Keywords
antiphospholipid syndrome - autoimmune disease - intestinal microbiome - gut microbiome
Introduction
Antiphospholipid syndrome (APS) is a prothrombotic autoimmune disorder that is characterized
by obstetric or thrombotic morbidity in the persistent presence of antiphospholipid
antibodies (APLAs).[1] The main autoantigen these antibodies are directed against is circulating plasma
protein β-2-glycoprotein 1 (β2GP-1). The etiology of APLAs remains elusive. Consequently,
no curative treatment is available and patients are treated with anticoagulants, a
treatment with both suboptimal safety in terms of bleeding and suboptimal efficacy,
with recurrent thrombosis and obstetric complications occurring under treatment.
The APS-defining antibodies are characterized by persistence over time, but transient
APLAs are relatively common in individuals without APS. These transient antibodies
are often triggered by infections.[2] The intestinal microbiome, the community of microorganisms that colonizes the intestinal
tract, poses a chronic exposure to a broad variety of microbial antigens and is thus
hypothesized to drive the formation of persistent APLAs.[3] The gut microbiota indeed can trigger autoimmunity through antigen cross-reactivity.
For several autoimmune disorders, including APS, gut bacteria that contain peptide
sequence homologues to the epitopes of autoantigens have been identified, resulting
in antigen cross-reactivity.[4]
[5]
[6] For instance, the gut commensal Roseburia intestinalis contains peptide sequences homologous to the predominant B cell and T cell epitopes
of β2GP-1.[5] Both APS patient plasma and a patient-derived monoclonal APLA indeed show cross-reactivity
with this B cell epitope-mimicking bacterial protein.[5] Similarly, APS patient-derived T cells cross-react with the R. intestinalis T cell epitope homologue. Further experiments with (NZW x BXSB)F1 mice showed that
immunization with and (immunization with and oral administration of) oral administration
of the bacterium led to β2GP-1 cross-reactive antibodies and thrombosis, supporting
a causal contribution of gut microbiota to murine APS.[5]
The emerging evidence on the etiological role of the microbiome in autoimmunity provides
new possible therapeutic targets. Treatment with oral vancomycin of APS (NZW x BXSB)F1
mice reduces anti-β2GP1 immunoglobulin G (IgG) titers, decreased cerebral, cardiac
and pulmonary thrombi, and improves survival.[7] In systemic lupus erythematosus, a disease that strongly overlaps with APS, transfer
of healthy microbiome to patients with therapy-unresponsive disease flares appears
to diminish disease activity.[8]
Both findings underline the therapeutic potential of microbiome-targeted therapies
in autoimmune disease.
To this end, the current study aims to translate the above observations to human APS
by determining whether the intestinal microbiota affect the biochemical disease activity
in APS patients.
Methods
Study Population
We recruited patients with APS, whose diagnosis met the APS Sydney research criteria,
from our tertiary Amsterdam University Medical Centers Vascular Medicine outpatient
clinic and through advertisements with the Dutch APS patients association.[1] All subjects were included based on their historical APLA values and were not re-screened
before start of the study. Positive APLAs were defined as positive lupus anticoagulant
or anti-β2GP-1 IgG or IgM or anti-cardiolipin IgG or IgM titer above 40 GPL or MPL
or above 99th percentile, on at least two occasions 12 weeks apart. Exclusion criteria
were: age below 18 years, current use of antibiotics, current use of a vitamin K antagonist,
history of gastroenteritis in the past month, history of inflammatory bowel disease,
current pregnancy or pregnancy in the past 6 weeks, arterial or venous thrombosis
in the past month, allergy to vancomycin or planned change in platelet aggregation
inhibitors, anticoagulants or hormonal therapy during the study period. Subjects using
vitamin K antagonist were excluded because of a potential effect of the intervention
on the international normalized ratio.
The Medical Research Ethics Committee of the Amsterdam University Medical Centers
approved the study. All participants provided written informed consent. The study
was conducted in accordance with the Declaration of Helsinki. The study was registered
in the Dutch trial register prior to initiation (NTR 7662).
Study Design
All study participants received oral vancomycin 500 mg four times per day for 7 days.
Vancomycin is a broad-spectrum antibiotic mainly effective against gram-positive bacteria
that is poorly absorbed from the gut and has previously been shown to elicit a distinct
shift in gut microbiota.[9] Blood, fecal and urine samples were taken at 7 days before start of treatment (day
−7), at baseline (day 0), at end of treatment (day 7), and 6 weeks after start of
treatment (day 42).
Venous blood was drawn with an 18G needle; all blood samples were processed within
3 hours after collection and stored at −80°C. Participants collected fecal samples
at home within 24 hours preceding the study visit. Samples were stored at 4°C until
study visit and then stored at −80°C until processing.
Outcome
Given the incidence rates, clinical endpoints (i.e., thrombotic or obstetric events)
were not a suitable outcome in this study design. Antibody profiles correlate with
clinical outcomes but no single antibody test correlates strongly enough to function
as a surrogate endpoint. Moreover, since the mechanism through which the intestinal
microbiota influence APS remains elusive, it is unclear which are the optimal biomarkers
to study such an effect. Therefore, we composed a comprehensive biomarker panel of
parameters identified from the literature to be associated with the clinical APS phenotype
([Table 1]).[1]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18] The predefined primary outcome of the study was the first principal component of
the biomarker panel data. This value collectively captures the maximized variance,
while it is adjusted for inter-individual variance, in the combined clinical APS phenotype-related
biomarkers and is thus considered to reflect the changes in disease activity.
Table 1
Biomarker panel
|
Biomarker
|
|
Antiphospholipid antibodies
|
Lupus anticoagulant
|
|
Anti-β2GP1 IgG, IgM, IgA
|
|
Anti-cardiolipin IgG, IgM, IgA
|
|
Anti-phosphatidylserine/prothrombin IgG, IgM
|
|
Primary hemostasis
|
Platelet count
|
|
Light transmission aggregometry
|
|
Platelet flow cytometry
|
|
Von Willebrand factor
|
|
Secondary hemostasis
|
Activated partial thromboplastin time, prothrombin time
|
|
Calibrated automated thrombogram
|
|
Activated protein C resistance
|
|
NETosis
|
Citrullinated histone H3
|
|
Fibrinolysis
|
Clot lysis
|
|
D-dimer
|
|
Complement
|
C3a
|
|
C5a
|
|
Inflammatory markers
|
C-reactive protein
|
|
Tumor necrosis factor α
|
|
Interferon γ
|
|
Interleukin-6
|
Abbreviation: IgA, immunoglobulin A; anti-?2GP1: Anti-β-2-glycoprotein-I.
Biomarker Assays
In addition to APS criteria APLAs, we evaluated anti-cardiolipin IgA and anti-β2GPI
IgA antibodies, because of their role in intestinal immunity, and we measured anti-phosphatidylserine/prothrombin
antibodies. The APLA and lupus anticoagulant assays are described in detail in [Supplementary Materials and Methods]. Quantitative antibody titers and the numerical value of the normalized lupus anticoagulant
ratio for both Russell's viper venom time and silica clotting time were included in
the analysis.
Light transmission aggregometry was performed by inducing aggregation with three concentrations
of adenosine diphosphate (ADP): 2, 5, and 10 μM. The following parameters of the aggregation
curve were included in the analysis: primary slope, primary aggregation, maximal aggregation,
final aggregation.
For platelet flow cytometry, platelets were either unstimulated or stimulated with
a receptor-specific agonist either ADP, CRP-xl, PAR-1 AP, PAR-4 AP, or U46619. Platelets
identified based on forward and side scatter were selected on GPIbα positivity or
integrin αIIbβ3 positivity. Of the unstimulated integrin αIIbβ3-positive population,
the percentages positive for glycoprotein VI and integrin α2β1 were included in the
analysis. For the unstimulated and agonist-stimulated platelets, the percentages positive
for anti-P-selectin and anti-fibrinogen of the GPIbα-positive population were included
in the analysis.
For the calibrated automated thrombogram, thrombin generation curves were generated
with 1 pM and respectively 5 pM of tissue factor. The following parameters were included
in the analysis for both tissue factor concentrations: lag time, peak height, time
to peak, velocity, and endogenous thrombin potential (ETP). Activated protein C (APC)
resistance was measured by evaluating ETP in the presence of APC and 5 pM tissue factor.
APC sensitivity ratios (ETP values measured in the presence of APC/ETP values in absence
of APC) normalized to pool plasma were included in the analysis.
Clot lysis test was performed with and without carboxypeptidase inhibitor (CPI) to
evaluate contribution of activation of thrombin activatable fibrinolysis inhibitor.
Clotting time, clot lysis time without CPI, and clot lysis time with CPI were included
in the analysis.
Secondary Outcomes
We evaluated gut microbiota composition by 16S sequencing (Illumina), assessed intestinal
permeability with lactulose mannitol test and plasma marker lipocalin-2 and measured
fecal calprotectin and fecal short-chain fatty acids (SCFAs).[19] These assays are described in the [Supplementary Materials and Methods] section.
Statistical Analysis
All analyses were performed in R statistics version 4.0.3. We performed multilevel
principal component analysis using the mixOmics package.[20]
[21] Principal component analysis, a widely used form of machine learning, is a dimension
reduction technique, used to enhance analysis and visualization of high dimensional
data, i.e., high variable number datasets. It reduces the variables by capturing the
maximum amount of variability in the data on new composite virtual variables, called
principal components. Multilevel principal component analysis is a variant that is
specifically applicable to repeated measurement data. This multilevel analysis deals
with the high inter-individual variation relative to the intra-individual variation,
which is the relevant parameter in this pre–post intervention design. For the principal
component analysis, the biomarker panel data were centered and scaled. Nonnormally
distributed data were transformed using log transformation. We replaced assay results
below the lower limit of quantification with the lower limit divided by two. Variables
with over 20% missing values were excluded from the analysis. The remaining missing
data were imputed using the Non-linear Iterative Partial Least Squares (NIPALS) algorithm
within the mixOmics package. Differences in the principal components, which followed
a normal distribution, were tested using t-test comparing the averaged baseline values (time point day −7 and day 0 combined)
to day 7 and to day 42.
Secondary outcomes (APLA titers, lipocalin-2, fecal SCFAs, and fecal calprotectin)
were evaluated with paired t-test. Considering APLA profiles differ amongst APS patients, APLA titers were assessed
as a composite value. For each subject we selected the APLAs that subject was positive
for at baseline. These APLA values were normalized to the baseline value of that APLA
in the individual subject. These normalized values were then averaged to create a
composite APLA.
Microbiota data were analyzed using phyloseq, vegan, and mixOmics packages. Microbiota
abundance was visualized using multi-level principal component analysis of the centered
log-ratio transformed amplicon sequence variant counts.
Results
A total of 16 APS patients were included in this study. One subject was excluded before
start of treatment because of pregnancy. In total, 15 subjects completed the study,
of which 14 were women. Baseline characteristics and APLA profiles of the participants
who underwent treatment are depicted in [Table 2]. Notably, 4 out of 15 study subjects tested negative for all APLA at baseline. Patients
were included based on previous diagnostic APLA values and not rescreened before inclusion.
Table 2
Baseline characteristics
|
Characteristic
|
N = 15
|
|
Age, mean ± SD
|
39.3 ± 8.6
|
|
Female, n (%)
|
14 (93)
|
|
BMI, mean ± SD
|
26.6 ± 5.6
|
|
Ethnicity, n (%)
|
|
Caucasian
|
13 (87)
|
|
Mixed
|
2 (13)
|
|
History of pregnancy morbidity, n (%)
|
|
Fetal loss ≥ 10 weeks of gestation
|
8/12 (67)
|
|
> 2 early pregnancy losses
|
3/12 (25)
|
|
Preterm birth
|
2/12 (17)
|
|
History of thrombosis, n (%)
|
|
Venous thrombosis
|
3 (20)
|
|
Arterial thrombosis
|
3 (20)
|
|
Concurrent autoimmunity, n (%)
|
|
SLE
|
0 (0)
|
|
Other autoimmune disorder
|
2 (13)
|
|
Use of DOAC, n (%)
|
4 (27)
|
|
APLA positivity at diagnosis, n (%)
|
|
Lupus anticoagulant
|
10 (67)
|
|
Anti-β-2-glycoprotein-I IgG
|
8 (53)
|
|
Anti-β-2-glycoprotein-I IgM
|
2 (13)
|
|
Anti-cardiolipin IgG
|
10 (67)
|
|
Anti-cardiolipin IgM
|
4 (27)
|
|
APLA positivity at baseline, n (%)
|
|
Lupus anticoagulant
|
6 (40)
|
|
Anti-β-2-glycoprotein-I IgG
|
6 (40)
|
|
Anti-β-2-glycoprotein-I IgM
|
4 (27)
|
|
Anti-β-2-glycoprotein-I IgA
|
4 (27)
|
|
Anti-cardiolipin IgG
|
6 (40)
|
|
Anti-cardiolipin IgM
|
1 (6.7)
|
|
Anti-cardiolipin IgA
|
5 (33)
|
|
Anti-phosphatidylserine prothrombin IgG
|
6 (40)
|
|
Anti-phosphatidylserine prothrombin IgM
|
5 (33)
|
|
Triple positive (LAC, any aβ2GP1, any aCL)
|
3 (20)
|
|
No APLA
|
4 (27)
|
Abbreviations: APLA, antiphospholipid antibody; BMI, body mass index; DOAC, direct
oral anticoagulant; IgA, immunoglobulin A; LAC, lupus anticoagulant; SD, standard
deviation; SLE, systemic lupus erythematosus; aCL, anti-cardiolipin; a?2GP1, anti-β-2-glycoprotein-I.
Gut Microbiota Composition
Treatment with vancomycin induced a distinct shift in gut microbiota composition directly
after end of treatment which faded at 6 weeks after treatment ([Fig. 1]). [Fig. 1] shows the difference in the first two principal components (PC1 and PC2) of the
abundance of the sequenced microbes between day 0 (blue), day 7 (purple), and day
42 (yellow).
Fig. 1 Multi-level PCA of the centered log-ratio transformed amplicon sequence variant counts.
This provides a visual representation of the fecal microbiome abundance data at the
three time points. Genera with the strongest ordination associations are indicated
with arrows. PC, principal component; PCA, principal component analysis.
Primary Outcome
The primary outcome, the first principal component (PC1) of the biomarker panel, is
depicted in [Fig. 2A] at the four time points. C-reactive protein, tumor necrosis factor-α, and interferon-γ
were excluded from the analysis because the majority of measurements was below the
limit of detection. PC1 differed between baseline and day 7 (absolute difference:
1.65, p = 0.03). There was no difference between baseline and day 42 (absolute difference:
1.89, p = 0.10).
Fig. 2
Principal component analysis of the biomarker panel. (A) Primary outcome principal component 1 differs between baseline and day 7 (absolute
difference: 1.65, p = 0.03). (B) Multilevel PCA plot for all subjects at four time points. (C) Post-hoc analysis: principal component 1 for APLA-positive subjects (n = 11) significantly differs between baseline and day 7 and between baseline and day
42 (absolute difference: −2.12, p = 0.02, and absolute difference: −3.00, p = 0.03, respectively). (D) Multilevel PCA plot for APLA-positive subjects at four time points. APLA, antiphospholipid
antibody; PCA, principal component analysis; PC, principal component.
A post-hoc analysis of the subgroup of patients with positive APLAs at baseline showed
a significant difference in PC1 at both day 7 and day 42 compared with baseline ([Fig. 2C]) (absolute difference: −2.12, p = 0.02 and absolute difference: −3.00, p = 0.03, respectively). Of note, the reverse direction in absolute difference in the
post-hoc analysis derives from the difference in principal coordinates in the subgroup
data and does not signal an opposite biological effect.
The finding of 4 out of 15 patients testing negative for all APLA was unanticipated.
This did however create the possibility of investigating whether the effect on the
main outcome (PC1) was APS-specific. We performed a post-hoc comparison between the
subjects that had APLAs at baseline and those that tested negative for all antibodies.
Delta values for PC1 were calculated by subtracting the PC1 value at day 7 and 42,
respectively, from the combined preintervention values. The groups were compared on
these delta-values using unpaired t-test. In the APLA positive group, the delta in PC1 was −3.44 at day 7, whereas in
patients with negative APLAs, the PC1 delta was −0.03 (p = 0.03) ([Fig. 3A]). A similar difference between these groups was found at day 42 with a PC1 delta
of −4.46 in the patients with positive APLAs and 1.89 in subjects with absent APLAs
at baseline (p = 0.03) ([Fig. 3B]).
Fig. 3
Differences in the first principal component at two time points in patients with (
n
= 11) and without (
n
= 4) positive APLAs at baseline. (A) Delta in PC1 at day 7 for patients with and without APLAs significantly differed
(p = 0.03). (B) Delta in PC1 at day 42 for patients with and without APLAs, significant difference
(p = 0.03). APLA, antiphospholipid antibody.
Secondary Outcomes
Antiphospholipid Antibody Titers
Because APS patients have distinct APLA profiles, APLA levels were analyzed as a composite
value reflecting the change in an individual's profile. This composite value did not
differ between baseline and both day 7 and day 42 ([Fig. 4A]) (paired t-test, p = 0.25 and p = 0.56, respectively).
Fig. 4
Secondary outcomes at four time points. (A) The normalized APLA composite for subjects with positive APLA at baseline (n = 11). (B) Plasma lipocalin-2 levels for all subjects (n = 15). (C) Fecal calprotectin levels for all subjects (n = 15). (D) Fecal acetate levels for all subjects (n = 15). APLA, antiphospholipid antibody.
Intestinal Permeability
Neither plasma lipocalin-2 levels nor fecal calprotectin ([Fig. 4B, C]) levels showed a difference postintervention compared with baseline indicating no
effect of vancomycin in gut permeability. All but one lactulose measurements were
below the limit of detection, prohibiting comparison between the time points.
Fecal Short-Chain Fatty Acids
Of the fecal SCFAs, only acetate was different between day 0 and day 7 ([Fig. 4D], p = 0.04), but this was not statistically significant after Bonferroni correction for
multiple testing.
Discussion
The current study evaluated whether experimentally changing the intestinal microbiome
affects the disease activity in human APS. The intervention of oral vancomycin, which
is not absorbed from the gut, was effective in eliciting a substantial shift in the
gut microbiota composition. This indeed resulted in a change in biochemical phenotype
at the end of treatment, as collectively reflected by the panel of clinical APS phenotype-related
biomarkers. The effect diminished after 6 weeks, but was prolonged in the subgroup
with APS-defining antibodies present at baseline. The unanticipated subgroup without
APLA at baseline showed no response to treatment in a post-hoc analysis. This suggests
that the effect of this intervention is specific to APS patients. Overall, the data
support a pathophysiological role of the microbiota in human APS.
The findings in APS patients complement in vitro data, mouse studies, and human observational
studies on the role of the gut microbiota in APS.[3]
[5]
[7] Measuring disease activity poses a challenge in smaller human APS studies, given
the infrequency of clinical events and the range of cell types and extracellular proteases
involved in the pathogenesis. This novel approach represents an integral and unbiased
method to assess the APS biochemical phenotype. The wide selection of biomarkers,
which have been associated with the APS phenotype previously, ensures that the broad
spectrum of APS pathophysiology is captured. The application of principal component
analysis, with multilevel variant to account for the repeated measurement data structure,
enables us to combine these data into a single representative outcome measure in an
unbiased way.
The choice of experimental study design merits consideration, and was determined by
the fact that APS is a rare disease. The prevalence of APS makes randomized studies,
requiring a larger sample size, unfeasible for proof-of-concept studies such as performed
here. The pre–post design remains valid for the primary objective of investigating
any effect of the gut microbiome on disease biomarkers in APS patients. A pre–post
design is sensitive to bias when a time-outcome effect is present. However, our inclusion
criteria ensured patients in a stable disease state. For this chronic condition, variation
within the short time span of the 7-week study duration is not to be expected. Moreover,
as this was not a therapeutic study, randomization was not necessary to ensure comparability
between exposure states (treated vs. untreated). Instead, patients functioned as their
own controls maximizing statistical power to investigate our hypothesis in this relatively
rare disease.
This study has some limitations. The use of the composite outcome hampers the interpretation
of the effect size. The value is dimensionless and cannot be used to inform us on
the clinical impact. A sample size calculation was not possible because the effect
size could not be estimated with a dimensionless primary outcome. This methodological
tradeoff was deemed necessary and functional, given that the aim was an assessment
of whether, rather than how much, the microbiota affect APS in humans.
Another limitation is that four subjects tested negative for all APLAs. All subjects
had a historic diagnosis of APS according to the Sydney criteria[1] but were not re-screened for APLAs directly before start of the study. Possible
explanations for the discrepancy include seroconversion and interlaboratory variation,
both of which are known to occur with APLAs.[22]
We investigated several potentially involved mechanisms. The shift in disease activity
is likely not to be mediated by a direct effect on APLA production or clearance as
the intervention did not affect antibody titers. Previous work has shown an altered
intestinal permeability in autoimmune disease in general and APS in particular.[5]
[23] However, we in fact found no evidence of increased gut permeability in our APS cohort.
A modest effect too small to detect with the markers we used cannot be excluded. The
lactulose mannitol test however, a gold standard for functional measurement of gut
permeability, resulted in concentrations below the detection limit of our sensitive
mass spectrometry assay for virtually all subjects.[19]
[24] However, since there is extensive interplay between the immune system and intestinal
microbiota with an intact barrier function, increased intestinal permeability is not
a prerequisite for an intestinal microbiota-mediated effect on disease activity.
A direct systemic effect of vancomycin on the biomarker panel is unlikely because
of its poor intestinal absorption. Moreover, the effect appeared to be APS-specific,
based on the post-hoc analysis in which we regarded the APLA-negative subgroup as
non-APS controls. Lastly, SCFAs are linked to autoimmunity, for instance through effects
on T helper cells and regulatory T cells.[25] Of the four measured SCFAs, we observed a small effect on fecal acetate levels only
(p = 0.04), but this was not statistically significant after correcting for multiple
testing.
In conclusion, this study shows that the intestinal microbiota affect the biochemical
disease activity in human APS through a yet unknown mechanism. Future studies should
focus on unraveling this relation in pursuance of possible gut microbiome-directed
treatment strategies.
