data in chapter 9. We observed an upregulation of Bacteroidetes in children with symptomatic respiratory infections relative to controls. After recovery from these infections the overall abundance of Bacteroidetes was still up-regulated. Based on figure 3 it is tempting to conclude this represents chronic colonization of the nasopharynx by this phylum. Upon closer examination we however established that only a fraction of patients were colonized at both occasions. This radically changes the interpretation of the data; Bacteroidetes are associated with (former) symptomatic respiratory tract infections on a population level, but dynamically occur in individual patients.
At first glance this issue appears to be less relevant for pattern-recognition based techniques such as the eNose which do not allow correlation with individual compounds. This issues is however most relevant for such techniques as the pattern-based nature of the outcomes can make it difficult to assess whether outcomes hold relevance for individual patients or represent a phenomena on a population level.
Another consequence of such high dimensional datasets is the data-analysis itself, especially when integrating multiple approaches into a systems medicine approach108. Such multi-scale ‘big data’ has a particularly bad signal to noise
ratio and a low number of cases relative to the observations, both of which may undermine the reliability of study results by posing a risk for false discoveries109,110.
This is unfortunately illustrated by the relative lack of externally reproduced results in the literature.
Besides being conscious of the issues stated above there are several possible mitigating actions which are increasingly incorporated into omics studies. The common goal of these approaches is to validate the robustness and plausibility of study findings.
Statistical validation - First and foremost all findings should be validated in
a novel population. As a startingpoint this requires internal (bootstrapped) cross-validation of datasets41. Preferable is to externally validate findings in a
newly recruited population. Both approaches work from the same basic concept: a diagnostic algorithm is constructed based upon a training population, this algorithm is subsequently tested in a novel population of similar patients whom were not used to construct this algorithm. This approach allows testing of the value of the devised algorithm for individual patients partly emulating its incorporation in clinical practice.
Data-integration – An attractive way to address some of these issues is to perform
large multi-centre trials such as the recently initiate U-BIOPRED project on airways disease111. By combining data from multiple study centers the number of
cases relative to the number of observations is improved. Furthermore integration of multiple omics techniques in a systems biology approach may provide more robust and biologically plausible outcomes (see below). A prerequisite is that the sampling and analysis procedures are standardized and preferably centralized (chapter 10).
Although uncommon today, the progress in medicine would be greatly expedited if all researchers would make their omics data publically available. This would allow swift external validation of results and the collaborative building of disease maps such as recently initiated for Parkinson’s disease112. Similarly
databanks profiling the entire proteome113 and volatile metabolome114 are
currently being build. Eventually such approaches may enable construction of cloud based datasets. Such centralized databases holding anonymous omics profiles of patients may help to create self-learning systems in which every novel patient contributes to the accuracy of the diagnostic algorithm by contributing its data. The potential to speed up the development of omics based applications is considerable. Unfortunately, the current competitive nature of science prevents data-sharing from being common practice.
Biological plausibility – If compounds underlying a molecular signature can be
identified, their biological plausibility should be tested by comparing it to known cellular pathways115. A molecular signature matching such a pathway may have
more validity compared to one which doesn’t. Various of such databases are currently available or being built for genomics, transcriptomics, proteomics and metabolomics approaches6,116,117. Besides these relatively complex analytical
approaches a more pragmatic approach may provide a similar biological validation. In chapter 9 we associated the presence of Bacteroidetes with children who were born by c-section and/or had a lower gestational age. As similar observations were previously established in experimental studies this provides a biological foundation for this observation118. It is important to realize that these
type of analyses have the generation of plausible hypotheses as their primary goal. As discussed earlier such findings need to be evaluated in a study properly designed for this goal.
Discarding results uncorrelated to clinical outcomes
Irrespective of the previous statements it is important to not discard any results that can’t be explained by our current biomedical knowledge. Most disease definitions, including most studied in this thesis, rely heavily on clinical
presentation. The underlying pathophysiological mechanisms may however be very heterogeneous. This may cause researchers to discard clinically relevant results as they do not appear relevant for all patients with a specific diagnosis.
Unsupervised clustering - This risk may be mitigated by applying unsupervised
clustering. When applied to omics dataset such techniques can identify clusters of biologically similar patients. These ‘biological’ phenotypes can subsequently be correlated to both clinical and physiological outcomes (also see 5.3) providing plausibility of such phenotypes76.
Assessing biomarker presence without function
Another potential pitfall of omics based studies is considering biomarker presence with disregard to its function, in relation to the physiological processes inside the body. This is especially true for genomics studies, but also applies to other omics techniques. As an example a microbiome study may not be able to differentiate between two clinical conditions. This does however not imply that the microbiome does not carry relevance to this disease as significant differences in metabolic activity may occur. Conversely shifts in the microbiome may not relate to relevant shifts in overall microbial activity119. Such emergent phenomena are not captured
by merely describing biomarker presence2.
Combining techniques – By combining multiple omics techniques both the function and presence of molecules can be assessed. Examples are the combination of single cell genomics and transcriptomics and the study of both microbiome and volatile metabolites.
Incorporating physiological and environmental parameters – By incorporating
these parameters into omics based studies the relevance of these biomarker profiles for the homeokinesis of our bodies in reaction to the environment can be assessed. This may provide important insights not identified when merely assessing biomarker presence. An example of this is the identification of genetic variation on the 17q21 locus which were only associated with asthma development in children with Rhinovirus induced wheezing91. As illustrated by these data,
longitudinal studies may be key to identify the functional ramifications of the established biomarker profiles because biomarker patterns are highly individual and vary over time6,120.
Not assessing the intention to diagnose population
As discussed in paragraph 2.1, those biomarkers that discriminate between gold standard healthy controls and gold standard cases are likely to be different from those that discriminate cases and controls in the intention to diagnose
population. Whilst evaluation of gold standard cases and controls is valuable as a proof of concept, it is important to realize that the resulting algorithm cannot be externally validated in the intention to diagnose population and may overestimate the diagnostic potential of the technique.
Studying intention to diagnose population – The best way to evaluate the potential
of omics based techniques in clinical practice is often to immediately study it in the intention to diagnose population despite the fact that these studies are often more challenging.
Technological challenges
As most omics techniques are fairly new there often is a lack of understanding with respect to possible confounding conditions and the best sampling protocols. This induces variability between studies which hampers combining of data and prevents external validation. This lack of standardization is especially worrying as good reference standards often lack. Therefore our ability to detect sampling artifacts and outliers is strongly limited. Combined these factors likely contribute to false positive associations.
Technological advancement – Omics based analytical techniques are rapidly
evolving. Challenges for the coming years are to devise techniques that allow full characterization of the proteome and metabolome in a single sample allowing true omics studies in these fields. Furthermore the knowledge generated by clinical studies should be used to develop targeted sensors for specific clinical applications. This may help to strongly improve the reliability, reproducibility and accuracy of these diagnostic tests, possibly even down to the mutation level121.
Sampling standardization – It is extremely relevant to conduct careful
methodological experiments that allow standardization of measuring protocols. Fortunately international taskforces seeking to fulfill this need are currently underway for exhaled volatile analysis. These taskforces have the challenging task of balancing careful control of experimental conditions without complication the technique to a point that it bypasses its clinical applicability.
In summary, besides implicit strengths, omics techniques have several limitations and potential pitfalls. By carefully abiding to guidelines regarding the development of diagnostic tests122, omics study design123 and the interpretation
of results108 most limitations can be overcome. From these recommendations a
1. Validation and standardization of analytical technique and sampling method 2. Biomarker discovery in intention to diagnose population
3. Validation: External validation, understanding of biological pathway 4. Development of low-cost, targeted high-throughput assay
5. Validation and refinement of assay in targeted clinical setting 6. Formal incorporation of assay into clinical practice
7. Continuous improvement of assay by cloud based integration of data 8. Reiteration based on improved algorithm from step 4 onwards
5.3 Future or fantasy? : Beyond personalized medicine; individualized medicine
As discussed earlier, the patient heterogeneity found within disease populations may mean our current disease definitions do not adequately represent the underlying biochemical processes. In fact we know very well from clinic that multiple disease processes lead to similar clinical outcomes and vice versa. If the unsupervised clustering techniques discussed in section 5.2 are applied to ever bigger datasets that include an increasing variety of clinical presentations (e.g. not asthma but obstructive lung disease in general) we might be able to build novel disease classifications based on the underlying disease processes. As illustrated in chapter 5 this may help to personalize interventions thereby improving the therapeutic response for patients.
Even though technology is currently not at such a state, we could take this even further by means of a thought experiment, which might solve the problem of population heterogeneity alltogether. Through increasing integration of omics data researchers could be able to create ever more detailed phenotypes until the hypothetical situations occurs in which each person is his or her own phenotype. Individuals could have their entire interactome profiled to detect what biochemical processes are changed relative to a healthy status.
This scientific progress will need to be paralleled by an ethical discussion regarding the potential ramifications of such an approach. What do we want to know? And how well in advance? Who is allowed to have access to our information; the patient, the doctor, the government or even the insurance company? While such detailed information has powerful beneficial applications, its dual use may have undesired adverse effects such as stigmatization and exclusion of individuals from healthcare and society. It is likely that our ability to correlate biomarker profiles to clinically relevant outcomes will precede our abilities to efficiently and cost-effectively manage these conditions. The challenge for the coming years will be for society to effectively address this balance between privacy and preparedness.
If done successfully this may help to break down the current undesirable distinction between research data and operational clinic data. Therapeutic agents could then be targeted directly at the disease process, perhaps even irrespective of the symptoms, as a type of individualized medicine. In this respect it might be worthwhile to explicitly study those individuals that do not develop disease despite having the same environmental exposures that individuals whom did attract the disease have. This may yield powerful new insight into what genetic or environmental factors are protective. Especially the latter could provide novel therapeutic strategies previously not considered such as deliberately exposing individuals to a ‘healthy’ microbiome100.
In this respect early life21,124 and even ante-natal125,126 exposures are key to the
way our interactome develops . Strikingly this ‘learning’ effect even appears to be preserved across generations through so-called epigenetic phenomena127 meaning
that environmental exposures of your ancestors may affect your phenotype. This once again stresses the importance to perform longitudinal cohort studies in the pediatric population.
In summary the research in this thesis displays both the potential merits and pitfalls of omics studies. If the current challenges are overcome it’s likely these techniques will strongly affect the way we diagnose, phenotype, monitor and treat patients over the next decades. This is something I hope to actively contribute to during the forthcoming years.
