Introduction
In the light of our long co-evolution with microbes [1 ], the human microbiome can be seen as an
accessory genome including diverse bacteria, viruses, archaea, and fungi, expanding
over several bodily niches and extending our functional potential. While most
bacteria are commensal, many inhabiting our body are mutualistic, and some are
detrimental. They are kept in check by the immune system and the surrounding cell
communities contributing to the compositional niche, whose survival depends on the
livelihood of the entire “human meta-organism”. After a period of
what can be considered a prolonged cold war with bacteria, brought on by the advent
of the hygiene hypothesis [2 ] and the
golden era of the antibiotics, growing evidence of microbial “ecosystem
services” [3 ], the inextricable
link between our health and our microbiome, including efficacy and response to
medication such as cardiac drugs and chemotherapeutics, have led to a rekindling of
interest in mucosa-associated bacterial microbiomes [4 ]
[5 ]
[6 ].
For the last two decades, this area of research has been propelled forward by several
consortial efforts, including MetaHit [7 ],
Human Microbiome Project (HMP) [8 ],
Metacardis [9 ]
[10 ], and Flemish Gut Flora Project [8 ]
[11 ], in which the intestinal microbiome and, in a minority of cases
(HMP), the oral, dermal and vaginal microbiome have been sequenced in large cohorts
to define: a) a “healthy microbiome state”, b) factors contributing
to microbiome composition (including medication, nutrition, and environment)[11 ]
[12 ]
[13 ], and c) the
associations of these microbiomes with disease [7 ]
[14 ]. This has been further
facilitated and expedited by the introduction of whole metagenome shotgun sequencing
as well as accessible and scalable analyses methods. These population studies have
linked dysbiosis – a state of deviation from a core functional and taxonomic
composition- with the pathogenesis of several diseases, including obesity,
non-alcoholic fatty liver disease, and cardiovascular disease [15 ]. While a transient presence of bacteria
in the blood has been described after teeth brushing or gingival manipulation [16 ]
[17 ], their consistent presence has been interpreted as indicative of
infection or sepsis. This notion changed only recently with the description of low
grade or subclinical inflammation in metabolic disease.
Low-grade inflammation is consensually perceived as a potential trigger for metabolic
dysfunction, insulin resistance, and diabetes [18 ] and is often reflected in subtle increases in systemic inflammatory
markers such as C-reactive-Protein (CRP). Increased translocation of
lipopolysaccharides (LPS), mainly from the gut to the blood, defined as endotoxemia
[19 ], has also been described as a
potential trigger for chronic inflammation in metabolic disease, supported by the
onset of insulin resistance in healthy subjects after LPS-transfusion [19 ]
[20 ]. Therefore, it is surprising that the notion of a core blood
bacterial signature in health and its divergence in disease with accompanying
systemic response remains contentious. However, converging evidence for microbial
signatures in metabolic tissues, including the blood, suggests a novel paradigm for
metabolic disease development [19 ]
[20 ]
[21 ]
[22 ]
[23 ]
[24 ]. In this review, we address the potential clinical relevance of a
blood bacterial signature pertinent to disease development and outcomes and new
avenues for translational approaches. We, moreover, discuss pitfalls related to
research in low bacterial biomass while suggesting mitigation strategies for future
research and application approaches.
Schrödinger’s blood bacteria: ‘you don’t see
me, now you do.’
The concept of bacterial presence in the blood has been reported as early as
1969, where living and metabolically active bacteria [25 ] were localized in the blood of
healthy human subjects. In the meantime, a commendable body of work has combined
classic molecular and microbiological techniques to evidence the presence of
bacterial cells and sequences in the blood of presumably healthy humans, either
as a study cohort or as healthy controls for specific disease states (recently
reviewed in [26 ]). These methods
include quantitative polymerase chain reaction (qPCR) [27 ]
[28 ]
[29 ]
[30 ]
[31 ]
[32 ]
[33 ]
[34 ]
[35 ] of 16 S
rRNA or targeted bacterial genes [29 ]
[35 ], fluorescence
in-situ hybridization (FISH) and other employment of fluorescent probes [35 ], transmission electron microscopy
[36 ], dark field microscopy[36 ], PCR followed by electrospray
ionization-mass spectrometry [37 ] as
well as classical bacterial culture [32 ]
[38 ]
[39 ]. Several of these results have been
met with criticism, considering the long-standing dogma of the blood being an
innocuous environment and controversy regarding visual confirmation of bacteria
deemed by some authors as L-form bacteria and by others as merely membrane
vesicles or aggregated proteins [40 ].
The advent of efficient and scalable technologies, including next-generation
sequencing of the 16 S rRNA gene (present in all bacterial cells), whole
metagenome sequencing, RNA-sequencing [38 ]
[40 ]
[41 ], and the ongoing expansion of
microbial reference genomes have facilitated the detection of non-culturable or
hard-to-culture bacteria across several tissues and diseases. This is especially
important considering several blood-borne bacteria can persist in a dormant
state [42 ]. Moreover, the
democratization of analytical tools, including the development of microbiome
tools with friendly graphic user interfaces [43 ] and comprehensive documentation [44 ], have made this research area
accessible to a broader scientific community.
Blood bacterial signatures in health and cardiometabolic disease
While the role of systemic inflammation in cardiometabolic disease, partially
induced by bacteria, has been widely appreciated, most of the published studies
related this observed systemic inflammation to bacterial components
and/or selectively analyzed surrogate parameters. This includes
measurements of host response patterns (e. g., LPS binding protein
(LBP)) or wall components of gram-negative bacteria (LPS) in the circulation to
assess bacterial burden indirectly. However, there is scarce but increasing
research investigating direct bacterial presence, that is, bacterial cells or
bacterial nucleic acids, and its quantitative and qualitative potential to
contribute to inflammation and dysfunction of both the immediate
microenvironment and the whole system. Several studies have evidenced the
presence of bacterial genetic material, whole bacteria, or even bacterial RNA in
the blood of healthy subjects [26 ]
[42 ] ([Fig. 1a ]), and research dissecting the
contribution of circulating bacteria or bacterial genetic material to
cardiometabolic disease has been accumulating steadily ([Fig. 1b ]). From a clinical
perspective, it has been noted early on that bacteremia of unknown origin was
more prevalent in subjects with type 2 diabetes (T2D) compared to non-diabetic
subjects [45 ], with higher occurrences
of Staphylococcus aureus and Klebsiella in the blood along
relatively ‘harmless infections’ such as urinary tract
infections, leading to septic shock more often (OR 1.9) [46 ] and motivating the early
application of targeted empiric antibiotic treatment. A hallmark study was
published in 2011 by Amar et al., analyzing blood bacterial load by measuring
16 S rRNA copy numbers in more than 3000 subjects followed over nine
years. The authors report an increased abundance of Proteobacteria in visceral
obesity as well as onset of T2D [30 ]
and cardiovascular disease at follow-up [47 ]
Fig. 1
a ) Overview of phyla proportions, controls inclusion, and the
number of included subjects in selected study cohorts reporting
blood bacterial signatures in healthy cohorts and healthy subjects
of cohorts with cardiometabolic and cardiovascular disease .
Names of the studies are to the left of the bar. Bar length equals
100% (total proportions of reported phyla), and each color in
the bar represents a specific phylum (color code specified in the legend
to the right). Controls: specifies whether experimental and collection
controls were included to account for contamination. A red box
corresponds to a lack of controls, a green box to the availability of
adequate controls. Ref.: refers to the reference number in the
bibliography (Paisse and Gosiewski et al. are reviewed in ref [25 ]) and N the number of
healthy subjects in the cohort. An average distribution for all studies
is given in the lowest bar, and N on the right refers to the total
number of subjects included. b ) Overview of phyla proportions,
included controls, and the number of included subjects in selected
study cohorts reporting blood bacterial signatures in disease
(focusing on the cardiometabolic disease). Names of the studies
are to the left of the bar. Bar length equals 100% (total
proportions of reported phyla), and each color in the bar represents a
specific phylum (color code specified in the legend to the right).
Controls: specifies whether experimental and collection controls were
included to account for contamination. A red box corresponds to a lack
of controls, a green box to the availability of adequate controls. Ref.:
refers to the reference number in the bibliography and N the number of
diseased subjects in the cohort. Conditions: refers to the medical
conditions of the cases in the studies. An average distribution for all
studies is given in the lowest bar, and N on the right refers to the
total number of subjects included. Abbreviations : NAFLD:
non-alcoholic fatty liver disease, CVD: cardiovascular disease, CHD:
chronic heart disease, STEMI: ST-elevation myocardial infarction.
In 2014, Sato et al. undertook a nested case-control study reporting that
prevalent T2D was associated with an increased inflammatory response with higher
levels of LBP independently linked to HbA1c and BMI. Higher abundances of
Clostridium coccoides , Atopobium clusters [31 ]
[47 ], Sediminibacterium spp.[48 ], and higher detection of the
16 S rRNA gene was positively associated with the disease [31 ]. This is particularly important, as
increased bacterial DNA detection after bariatric surgery in subjects with T2D
and obesity revealed a blunted response to the intervention with minor
improvement in glucose tolerance or inflammation [49 ]. In contrast, subjects with a high
abundance of the genus Bacteroides were less likely to develop T2D during
a two-year follow-up [48 ]
[49 ].
Likewise, there is increasing evidence for a link between circulating bacteria in
liver fibrosis and steatosis, which is associated with an impaired immune system
and glucose metabolism and major cardiovascular events [50 ]. In 2016, Lelouvier et al. reported
increased levels of circulating 16 S rDNA and proportions of
Proteobacteria (specifically Sphingomonas, Bosea, and
Bradyrhizobiaceae taxa) in liver fibrosis. This was further
substantiated by other studies showing similar results in liver cirrhosis in
several blood compartments [51 ] and
the increased detection of bacterial DNA via in situ hybridization in
particularly severe cases of liver decompensation [52 ]. Chronic kidney disease (CKD), a
well-established cardiovascular risk factor [53 ], has similarly been linked to changes in blood bacterial
signature, as it was associated with a decreased alpha diversity – a
hallmark of several non-communicable diseases – and the expansion of
bacterial taxa expressing uricase as well as indole- and p -cresyl-forming
uremic enzymes [54 ]. These bacterial
uremic toxins have been linked to worsening kidney function, endothelial
dysfunction, cardiac fibrosis, macrophage activation, and insulin resistance
[55 ]. CKD is also linked to a
dysbiosis in the gut, which has been associated with impairment of gut
epithelial barrier via depletion of claudin-1, occludin, and zonula occludens-1
proteins in addition to dysfunction of colonic T-regulatory cells [56 ] as well as further aggravation by
direct effects of uremic toxins [56 ]
[57 ]. Therefore, it is
not surprising that CKD is reflected in a decreased bacterial diversity in the
blood and associated with a higher abundance of proteobacteria, which were
inversely correlated with kidney function [58 ]. Comprehensive studies on the specific compartmentalization of
bacterial DNA in metabolic disease remain rare but promise a more holistic
approach. For example, Anhê, Jensen et al. compared the bacterial DNA
load and taxonomy in plasma samples, liver, subcutaneous, visceral, and
mesenteric adipose tissues, noting a relatively reduced bacterial load in plasma
samples compared to other tissues and a preferential presence of two specific
genera (Rodoferax and Polaromonas ) in the blood, but did not find
distinct differences in the tissues between subjects with and without T2D [23 ]. Results were supported and
expanded by findings from our research group, where 16 S rRNA gene
content was quantified and sequenced in 75 patients with obesity and with or
without T2D. Bacterial composition in the blood was associated with circulating
tumor necrosis factor-ɑ, CRP, LBP, and interleukin-6. Moreover, amounts
of Lactococcus in the blood correlated negatively with homeostasis model
assessment-estimated insulin resistance while Acinetobacter and
Tahibacter were found more often in the blood than in several studied
adipose tissues and correlated positively with diabetes status [21 ]. In another recent study, we
evidenced reduced bacterial diversity in the blood of subjects with T2D vs.
non-T2D, but a reduced richness after bariatric surgery paralleled with an
increased bacterial quantity. The latter was surprisingly negatively associated
with leukocyte counts. We also evidenced early shifts in taxa after surgery with
a reduction in Rhizobacter , Anoxybacillus , Streptococcus ,
and Ureibacillus. In our cohort, several of these genera were positively
correlated with inflammation, fat mass, and body mass index (BMI). Most
importantly, 40% of the variance in the blood bacterial composition
could be explained by 27 host variables, including medication intake [22 ]. Similarly, bacterial composition
predicted the clinical classification of patients according to metabolic disease
robustly. The strength of the last three studies is the extensive experimental
and bioinformatic contaminant control, which has since become more common in
recent studies [59 ].
At the other end of the cardiometabolic disease spectrum, infections in the
hospital setting as well as outpatient infections very often preceded coronary
heart disease (CHD) and stroke in the ARIC study, which analyzed cardiovascular
risk in 15792 subjects, showing that in a total of 1312 incident CHD and 727
incident stroke cases, infections were associated with an up to 12.83 OR for CHD
(inpatient infections, up to 14 days prior to CHD presentation) [60 ]. Increased strain can lead to overt
manifestations of cardiovascular events in predisposed subjects. This can be
exemplified in type 2 myocardial infarction (MI) in infection and anemia, where
an imbalance between myocardial oxygen demand and supply becomes apparent. That
being said, one can not preclude that systemic inflammation in the realm of
infection can further trigger immunological processes in local sites of
preexistent vascular lesions in the heart or carotid arteries. In line with a
potential mediating effect of blood bacteria and increased bacterial
translocation from the gut, Zhou et al. demonstrated higher levels of LPS and
D-lactate (bacteria-derived) in the blood of subjects two days post-MI along
with an increased alpha diversity in ST-elevation MI (STEMI) and enrichment of
gut-derived bacteria. At the same time, chronic CHD presented with lower
bacterial diversity in the blood compared to controls. Mediation analysis
revealed a mediating role for bacterial translocation in inflammation (CRP and
monocytes), left ventricular ejection fraction, and major adverse cardiovascular
events. More importantly, in an experimental MI mouse model, the reversal of
bacterial translocation via antibiotic treatment reduced serum LPS, alleviated
monocytosis, and reduced cardiomyocyte injury post-infarction [61 ]. More recently, Amar et al. showed
that subjects with an acute onset MI displayed a 1.3-fold increase in blood
bacterial load compared to control subjects with concurrent metabolic risk
factors, which was driven by high LDL cholesterol. In the 99 subjects with MI,
several genera known to metabolize cholesterol, such as Nocardiaceae ,
Aerococcaceae , Gordonia , Propionibacterium ,
Chryseobacterium , and Rhodococcus , were depleted [62 ]
. In conclusion, most studies
found a similar pattern of phyla distribution in the circulation, which was
dominated by Proteobacteria (~65%), followed by Actinobacteria
(16%), Firmicutes (10%), and Bacteroidetes (4%) ([Fig. 1 ]). With some exceptions,
changes in cardiovascular disease (CVD) followed the same trend, showing a
relative increase of Proteobacteria by 7% and a 10% reduction in
Firmicutes on average ([Fig. 2 ]). As
depicted, individual studies show more considerable changes but are not always
congruent. No objective conclusion can be drawn on lower taxonomic levels, as
reported results vary greatly, but for example, Pseudomonadaceae and
Streptococcus were enriched in CVD in more than two studies ([Table 1 ]).
Fig. 2 Heatmap referring to the differential abundance [increased
(in red) or decreased (in blue)] of phyla in studies with disease vs.
healthy controls. Studies are given to the left of the heatmap rows;
phyla constitute the columns of the heatmap. A look at differentially
abundant genera in CVD in selected studies are shown in the red box (for
increased genera) and the blue box (for reduced genera) next to the
survey referred. Abbreviation: CVD: cardiovascular disease.
Table 1 Studies showing evidence of blood bacterial
signatures in cardiometabolic disease in human
Author
Method
Matrix
Conditions and number
Year
Reference
Rajendhran
16 S V3
Whole blood
133 CVD vs 118 controls, thereof positive for 16 S:
31 vs 10 (not matched)
2013
[153 ]
Dinakaran
Metagenomic, 16 S quantification
Plasma
80 CVD vs 40 controls for quantification, 3 vs 3 for
sequencing
2014
[29 ]
Lelouvier
16 S V3/V4
Buffy coat
Cohort 1: 11 fibrosis vs 26 controls; Cohort 2: 11 fibrosis
vs 60 controls
2016
[154 ]
Whittle
16 S V4
Plasma
5 atopic asthmatic vs 5 controls
2018
[38 ]
Schierwagen
16 S
Buffy coat
3 variceal bleeding, 4 refractory ascites
2018
[51 ]
Zhou
16 S V4
Buffy coat
100 ST-segment elevation myocardial infarction, 50 CHD, 49
controls
2018
[61 ]
Qiu
16 S V5/V6
Whole blood
50 T2D vs 100 healthy (matched for age, sex)
2019
[48 ]
Alvarez-Silva
16 S V3/V4
Whole blood and ascitic fluid
33 cirrhosis
2019
[155 ]
Shah
16 S V3/V4
Buffy coat
20 CKD vs 20 healthy (not matched)
2019
[58 ]
Amar
16 S V3/V4
EDTA blood
103 high CVD risk, 99 with myocardial infarction
2019
[62 ]
Anhê
16 S V3/V4
Plasma
20 T2D, obese vs 20 normal glucose tolerance, obese
2020
[23 ]
Massier
16 S V4/V5, CARD-FISH
EDTA blood
75 with obesity, 42 no T2D vs 33 T2D
2020
[21 ]
Chakaroun
16 S V4/V5, CARD-FISH
EDTA blood
64 at baseline with obesity, 24 no T2D vs 24 T2D with
12-month follow-up (matched for age, sex, and BMI)
2021
[22 ]
CARD-FISH: catalyzed reporter deposition – fluorescence in situ
hybridization; CVD: cardiovascular disease; CHD: chronic heart disease;
CKD: Chronic kidney disease; EDTA: ethylenediaminetetraacetic acid; T2D:
type 2 diabetes
Translational avenues: are we there yet?
High throughput sequencing of blood-borne bacteria and other microorganisms is
relevant as a potential extension of the repertoire of current assessments in
health and disease. It has already shown paradigmal potential for the gut
microbiome and promises a similar potential for blood-derived disease markers
[15 ] or liquid biopsies [24 ] in non-communicable and infectious
diseases. Moreover, it will be critical for downstream applications, including
novel, cost-effective techniques to detect new pathogens in an ever-evolving
environment [63 ]
[64 ]. Exploring functional potential
using highly granular whole genome sequencing and genome assembly paired with
metabolomics and metaproteomics will further underpin disease mechanisms [65 ], allowing targeted and possibly
personalized treatment options.
First things first: traditional applications for non-classical
diagnostics
A seemingly immediate application for the analysis of bacterial sequences in
the blood, along with other bodily niches, is the potential to precisely
identify bloodstream pathogens at the strain level. This helps account for
pathogenicity, transmission potential, and antimicrobial resistance [66 ] and define their source early
on in order to inform targeted interventions in clinical infection
management. Considering that the latter is in most cases dictated by
assumptions of infection source and the fact that culture-based methods are
slow and fail to identify potential pathogens in 40% of cases [67 ], this application seems
particularly central in subjects with sepsis [68 ] and high-risk subjects such as
immunocompromised patients and those after hematopoietic cell
transplantation (HCT) [69 ] or
subjects with liver cirrhosis [52 ], as well as patients with diabetes. In the latter, the higher
risk of primary bloodstream infections may arise from the translocation of
organisms from damaged mucosal sites like the oral cavity or the gut.
Additionally, it might be helpful in patients who are already treated with
antibiotics, and in whome blood cultures usually fail. A promising approach
was published by Tamburini et al., who developed a bioinformatic tool, which
utilizes the potential of whole-genome sequencing to profile strain
variations between metagenomes to match bloodstream pathogens in HCT to a
candidate source. They found concomitant dominant gut colonization with
known enteric pathogens a few days prior to bloodstream infection becoming
clinically apparent. Interestingly, they found cases where typically
non-enteric pathogens (e. g., Pseudomonas aeruginosa and
Staphylococcus aureus ) were present in the gut microbiota before
infection and ascertained functional relatedness by showing concordant
predicted and clinical antibiotic resistance [70 ]. While turnaround time in
sample preparation and sequencing and the high cost remain hindersome for
efficient integration of whole-genome sequencing in clinical care,
strain-level identification of pathogens and their reservoirs has the
potential to improve infection prevention by reconstructing the trajectories
and timing of colonization, the evolution of pathogenicity, and microbial
adaptation [71 ]. This can further
enhance management procedures by rapidly identifying resistance and escape
mechanisms in real-time [63 ].
Considering this manuscript was written amid the coronavirus 2 (SARS-CoV-2)
pandemic, it would almost be tone-deaf not to stress how metagenomics has
been pivotal in helping us develop a holistic ecological understanding of
microbial evolution over time and geography. This is particularly true for
detecting subclonal mutations (used routinely in cancer research), which
have helped delineate the chains of infection and interspecies transmission
and the higher frequency of which has been associated with severe acute
respiratory distress syndrome in SARS-CoV-2 [72 ]. The same holds for bacterial
infection: bacterial DNA in the bloodstream more effectively identifies
subjects at higher risk of death or more severe illness from suspected
sepsis [73 ], regardless of the
isolation of viable bacteria via blood culture alone [37 ]. While immunotherapy is
promising but practically nonexistent in cardiometabolic disease [74 ], the implementation of
screening for blood bacterial DNA might be beneficial in critically ill
subjects, where adjunctive immunotherapy [75 ] or treatment with specific
toll-like-receptor (TLR) antagonists is an option [76 ].
Bacteria in the blood: liquid biopsies of the new era?
As liver cirrhosis has been associated with a circulatory bacterial
signature, it is not surprising that hepatocellular carcinoma (HCC) displays
a specific alteration of the bacterial sequences found in the blood of
diseased subjects. Specifically, the blood of subjects with HCC had a lower
bacterial diversity. Moreover, significant differences in the relative
abundance of Pseudomonas , Streptococcus ,
Bifidobacterium , Staphylococcus , Acinetobacter ,
Klebsiella , and Trabulsiella were noted. The latter four
bacteria were heavily enriched in HCC, and Staphylococcus had the
most significant association with HCC
(p +=+4.0e-08), showing a 4.3-fold increase
compared to controls. The authors then identified a marker-based model
depending on five bacterial genera, distinguishing HCC from controls (AUC of
0.879, accuracy 81.6%). Validation in a subgroup of the cohort
confirmed that the model accurately differentiated HCC with an AUC of 0.875
and an accuracy of 79.8% [77 ]. Substantial contributions of local tissue and gut microbiome
have been demonstrated for several cancer types [78 ]
[79 ]
[80 ]
[81 ]
[82 ]
[83 ]
[84 ]
[85 ], which further motivates the
search for microorganism-derived molecules to diagnose noncommunicable
diseases such as cancers. In a recent publication, Poore et al. reassessed
whole-genome and whole-transcriptome sequencing studies in The Cancer Genome
Atlas accounting for over 30 types of cancer from 104,814 treatment-naive
subjects, detecting cancer-specific microbial signatures (viral and
bacterial) in the tissues and blood, and benchmarking signatures from
microbial DNA from the plasma against cell-free tumor DNA. They, moreover,
used deep metagenomic sequencing on plasma samples from 100 tumor patients
vs. 69 healthy subjects and showed that cell-free microbial profiles were
discriminatory between healthy and diseased subjects and between different
types and stages of cancers. This was only possible for specific cancer
types (colon, stomach adenocarcinomas and renal clear cell carcinoma), and
the microbial signatures failed to differentiate intermediate cancer stages,
suggesting that the microbial community structure might not be associated
closely with cancer stages. They further trained a microbial source tracking
algorithm and showed that the gut microbiome was most relevant for the
bacterial signature in cancers of the gastrointestinal tract, further
indicating that Fusobacterium spp. was overabundant in GI- vs.
non-GI-tumors. That being said, the authors did not replicate the bacterial
signatures found in HCC in other studies and relied on tissues more than
blood samples, which were very often used as negative controls and did not
generally show a differential microbial signature [24 ]. While bacterial signatures
have been associated with cardiometabolic disease and cancer, this has been
done in heterogeneous populations, and studies differed extensively in how
they processed samples and what analytical tools were used, making it
impossible to benchmark a ‘meta-population blood bacterial
signature’ for cancer or cardiovascular disease. Currently,
‘good old’ circulatory tumor markers and imaging studies
will need to suffice. It is desirable to standardize studies on blood
bacterial signatures in several diseases and pair them with traditional
disease assessment to apply them in the clinical context of prevention in
the future.
Blood bacterial signatures derived therapeutic avenues
Targeting bacterial signatures in disease requires an understanding of how
these signatures come about in the first place, whether they are a mirror of
a common immunological denominator contributing both to metabolic health and
increased susceptibility towards bacterial translocation, or whether
translocated bacteria (alive or dead) are at the base of immunological
cascades contributing to metabolic disease. Moreover, even if bacteria in
the blood do not make up an ecological niche, bacterial functionality
related to host-bacterial co-metabolism such as cholesterol degradation
might still reflect the severity of cardiometabolic diseases. In this sense,
consolidating a new therapeutic strategy might not directly require
targeting bacteria in the blood but other upstream mechanisms associated
with an increased bacterial translocation. Downstream strategies may ensue
over time but are currently inaccessible due to the lack of informative
mechanisms linking circulating microbial signatures to disease.
Examples for upstream strategies could be targeted at correcting altered gut
microbiota via pro-, pre-, syn- or antibiotics, an impaired gut barrier, or
the modulation of the crosstalk between the host and the microbiome by
harnessing bacterial metabolites but also by targeting the bacterial
metabolism of xenobiotics.
Elimination strategies such as via antibiotics
When it comes to antibiotics, studies in humans have been contradictory,
showing that children treated with antibiotics are more likely to
develop obesity and diabetes in a dose-dependent manner [86 ]
[87 ]
[88 ], while in mice at least,
antibiotics treatment-induced improved gut barrier, insulin sensitivity
and weight loss [89 ].
Furthermore, gram-positive bacteria-targeted antibiotic treatment of
mice with systemic lupus erythematosus reduced bacterial growth in
mesenteric lymph nodes and the liver [90 ].
From a clinical perspective, antibiotic treatment not seldom leads to a
selection of pathogens such as C. difficile . Considering this,
the lack of possibility to target only one specific strain/taxon
or bacterial function, the increase of multiresistant pathogens, and
lack of informative mechanistic studies, we believe the use of
antibiotics to treat metabolic disease is currently questionable. Other
strategies such as bacteriophage implementation, growth inhibition via
bacteriostatic non-antibiotics, or even blood UV-Irradiation [91 ] might be alternatives but
necessitate very nuanced testing for safety and efficacy. Moreover,
these rather experimental approaches are hard to justify in metabolic
disease, where pharmacotherapies with clinical evidence for hard
outcomes are available.
Supplementation strategies via replenishing missing bugs
Fecal microbiota transplants (FMTs) FMTs has proven to be a valuable
tool for the discovery of next-generation microbiota targeted
therapeutics and has now been successfully implemented in the
effective treatment of recurrent C. difficile either via
colonic or duodenal infusion or in the form of oral frozen capsules
containing fecal material from healthy donors [92 ]
[93 ]. FMT, on the other
hand, appears to only transiently improve insulin sensitivity and
insulin secretion without ameliorating obesogenic phenotypes due to
the failure of donor’s microbiota to colonize the gut of the
recipient’s long-term [94, 95 ]. Interestingly, the application of FMT in
newly diagnosed type 1 diabetes stabilized residual Beta-cell
function mainly seen after autologous as compared to healthy donor
FMT [96 ]. There are no
studies proving a reduced translocation of bacterial fragments
through the intestinal barrier after FMT. Alas, a case of bacteremia
with a multiresistant bacteria with subsequent death has been
recently described [97 ],
encouraging the search for more targeted and safe strategies than
FMTs in metabolic disease.
Supplementation of single bacteria to alleviate metabolic
disease
Given the medical risk of FMTs and their short-lived success, there
has been increasing focus on the therapeutical use of single taxa or
consortia. Akkermansia muciniphila has received much
attention, proving to alleviate obesity, insulin resistance, and
increased gut permeability in mice [98 ]. Supplementation with
pasteurized rather than live A . muciniphila mildly
improved insulin sensitivity [99 ] (phase 1–2 Study) and improved gut barrier
function via the thermostable outer-membrane protein
Amuc_1100’s interaction with TLR2 in the host [100 ]. Similarly, evidence
of improved insulin sensitivity being associated with Eubacterium
hallii (Anaerobutyricum ) [101 ] after FMT and in mice
[102 ] has led to a
pilot study to investigate the effects of direct supplementation of
E. hallii in humans, which implicated the supplementation
in improving peripheral insulin sensitivity [103 ]. Similarly,
Faecalibacterium prausnitzii is relatively increased in
metabolic health [14 ]
[104 ], which led to further
studies to delineate potential beneficial effects on metabolism.
While a microbial anti-inflammatory molecule derived from F.
prausnitzii reduced inflammation in mice models of colitis
via nuclear factor-kappaB inhibition [105 ], there is, to date, no
direct evidence for an anti-hyperglycemic effect. Several bacteria
could be considered in the quest for precision medicine in metabolic
disease [106 ] but
currently do not embed the notion of bacterial translocation as a
clearly defined outcome.
Bacterial metabolites and postbiotics
Several bacterial metabolites have been shown to contribute to metabolic
regulation in the host. Secondary bile acids, produced after hydrolysis
of primary bile acids by the bacteria in the gut, seem to increase body
expenditure and tissue-specific glucose [107 ]
[108 ]
[109 ] uptake as well as pancreas
secretion of insulin [110 ].
But these effects have not been associated with bacterial translocation
and are hard to track intracellularly considering the target farnesoid X
receptor can be both activated and inhibited in different tissues with
similar effects on the metabolism, suggesting further exploration of
tissue-specific receptor pathways [109 ]
[111 ]. On the
other hand, short-chain-fatty acids are well-known microbial metabolites
with pleiotropic effects on the host produced by bacterial fermentation
of fibers [112]. Importantly, butyrate, observed to be ubiquity reduced
in T2D (along with a decrease in butyrogenic groups [14 ]
[113 ]), can directly reduce
colonic inflammation by regulating interleukin-18 secretion in the
epithelium and immunosuppressive T-cells [114 ] and hence maintain
intestinal homeostasis and integrity via inhibition of histone
deacetylase activity [114 ].
Several other metabolites have been clearly and robustly associated with
metabolic diseases such as imidazole propionate in diabetes [115 ], trimethylamine N-oxide
and tyrosine derived metabolites with cardiometabolic disease [116 ]
[117 ] but also
tryptophan-derived microbial metabolites, which improve inflammation and
the gut barrier function [118 ]
[119 ] as well
as microbially produced polyphenols such as urolithin A, which have been
shown to reduce increased gut permeability [120 ] and improve mitochondrial
health in humans [121 ],
motivating the identification and application of postbiotics
(bacteria-derived non-living components) or other therapeutic strategies
targeting deleterious bacterial metabolites.
In summary, there is a plethora of microbially based next-generation
therapeutics which need to be further tested in humans in the specific
condition of disease, where their actions could indeed be scrutinized.
In which case, it would be helpful to analyze whether these treatments
modify bacterial translocation and bacterial blood signatures while
changing the course or severity of the metabolic disease. This will
further substantiate the contribution of these signatures to disease and
place them in a more strategic context for diagnostic or therapeutic
implementation.
Important considerations: Contamination, biases in assessment and mitigation
strategies
While the technological and analytical developments in the microbiome field can
be seen as overall monumental for the scientific landscape, inspiring curiosity,
leading the way into personalized medicine of both communicable and
non-communicable diseases [121 ]
[122 ], and helping relinquish
established dogmas, low bacterial biomass samples such as the blood require
careful considerations of pitfalls and measures to mitigate for several biases
[121 ]
[122 ]
[123 ]. Specifically, these samples are
highly prone to contamination from background environmental DNA and to technical
biases, not the least of them being over-amplification during PCR [122 ]
[123 ]
[124 ]. The lack of standard operating
procedures, benchmarked techniques, and best practices both on the experimental
and analytical sides of research in working with low biomass samples has led to
sensationalized science and wide controversy [125 ]. Hence, with increasing bacterial
signature mining in low biomass samples, it is vital to unify experimental and
analytical procedures aiming to overcome limiting biases, which should be fully
understood and accounted for during study design, data interpretation but more
importantly, during the development of diagnostic and therapeutic tools based on
these results.
Technical aspects
Routinely, microbiome studies will usually consist of the following
procedures before any analytical pipeline can be implemented: sample
collection, DNA extraction, and sequencing library preparation (possibly
with or without bacterial enrichment techniques such as culture [39 ]
[126 ]
[127 ] or bacterial DNA enrichment),
high-throughput sequencing technologies [128 ] such as amplicon-based sequencing (e. g.,
16 S rRNA gene), shotgun metagenomic sequencing [129 ], as well as RNA
sequencing-based approaches [38 ].
Because procedural controls, which take contamination of samples during
collection and cross-contamination during processing and sequencing into
account, are rarely implemented, studies in low microbial biomass samples
relying on sensitive metagenomic techniques highly confound and inflate the
diagnostic implications for microbial contribution to disease [130 ]. Therefore, it is pivotal to
identify the sources of contamination during sampling and laboratory
procedures. These are 1) personal (doctors, nurses, doctor assistants, and
technicians) and their direct ‘personal cloud’ [131 ], 2) the immediate environment
of sampling, storage, and processing, and 3) laboratory reagents and
equipment.
Accordingly, they can be mitigated: 1) contamination from a
sampling/processing person could be reduced by wearing sterile
masks, gloves, surgical headdress, surgical robes (for single use) and
relying on automation where possible, which can be later better accounted
for by correction for batch effects [132 ]. 2) Environmental contaminants are hard to control and
account for, primarily because the sampling usually occurs in a different
place from where the samples are stored and processed, and the environmental
signatures of these locations can very well depend on the subjects working
there [133 ]. Because these
contaminants are most likely to be found in the air and on surfaces, it
might be worthwhile to sample these environments during subject inclusion
and processing times. 3) low biomass samples are most amenable for
contamination during processing with lab reagents and equipment, which might
still contain bacterial DNA even when they are ‘bacteria
free’ with even the most promising kits for DNA extraction
containing clear DNA signatures [124 ] and can overwhelm genuine bacterial signatures from the
sampled tissue [134 ]. More
interestingly and maybe less intuitively, low input DNA concentration can
highly bias population levels because it results in overamplification of the
template DNA, increases duplication rates, and favors AT-rich sequences
[135 ], which does not, for
example, occur in negative controls, leading hence to the overrepresentation
of a particular species or even a contaminant and confounding
population-level analyses. This requires a multi-pronged approach whereby a)
reagents and equipment are subjected to UV radiation, b) controls
(extraction blanks) from the sampling and laboratory environments, equipment
and reagents are carried through the entire procedures as the actual
samples, samples are randomized, and standard operating procedures are
consistently followed during lab work with the inclusion of no-template
amplification (possibly by the same subjects for a specific project) and c)
bioinformatics approaches are included to assess contamination [136, 137 ], track its source
[138 ]
[139 ], account for it, and remove it
[139 ]. Similarly, reviewers,
editors, and readers should be versed when looking at studies analyzing the
impact of bacterial signatures in disease. For the interested experimenter
and reader, we wholeheartedly recommend reviewing the RIDE-criteria and the
recent publication by Eisenhofer et al. [130 ]
[140 ].
Scientific aspects
While this area of research has immense untapped potential, it is vital to avoid
methodological pitfalls, which would lead to an inevitable waste of time, money and
highly tarnish the credibility of the research. Moreover, it would be important to
address limitations in the design and interpretation of current research. For
example, while comparing studies, it is essential to understand the effect of the
matrix from which the bacterial signature is derived: is it plasma, whole blood, or
peripheral blood mononuclear cells? The distribution of microbial DNA between
compartments of the blood is not clear. This information is important because the
matrix can highly influence technical procedures such as PCR efficiency. This has
overarching implications pertaining to the source and the physiological role of the
bacteria/bacterial sequences. Similarly, studies, which controlled and
corrected for contamination, did so very differently, making the comparison between
them non-conclusive. It will therefore be important to establish an optimal method
to control for contaminants that goes beyond complete subtraction of flagged taxa.
Specifically, a prevalent risk with stringent decontamination is that real signals
reflecting commensal, tissue-specific microbial communities and concomitant
predictive microbial profiles may be discarded. Poore et al. showed, for example,
that a stringent filtering approach might be desirable with the possible downside,
if scalable/universal, to preclude biologically relevant and informative
results [24 ].
It is also safe to say that most works in this area currently fail to address
causality and rely on correlation. While correlational observations are an important
steppingstone and have often been the beginnings of great stories, they remain
exactly that: the beginning. Hence, it is important to address outstanding questions
to inform future research and applications. Some of these questions pertain to a)
whether sequences are indeed cell-free bacterial DNA or come from bacteria that are
either dead, alive, or dormant, b) whether bacteria in the blood occupy a real
ecological niche, c) the source of the bacterial DNA or bacteria for which examining
primary tissue specimens along matched gut epithelia and sampled orifices will be
pivotal, and d) the tissue selectivity of bacteria, which could potentially be
informative.
While microbial DNA has the potential of inherent pathogenicity because it activates
TLR9 in its unmethylated form resulting in inflammatory cascades [141 ]
[142 ], the critical question to answer is whether microbial DNA in itself
is a pathogenic finding, whether it is an epiphenomenon that reflects overall or
specific disease burden or whether both metabolic disease and bacterial signatures
are epiphenomena of lurking immunological conditions or active infections ([Fig. 3 ]), which we yet have to be
identified because our tests lack sensitivity.
Fig. 3 Bacterial translocation can be derived from several
sources such as the skin, the oral cavity, the gut, clinical
infection sources, medical procedures including dwelling catheters, and
contamination/infection from a hospital stay. The translocation can
include endotoxins, whole bacteria, or bacterial sequences leading to
systemic and local tissue inflammation and culminating in local tissue
dysfunction such as insulin resistance in adipose tissue, liver dysfunction,
or cardiovascular disease. Similarly, signaling metabolites from the gut can
contribute to cardiometabolic disease and specific tissue dysfunctions,
which in turn can aggravate bacterial translocation by increasing
susceptibility to colonization and infection and increased
‘leakiness of the gut.’
In moving towards mechanisms and granular interrogation of causality, it will also
be
crucial to understand whether the entire bacteria, bacterial cell components, or
metabolites are most important for the postulated effects of the identified
microorganisms. A question that lends itself to further exploration concomitantly
is
whether the observed taxa are alive or not. A drawback of culture-independent
sequencing methods is their inability to differentiate living from dead bacteria.
This is particularly important because DNA, including that from contaminants, can
persist in the environment. More importantly, while DNA sequencing and functional
annotation can predict bacterial metabolism, only viable bacteria can undertake
active metabolic processes. Therefore, methods that do not relay viability
information will always overestimate the importance of bacterial metabolism in the
host metabolic response, as long as they are not coupled with information on
transcription of metabolic modules or direct measurement of metabolites.
Hence, the application of methods that interrogate active metabolism and delineate
viability is warranted. Techniques exploring viability can characterize one or
several viability aspects, such as the presence of an intact membrane, the
replication of genetic material (via RNA detection), and the detection of metabolism
or energy. The main challenge being, whether these methods are compatible with
next-generation sequencing and whether they are applicable to low-biomass
samples.
Microbial culture caters to both these needs and is the gold standard to prove
viability because it displays the ability of bacterial cells to divide and
metabolize nutrients. On the other hand, several bacterial strains are painstakingly
hard to culture or require immense optimization to determine optimal culturing
conditions. Others are viable but non-culturable because they are either damaged or
maintaining the structural and metabolic properties observed or postulated in their
environment of origin is not possible [143 ]. The application of propidium iodine, propidium monoazide (PMA), and
ethidium monoazide – all dyes that can bind to DNA in membrane compromised
cells and extracellular DNA- enable the labeling of dead bacteria and can be coupled
with several other techniques, including epifluorescence microscopy, flow cytometry,
but also qPCR, digital PCR and, metagenomics among several other downstream
techniques [144 ]. PMA seems to be more
selective and less cytotoxic than other methods, and several of its selectivity
drawbacks are known (stains viable cells for some species while failing to stain
dead cells from other species). More importantly, it can be used for low-biomass
samples, assists in the removal of contaminant DNA (environmental or introduced by
the reagents including PCR reagents) [124 ], has the potential to enrich for rare microbial community members [124 ]
[145 ], and allows for qPCR quantification to distinguish the viable from
the total fraction of microbial cells in several settings [146 ] (reviewed in [144 ]). Unfortunately, the samples are
required to be in an aqueous solution, and ethylenediaminetetraacetic acid (EDTA,
found in specific blood collection tubes) can affect dye permeability through
membranes [147 ], limiting the application
of this method to other human low biomass samples such as the blood. Other methods
such as RNA analyses (e. g., metatranscriptomics) enable quantification and
identification of potentially ‘active bacterial community members’
and can delineate taxa as well as acute responses to stimuli relatively early on
because RNA has a very short half-life [148 ]. Unfortunately, RNA yield is highly dependent on the matrix it
derives from and the preservation method employed until RNA extraction is undertaken
[148 ]
[149 ]. Moreover, significant RNA losses have
been described during sample preparation, which is particularly problematic in
low-biomass samples, where yield is low to begin with [150 ]
[151 ]. Similarly, metaproteomics can also be applied for the detection of
active bacteria. Coupled with metagenomics, it can enable the identification of
active metabolic pathways of bacterial communities and connect them to substrate
availability and environmental conditions. A drawback is that only proteins with
exact matches in the search databases can be identified. Another drawback is the
potential persistence of proteins in specific environments.
Various techniques are available and allow their application in conjunction with
several other methods to detect viability, phylogeny, and metabolic activity,
lending themselves well to studies assessing the contribution of bacterial cells to
overall host health. Several bottlenecks remain, most prominently the necessity for
extensive optimization in a setting where human tissue is a matrix, the low
throughput of these methods paralleled with the high-cost and the need for careful
result interpretation, making this a challenging task but an excellent opportunity
for collaboration between groups across the scientific and medical spectrum.