Nutritional Metabolomics to Identify Biomarkers of Dietary Patterns and Specific Diet Exposures


In the NIH metabolomics interest group webinar
series. Today’s webinar from Dr. Mary Playdon will
focus on nutritional metabolomics to identify biomarkers of dietary patterns, and specific
diet exposures and its application to understanding breast cancer etiology. Before introduce our speaker I would like
to review a few logistics. Should you experience a technical difficulty
you can feel free to contact us through the question box and the expanded control panel,
by phone or e-mail. If you need to view live closed captioning,
please click on the link that will appear in the chat box. I believe it’s already there, so please
click on that link. And also at the end of Dr. Playdon’s presentation
we will allow time for questions. If at any time during the presentation you
have a question, though, please feel free to type it in the question box and then I
will ask it on your behalf at the end of the Q&A. Don’t wait, you can type your questions
in as the presentation is going on.>>Okay! Today it is my pleasure to introduce Dr. Mary
Playdon. Dr. Playdon joined the National Cancer Institute’s
Division of Cancer Epidemiology and Genetics in August 2014 as a pre-doctoral fellow through
the Yale University-NCI partnership training program. She completed a bachelor of science and a
Master of Public Health at the Queensland University of Technology in Australia. Dr. Playdon is a registered dietitian practicing
in Australia and England before coming to the United States. Her research interests include nutritional
epidemiology, metabolomics and etiology and survivorship of female cancers. As a post-doctoral fellow, Dr. Playdon continues
her work on nutritional metabolomics and cancer etiology and has expanded her portfolio to
include alcohol and obesity exposure and their relationship to hormone metabolism and other
female cancers including endometrial cancer. If you can all just give me a minute to get
Dr. Playdon’s slides up, I would appreciate it. K just give me one more second.>>All right, I would like to turn it over
to Dr. Playdon. Dr. Playdon: [2:22] Thanks Krista. And thanks everyone for being here. As Krista mentioned I’m going to present some
recent work under attritional metabolomics and its application to understanding breast
cancer etiology. But first I’d like to propose this question
to you, does diet matter for cancer prevention? It is currently estimated that around one
third of the most common cancers could be prevented through diet, weight control and
physical activity. This is very important because cancer is the
second leading cause of death in the U.S. and the leading cause of some age groups,
and the prevalence is increasing worldwide.>>If we can identify lifestyle strategies
that are modifiable for preventing cancer, we could dramatically impact cancer rates. Here on the screen is a summary of what we
know so far about foods and cancer risk based on the American Institute of Cancer Research
world expert report. On the left we have list of food exposures
and along the top we have a list of cancers. The dark pink squares show that the foods
convincingly increase cancer risk. The dark blue squares show convincingly decrease
risks and the yellow circles highlight where some of the evidence is stronger but clearly
there are lots of empty spaces and a lot of exposures where the evidence is not strong
either way.>>One area that is interested me personally
is breast Cancer prevention. Breast cancer is the second leading cause
of cancer death among women and it can lead to long term associated comorbidities. So far besides obesity alcohol intake is the
only definitive diaper dated risk factor for breast cancer. Research on many other diet-related factors,
e.g. fats, fruits, and vegetables, have been quite conflicting.>>[4:22] One possible reason for the mixed
finding is diet is generally measured using self-reported questionnaires. Now, these are prone to measurement errors
that can make it difficult to observe diet associations. One way that we might improve measurement
of diet in epidemiological studies is by using objective nutritional biomarkers. But the problem is only a handful of validated
biomarkers currently exist.>>Metabolomics is an emerging technology
and it can be used to identify nutritional biomarkers in human bio specimens, for example
blood and urine. Hundreds of molecules can be measured of one-time
including endogenous nutrition and energy balance metabolites and hormones, as well
as exogenous food breakdown products. You can see examples here on this slide.>>Metabolomics allows for hypothesis generating
agnostic type analyses to uncover nutrition-related metabolic pathways that may be associated
with disease and may not have been previously considered. Secondly, metabolomics provides us with an
opportunity to identify new and better nutritional biomarkers for potentially improving measurement
of diet exposures. Thirdly, mediation analysis can be used to
uncover mechanistic mediators of diet and disease relationships.>>There are two major questions that I’m
going to address today in this webinar. Firstly, there are very few studies that have
looked at biomarkers of overall diet quality. Secondly, few studies have looked at nutritional
metabolomics and its application to exploring breast cancer etiology.>>The first analysis that I’ll speak about
today looks at identifying biomarkers and dietary patterns using metabolomics.>>[6:26] Overall diet quality can be measured
by scoring how closely food intake patterns align with national dietary guidelines. There is mounting evidence that overall diet
quality is associated with lower risk of many chronic diseases and mortality. Even some diseases where no individual foods
were associated with risk have been shown to be strongly associated with overall diet
quality. Diet quality scores take into account the
complexity of an overall pattern of eating and this includes the fact that components
of your diet are often correlated and they might also interact to affect disease risk.>>There were two objectives for this first
study. The first was to identify metabolite associations
with four dietary patterns that are all based on national healthy eating guidelines as well
as the components that make up each of those diet indices. And the second objective was to gain insights
into the potential biological mechanisms that are influenced by overall diet quality. The study population were healthy male Finnish
smokers, aged 50 to 69 who were participating in the Alpha-Tocopherol, Beta-Carotene Cancer
Prevention Study (ATBC), which was a large randomized trial that was conducted in Finland. We used pre- randomization baseline data from
five nested case-control studies to do a cross-sectional analysis among 1336 men. We focused on over 1300 metabolites that were
measured in at least two out of five studies using the Metabolon platform, the commercial
platform. These included both known and unknown metabolites. And the Inter-Class Correlation Coefficients
(ICCs) over all metabolites were high across the studies and that is suggesting we had
good platform reliability. The diet for the 12 month prior to study enrollment
was measured using a self-administered food frequency questionnaire. It was very detailed and it was designed specifically
for the male Finnish population.>>[8:45] Now, we chose diet indices that
have been thoroughly characterized in the Dietary Patterns Methods Project (DPMP) which
developed a process to allow dietary patterns to be compared across very diverse cohorts. We used this standardized process to convert
the Finnish food variables into US equivalents before calculating the diet index scores. Here is an example of the four diet indices
we measured. Along the top are the indices and listed below
are the food components which are scored for each diet index. First the healthy eating Index is an index
of overall diet quality and based on the U.S. dietary guidelines for Americans. The Ultimate Mediterranean diet score was
based on epidemiological studies in Europe that investigated mortality risk factors. The Baltic Sea diet was based on the Baltic
Sea diet pyramid and this is specific to Finland to represent a regional diet. Finally, the World Health Organization’s healthy
diet indicator is based on the international food dietary guidelines for the prevention
of chronic diseases. This index has primarily macro and micro nutrient
components as opposed to whole foods. It represents a contrast to the other three
indices. Each index has a slightly different scoring
criteria that they each provide overall score of diet quality on either a continuous or
ordinal scale.>>There were three stages of this analysis. Firstly we look at the association between
it metabolites and each diet index using partial Spearman correlation. Then we looked at metabolites in each of the
components that make up the diet indices. We used fixed effects meta-analysis to drive
a summary estimate for each dietary index and metabolite correlation and we used Bonferroni-correction
for multiple comparisons.>>Finally we did a pathway analysis to look
at metabolic pathways associated with each of the diet indices. And we only looked at metabolic pathways from
which we had measured at least five metabolites.>>So, the results- the men in our study were
an average age of 57, slightly overweight, with less than elementary or eighth grade
education. They were physically active and heavy smokers.>>Here are the results for the average healthy
dietary pattern scores. For each diet index a higher score means better
compliance or healthier diet. The maximum scores are listed on the right-hand
column. The score for the healthy eating index, to
put it into context, was actually quite similar to a recent analysis among non-smoking U.S.
adults that was based on national health and nutrition examination survey data.>>[12:02] Overall there were 23 metabolites
that were associated with the healthy eating Index at the Bonferroni-corrected alpha level
and this included 17 with identifiable chemical structure. Here on the slide I’m only showing the chemically
identified metabolites. The absolute correlations ranged up to .2
and we found the metabolite profile overall reflected most of the underlying components
that we used to score adherence. You can see these components are represented
by the different boxes here.>>For the alternate Mediterranean diet there
were 46 associated metabolites including 21 with known chemical identity and the absolute
correlations ranged to .3. Most of these were correlated with unsaturated-saturated
fat ratio components but there were also metabolites that were associated with fruits and vegetables,
fish and nuts. And again, this shows us we had fairly good
representation of many of the components that go into scoring the index. For the Baltic Sea diet, there were 33 associated
metabolites. 10 of these were known metabolites. The absolute correlations ranged up to about
.2. Again, most metabolites were linked with polyunsaturated
to saturated and trans fat ratio but there were also metabolites associated with vegetables,
whole grains, fish and reducing fat percentage.>>[13:36] For the healthy diet indicator
there were 23 associated metabolites, 11 of those were known and the correlations ranged
up to about .2 again. As a reminder, the healthy diet indicator
components are based primarily on macro and micro nutrients versus whole foods. We saw correlations with polyunsaturated fats
and fiber but not the other components. So clearly you can see that dietary patterns
that are based on whole foods are associated with more diverse metabolite profiles compared
with those based on macro and micro nutrients.>>Here are the metabolic pathways that were
associated with each diet index with Bonferroni-correction. All of these pathways were significant at
that Bonferonni level. On the Y axis we have the metabolic sub pathways. On the X axis we have the number of measured
metabolites within the pathway overall in red, and then the number that was statistically
significant within that pathway in blue. A common theme here is that lysolipid and
food or plant xenobiotic pathways contained most metabolites that were associated with
diet quality. The lysolipid pathway includes many lysophospholipids
that are involved in things like cell signaling, energy metabolism and cell membrane integrity
and the food and plant xenobiotics include many plant phytochemicals, for example.>>Pathway analysis—metabolic pathway analysis—
really provides us with greater power to detect associations than looking at the individual
metabolites. Associations might be seen for an overall
metabolic pathway but not the individual metabolites. For example, the pentose pathway, which in
this case includes fruit-related threitol and xylonate, but not these individual metabolites
was associated with the healthy eating Index as an example. The same was true for histidine metabolism.>>Three of the diet indices had unique metabolic
pathway associations. For example, chemical metabolism was unique
to the healthy eating Index. This pathway includes, for example, 2-aminophenol-sulfate
which is a whole-grain biomarker. The ultimate Mediterranean diet score was
uniquely associated with dicarboxylate metabolisms and the Baltic Sea diet was uniquely associated
with benzoate metabolisms. You can see some of the associated metabolites
listed there on the slide.>>[16:32] Now, I showed you that each diet
index had a unique metabolite profile but there was some overlap which is listed here. We saw a common candidate biomarkers are fruits
and vegetables, fish, fatty acid ratios and whole-grains. Common metabolic pathways across the diet
indices included lysolipid and food and plat xenobiotic as well as essential and polyunsaturated
fatty acid metabolisms.>>In sensitivity analyses there were no differences
when we additionally controlled for cotinine and there were no differences between cases
and controls.>>We were able to replicate many metabolite
associations with specific food components that have been reported in the literature
in prior studies and some of those are shown here. Most of these prior studies also used the
Metabolon Inc. platform. We also have some potentially novel observations
shown here. For example, galactonate is the milk sugar
metabolite associated with dairy and homostachydrine which has previously been extracted from alfalfa
and coffee beans and associated with whole grains and fiber.>>[17:58] In another previous analysis of
the prostate, lung, colorectal and ovarian cancer screening study, five serum metabolites
were associated with the healthy eating Index. Most were vitamin related metabolites. Threonate, but not the other metabolites,
was associated with the healthy eating Index in the current analysis. This could potentially reflect population
differences and dietary intake, for example these two populations had very different intakes
of vitamin D supplements at baseline. Another previous study, the PREDIMED profile,
looked at metabolite profiles of Mediterranean diets. Again our results differed. The PREDIMED trial measured their metabolites
using NMR, and the Metabolon platform used liquid and gas mass spec, so we don’t have
complete overlap of the metabolites that are actually measured. There may have been a population difference
in the dietary intake such as olive oil use for example their study showed strong associations
with oleic acid, which is high in olive oil, and olive oil intake would have been extremely
low in Finland in the 1980s.>>This first study has a number of strengths
that build on prior studies including a larger sample size, a large number of identified
metabolites. We use strict control for multiple comparisons
and the metabolites were measured using fasting blood samples. A limitation was that the generalizability
might be limited since the study population was all male smokers. Although, we have been able to replicate many
of the diet component metabolite associations and other studies among women non-smokers.>>In conclusion, we found diet quality indices
representing healthy dietary patterns are associated with serum metabolite levels. Our results suggest while there are commonalities
there is not a single unifying metabolic association to a healthy diet; rather the metabolite profiles
of the diet indices that we evaluated were strongly associated with the diet index components,
and these components reflect different underlying constructs.>>[20:30] On replication, future studies
could apply these findings to gaining insights into the mechanisms that drive the health
effects of diet quality. Metabolomics could help us define and construct
new future diet quality indices that are likely to have a metabolic impact and finally, we
identified several candidate diet component biomarkers, so future studies could potentially
evaluate metabolite char— those particular metabolite characteristics and there measurement
error properties to determine if they have utility as dietary biomarkers.>>Now I’m going to describe another analysis
where we looked at nutritional metabolomics and breast cancer risk in a prospective study.>>For the second study, the objective was
to evaluate associations of diet-related metabolites with risk of breast cancer in the prostate,
lung, colorectal and ovarian cancer screening trial cohort, including estrogen receptor
subtype specific analyses. We used data from a nested case-control studies
within PLCO, which was a large population-based randomized screening trial. The postmenopausal women recruited for the
study were from the screening arm of the trial. There were 621 incident invasive breast cancer
cases in 621 controls matched on age, month and year of blood draw, year diagnosis and
hormone use. Pre- diagnostic non-fasting serum metabolites
were measured again using the metabolon platform. In this analysis, we focus specifically on
617 metabolites of known identities, so named based on chemical standards. The diet was measured at baseline using a
food frequency questionnaire, and we used an agnostic approach and we created 54 dietary
variables and measured the healthy eating Index –the 2010 version.>>[22:46] The outcome— The breast cancer
outcomes were ascertained using a number of methods that included state cancer registries
and annual questionnaire with medical record follow-up if breast-cancer was indicated. For the analysis, to identify the diet-related
metabolites, first we ran partial-Pearson correlations to measure the associations between
serum metabolites in the dietary items, again with Bonferroni correction.>>Any metabolites related to diet at the
Bonferroni level were carried forward to breast-cancer analysis. We used conditional logistic regression to
measure associations between these diet metabolites and breast cancer. The comparison was the 90th versus the 10th
percentile of metabolite levels. We also used principal components analysis
to identify multiple metabolite component profiles of diet exposures and then we carried
these components forward to measuring their association with breast cancer.>>The women in this study were an average
age of 64. The majority were never or former smokers,
current drinkers, not diabetic and overweight or obese. Altogether 34 of our 55 dietary exposures
were associated with at least one metabolite. 113 metabolites were associated with at least
one food or diet exposure, and this translated into 222 total correlations which ranged up
to an absolute correlation of about .7, which is reasonable for a biomarker study for self-reported
diet. We would anticipate higher correlations based
on the feeding study, for example.>>By far, we observed the most metabolite
associations with coffee intake. Many of these observations that you can see
here on the site are actually replicated from prior studies. You can see here that we were able to replicate
many other diet related metabolite associations from previous epidemiological studies as well
and we also found some potentially novel candidate biomarkers. I’m defining potentially novel metabolites
as those not having been reported on in the epidemiological literature to our knowledge. Here you can see at least 52 replicated to
our knowledge and there were many more potentially novel associations.>>[25:31] For example, theanine, an amino
acid extracted from tea-leaves that has been studied in vitro in the past, was correlated
with tea intake. 3-hydroxypyridine-sulfate, which is a phytochemical
found in coffee, was correlated with coffee intake. This plant phytochemical was associated with
citrus and other fruits. All of these candidate diet biomarkers are
exogenously derived phytochemical constituents so that provides us with a biological rationale
acting as diet biomarkers.>>Here are the alcohol-related metabolites
that replicated and they represent breakdown products of the alcohol but also changes in
endogenous metabolism with alcohol drinking. Specifically many metabolites that correlated
with drinking alcohol related to androgen metabolism.>>Now I’m moving on to the breast-cancer
analysis which measured the association between those 113 bonferroni-significant diet-related
metabolites and breast-cancer risk. Overall we found three metabolites to be associated
with overall breast-cancer, and that’s at the false discovery rate of .2. This slide here is color-coded by metabolic
super pathways. Two of those metabolites are in the lipid
pathway and one is a cofactor or vitamin.>>Here I am presenting the top 20 diet related
metabolites that were associated with overall breast-cancer in our multivariate analysis. We ran these analyses with and without control
for body mass index—this is including control for BMI. Only the top three metabolites were statistically
significant at a false discovery rate of .2. The top metabolite, caprate, was related to
consuming butter. Caprate is a medium chain saturated fatty
acid that’s found in dairy fat. The second top-ranked metabolite associated
with breast cancer was a gamma tocopherol excretion product, or gamma CEHC, this is
a form of vitamin E, the gamma tocopherol, found in vegetable oil. And thirdly an alcohol-associated metabolite,
an adrenal steroid precursor.>>[28:05] When we stratified by estrogen
receptor subtype, there were 19 diet associated metabolites that were statistically significantly
associated with ER+ breast cancer and you can see that most of these are in the lipid
pathway, indicated in blue.>>Here we see 19 metabolites were statistically
significant and the effect estimates ranged up to an odds ratio over two [28.31]. There were 12 alcohol associated metabolites
and most are related to sex steroid hormone metabolism.>>Specifically androgen pathway metabolites
and metabolites downstream of DHEAS. This overall pathway generates estrone and
estradiol, which are well-known to be associated with breast cancer. Less is known about the androgens. Alcohol appears to influence androgen metabolism
along this pathway with downstream cascading effects.>>There were three metabolites associated
with ER+ breast cancer that are related to fat containing foods specifically caprate
that I mentioned previously, as well as two others that were associated with consuming
dairy fats and fried foods. Two vitamin E related metabolites were also
associated with ER+ breast cancer: gamma-CEHC and delta-tocopherol. The greatest dietary sources of gamma and
delta tocopherol in the US are vegetable oils and margarine. It is possible that the gamma-CEHC here is
related to fats used in baking and frying. Here, since the dessert variable included
cakes, cookies, donuts and pies. And in other analysis at the NCI we previously
found that gamma CEHC may be a marker of fried food intake.>>[30:05] Another tocopherol isomer- the
alpha tocopherol, which is correlated with vitamin E and multivitamin supplement use—was
inversely associated with ER+ breast-cancer.>>There were no metabolites associated with
ER‑ breast-cancer at a false discovery rate of .2. I will point out where only had 144 ER- cases
as opposed to 418 ER+ cases, so that might have influenced our power to detect significant
associations for this subgroup.>>Here is a gaussian graphical model of how
the diet-related metabolites relate to each other conditional on the presence of the other
metabolites. The results here are overlaid with the breast-cancer
results. The metabolite names highlighted in pink were
positively associated with ER+ breast cancer and the ones in purple were inverse associations. Clearly, the metabolites related to breast-cancer
fell into three distinct metabolic networks. Looking now at some of the top metabolites
associated with ER+ breast cancer, we ran cubic splines to assess for linearity. And all the top hits were linearly associated
with ER+ breast cancer and here you can see the relationship.>>There were no differences in metabolite
breast-cancer associations by follow-up time. In a series of sensitivity analyses we also
adjusted for total cholesterol to see if findings changed for tocopherol metabolites since levels
of cholesterol affect levels of circulating tocopherols but we did not see any material
differences. We also adjusted alcohol-related steroid metabolites
for the non- steroid alcohol-related metabolites and vice-versa. The steroid metabolite ER+ breast cancer associations
were attenuated to some degree, but they still remained statistically significant.>>[32:16] Next we looked at the 113 diet-related
metabolites and how they grouped into principal components. The PCA analysis derived three diet-related
metabolite components which we named coffee, including 31 metabolites, healthy eating index
and multivitamins which included 23 metabolites, and alcohol, which included 9 metabolites. Of these only the alcohol component was associated
with overall and ER+ breast cancer but not ER- breast cancer.>>Some prior studies support that androgen
metabolites may mediate the alcohol-breast-cancer association or point in that direction. This includes a feeding study that provides
evidence for a causal association between alcohol and pre-androgens as well as pooled
meta- analyses showing that pre- androgen metabolites are associated with breast cancer. Androgens combine to the androgen receptor
in breast tissue and they can trigger breast cell proliferation. They can also be converted to estrogens by
aromatase in the breast and we know that estrogens are well known to promote breast carcinogenesis. We also found that dietary-fat-related metabolites
were associated with breast cancer. The role of dietary fat in breast-cancer has
been very controversial—it’s been a controversial question for many decades. The VITAL cohort more recently found that
the relationship between dietary fat and breast-cancer might differ depending on the fatty acid type. In particular, they also saw a relationship
with saturated fat. Many studies had previously only looked at
total fat intake which could potentially have attenuated the association. We also found that parent tocopherols and
their metabolites were associated with breast cancer and we saw both positive and inverse
associations but that depended on the tocopherol isomer. There is a literature on the effects of the
different forms of tocopherol on the hallmarks of cancer metabolism, but unfortunately the
reports are contradictory for both animal and human studies.>>Here is a graphic and it is exploring some
potential reasons why we may have seen a positive association of certain tocopherols with breast
cancer. One example is that we might be seeing an
overall diet quality effect because both gamma CEHC and delta-tocopherol were inversely associated
with the healthy eating index score.>>[35:01] There were several strengths and
limitations of this study. The major limitation here was that we were
possibly underpowered for the subtype analyses, particularly for the ER- breast cancer subgroup
which had one third the sample size of the ER+ group.>>To summarize, the findings for the second
studies support that there is a role for diet in breast cancer etiology, particularly for
ER+ breast cancer. Our biomarker results implicate the androgen
pathway as a mechanism of alcohol-induced carcinogenesis. They also suggest a role for fatty acid and
tocopherol metabolism.>>We plan to replicate this analysis in a
new data set from the cancer prevention study where we measured circulating metabolites
in a larger study sample of 782 cases and the same number of controls. I would just like to thank all of my collaborators
on these two studies. Dr. Zanetti: All right, I would like to thank
Dr. Playdon for a very interesting and informative webinar. I have several questions. Let me pull them up. Very quickly I just want to change what we’re
doing here -really quickly, so we have the right display up. Okay, wonderful.>>36:40. So, the first question is where are the diet
metabolomics data for the Finnish males archived—Metabolites or Metabolomics Workbench — so the community
can use the raw data for further meta- analyses? Dr. Playdon: [36:54] At this point the raw
data has not been archived. I’m not actually sure on the protocol of NCI
for releasing those data. So, I would have to get back to you on that. Dr. Zanetti: This is Krista Zanetti, we can
talk about it with you if you’re interested—if you and Dr. More? are interested, we can guide
you. I guess that will be “to be determined”
but thank you for the question. I think it is a very good one, for us in general,
the community to be thinking about depositing our data. OK, next question: Why was the metabolite
abundance not age-corrected given that the males are in an age range? Dr. Playdon: so, the Metabolon data we received
is based on relative intensities. And so, we controlled for age in all the analysis,
but we are not looking at quantitative abundance data. Dr. Zanetti: All right. Okay, next question. As all of these data are from untargeted LC
MassSpec and GC MassSpec platforms, how were false positives identified and discarded from
the data from Metabolon? Dr. Playdon: Do you mean false-positive associations? Dr. Zanetti: It says here: how are false positives
identified and discarded from the data? Dr. Playdon: Well, statistically we tried
to adjust for multiple comparisons, unless you are referring to whether we excluded certain
metabolites– various criteria. Dr. Zanetti: Whoever asked that question if
you want to put any clarifying points down, I can eventually get to that. I’m not sure if you do have them, but if
you type it in I can try to get back to that, as we get through some of the questions.>>Okay, the next question is: Functionally,
is lineolate in the diet beneficial particularly with respect to cancer risk? Dr. Playdon: [39:33] Functionally… So, as far as metabolomics data and breast-cancer
is concerned, there isn’t a lot of data available. In terms of cancer in general, I’m not actually
sure. I would have to get back to you on that. Dr. Zanetti: Right, okay. So, the next question, could you reiterate
whether the Finnish samples had been frozen and stored before analysis and how stable
these different metabolites are in stored samples? Dr. Playdon: Yes, they have been frozen and
stored – I mean the study- we’re using baseline pre- diagnosis samples so they’d
been stored, actually, for quite a long time. The study was conducted in the 80s. I don’t actually have quality control– access
to quality control data specifically for that. Dr. Zanetti: Next question. Could you reiterate what the absolute concen—sorry,
my apologies. Could you reiterate what the absolute correlations
represent and why they are so often less than 0.5? Dr. Playdon: So, a lot of the diet correlations
you saw, particularly for the dietary pattern analysis were on the range of 0.2 to 0.3,
so there is measurement error inherent in self-reported dietary questionnaires, and
this contributes to, potentially, attenuation of these associations. We are planning to do some follow-up studies
where we are looking at this diet-metabolite relationship based on feeding study data. So we will be able to get more of an accurate
idea of true diet metabolite correlations based on gold standard measures of diet. Dr. Zanetti: [41: 44] The next question is
actually for me, not Dr. Playdon, so we’re gonna give her a break, but I’m going to answer
it because I have gotten this question several times. I was asked where you all can watch the prior
metabolomics webinar presented by Dr. Playdon, if it was recorded. I just want to clarify that the previous presentation
by Dr. Playdon wasn’t seminar form here at the National Institute of Health. But because we got so many inquiries about
having it be broadcast via webinar, she presented the exact same thing today on the webinar
so the extramural community and people outside of NIH can see it. So what you saw today was more or less what
she presented, the same data, I think it was the exact same presentation as was presented
in that seminar, so if you were on the webinar today you have seen that seminar. So I wanted to answer that so you all know. On to the next question. Why use Bonferroni correction which is more
conservative as compared with Benjamini-Hochberg, which controls better for false discovery
rate? Dr. Playdon: We chose to be more conservative
with the diet Associations because we had run some prior studies where we had seen the
Associations. We’ve now replicated them over at least
three different study populations. We used the FDR rate with the breast-cancer
analysis since that was agnostic as well, but nothing had been done previously in that
area so we relaxed the criteria slightly. Dr. Zanetti: Great, thank you. Next question. Have you thought about doing the same analysis
in populations at low risk of breast cancer so as to identify risk-lowering metabolites? Dr. Playdon: I haven’t thought about that,
but that is something to think about. Dr. Zanetti: Thank you, Dr. Playdon. Next question. Did you look at the effects of smoking in
the diet and breast-cancer study? Dr. Playdon: [44:01] We controlled for smoking
in the analyses. We also used smoking biomarkers, for example,
cotinine, for additional adjustment to see if it made any difference in the diet-metabolite
associations, and it did not. Dr. Zanetti: Okay, so I’m going to jump back
to clarification regarding the functional question. In follow-up to the functionally relevant
question, we have observed the use of radiation to treat diets causes the appearance of many
lineolate metabolites including C8, C10, and C12 oxidized fatty acid. These, therefore, might be coming from food
sources other than butter. Dr. Playdon: Okay, thank you. Dr. Zanetti: Thank you. I’m just going to move back up, I’m trying
to manage a lot of questions here, so bear with me. Oh, okay. Considering those metabolite biomarkers, what
about specificity and sensitivity of them? Like a raw curve? Dr. Playdon: Yeah, that’s another approach
that we’ll be taking in our next analysis and doing more predictive analyses as well. Dr. Zanetti: Okay, next question. The alcohol Association looks a bit puzzling. Individuals that report drinking no alcohol
often do so because they suffer from severe disease. How would your associations look if you excluded
the tea-totallers? Dr. Playdon: I think they mean extreme alcohol
drinkers. That is something we didn’t specifically exclude,
extreme alcohol consumers, but generally the alcohol consumption wasn’t extreme in this
population. Dr. Zanetti: [46:10] Okay, next question-
you guys are just not going to give Dr. Playdon a break here now that you have her for the
next ten minutes! But you have her for the next ten minutes… Are you considering the role of the gut microbiome
in your research. There is a lot of chemistry and biochemistry
between the mouth and the blood and all the action in between will determine which food
components or their metabolic products show up in the blood. Dr. Playdon: I think that’s a good point and
that’s definitely a really hot area for future research particularly with the tocopherol
findings. I’m interested in following up on that. Dr. Zanetti. Okay. Somebody is asking, remind me when the diet
metabolites were analyzed, how long after storing? Dr. Playdon: So, it would have been many years
for the ATBC study that was conducted in the 80s and the analyses were done around 2009ish. So yeah, there are storage issues. A lot of his quality control issues we really
do need to explore in the field of Metabolomics to find out the degradation over time. It’s an important thing to consider. I don’t know specifically in this case the
effects on the metabolites, but even though those samples were stored for a very long
time, we were still able to replicate a lot of associations in that study and the PLCO
study, which those samples were not stored nearly as long. Dr. Zanetti: [47:58] K, onto the next question. Do you see any difference in the correlation
between metabolite and dietary quality index in the PLCO and ATBC studies: one used nonfasting
blood and one used fasting blood. Dr. Playdon: Yeah, so, the PLCO study, there
was one nested case-control study which was a lot smaller and I think I mentioned in that
one slide there was a little bit of overlap but not a whole lot of overlap. So, sorry, remind me of the question? Dr. Zanetti: The question was: Do you see
any difference in the correlation between the metabolite and the dietary quality index—
Dr. Playdon: Oh, yeah. The magnitude of correlations were similar. The actual metabolites that came out were
different somewhat, which could have something to do, as I mentioned, with the differences
in populations intakes. Also the metabolites were measured at different
time points and the sensitivity of the platform changes over time so the newer platforms pick
up a lot more metabolites as well. It does make it difficult to compare across
studies that have been measured at different times. Dr. Zanetti: K next question. Given all the known associations between carotenoids
and cancer risk I’m surprised there weren’t any carotene cancer relationships which emerged. Can you comment on whether the Metabolon analytical
platform used analyzed lipid-soluble components like carotenoids? Since the tocopherols were determined it seems
the extraction method and analytical method may have been able to measure lipophilic phytochemicals
like carotenoids. Dr. Playdon: [49:53] Yeah, I think a lot of
the carotenoids that have been reported in the literature, we did not have in our metabolite
profile, so we weren’t able to assess that. Dr. Zanetti: Can we use canonical correlations
multiple to multiple rather than one to one in the study? Dr. Playdon: I’m not sure about that. Sorry. Dr. Zanetti: That’s okay. You’ve gotten a ton of questions, so…
okay. Are all the metabolite IDs confirmed by MFMS? Dr. Playdon: Metabolon has its proprietary
library of chemical standards for identification, so its proprietary. Dr. Zanetti: Alright. Okay, and then, I think this goes back to
the alcohol question which would be… I think the previous questioner… I think it was the… they meant some people
may report that they don’t drink alcohol because they can’t due to interactions with certain
drugs, etc., so how would the results change if you excluded the nondrinkers? Dr. Playdon: That’s something we could look
at. I didn’t exclude nondrinkers. This was more of that agnostic analysis looking
at all of the dietary exposures, but it’s something, you know, if we were looking specifically
at alcohol as an exposure, where we could do a series of sensitivity analyses to see
how things changed. Dr. Zanetti: Okay. Okay, alright…. Next question. What are the advantages and disadvantages
of using Metabolon’s platform versus one of the NIH Regional Comprehensive Metabolomics
Resource Cores or university lab with adequate resources? Dr. Playdon: We’ve used Metabolon here for
a number of studies. They do have a very extensive chemical reference
library so we are able to get quite a large number of identifiable metabolites very quickly,
so that’s the benefit to this platform. I think there are pros and cons to using different
platforms that is probably beyond the scope of this talk. Dr. Zanetti: Okay, so, I don’t have any more
questions popping up in the feed. However, I do want to give people at least
another minute to ensure there are no additional questions because people could be typing so
I do want to… I’m just going to pause for a second if
anyone is typing and then we can ask final questions. We have about four minutes. Okay, we got one. Okay. Okay. How do you suppose that in future studies
that we are going to be able to account for dietary intake of participants and measuring
the metabolites and dealing with variability in the participants due to gut microbiota—I’m
sorry— to make dietary suggestions for certain health outcomes, do we need to know how many
different individuals may respond to different diets and foods, for example, two people consume
the same day and produced different metabolite outcomes. This may even be the case among healthy adult
comparisons. So there two questions there so if you need
me to restate any of it let me know. Do you want to start with the first? I’m going r— go ahead. Dr. Playdon: I think one of the previous question
is kind of alluded to this but it is definitely something of interest for follow up in future
studies to look at the effect of the microbiome on the variability of these metabolite levels. We did not have this data, so I couldn’t
look at that. But that is definitely something that could
be followed up on in the future. Dr. Zanetti: The second part was to make dietary
suggestions for certain health outcomes, do we need to know how different individuals
may respond to different diets, two people consume the same type of produce different
metabolite outcomes. This may even be the case among healthy adult
comparisons. Dr. Playdon: yeah, as research—or as guidelines
take a more personalized and individualized approach, it sounds like a promising area—the
microbiome— to explore some of those differences. Dr. Zanetti: [54:55] The network diagram went
by quickly, how are they generated—Are the distances related to the false discovery rate? Dr. Playdon: So, we did a GGM analysis and
then we visualized it in cytoscape and the edges were connected. There were metabolites that had direct relationship,
or direct correlations, conditional in the other metabolites of at least .2. Dr. Zanetti: Okay, was there accounting for
stability of these metabolites during the course of your analyses? For instance, were pooled quality control
samples analyzed with patient samples?>>Yes, QCs were inserted. Technical replicants and pulled QCs and we
ran ICCs and everything.>>Okay. I have one more quest—Up, that was the last
question and we are at 1:59, and for anyone that knows me, I like to stay on time, so
we’re going to do that today. I would like to thank you all for your attention
and active participation in today’s NIH Metabolomics Interest Group webinar. I think that this was a fantastic number of
questions and clearly people were engaged which is exciting for us since this is only
I think our third or fourth—third, I think—webinar so I’m really excited to see that and that
we are bringing the community together to have the opportunity to discuss some of these
things via this mechanism. I especially would like to take the opportunity
today to thank our presenter today for an excellent presentation and hanging in there
and answering more questions than I think I have ever seen anyone have to answer at
the end of a webinar. We would welcome your feedback to inform future
webinars and again thank you all very much for your participation.>>[Event Concluded]