Introduction:
Neonates are protected from colonizing bacteria by antibodies secreted into maternal
milk. Necrotizing Enterocolitis (NEC) is a disease of neonatal preterm infants with
high morbidity and mortality that is associated with intestinal inflammation driven
by the microbiota
1-3
. The incidence of NEC is significantly lower in infants fed with maternal milk, though
the mechanisms underlying this benefit are not clear
4-6
. Here, we show that maternal Immunoglobulin A (IgA) is an important factor in protection
against NEC. Analysis of IgA-binding of fecal bacteria from preterm infants indicated
that maternal milk was the predominant source of IgA in the first month of life and
that a relative decrease in IgA-bound bacteria is associated with the development
of NEC. Sequencing of IgA-bound and unbound bacteria revealed that prior to disease
onset, NEC was associated with increasing domination of the IgA-unbound microbiota
by Enterobacteriaceae. Further, we confirmed that IgA is critical in preventing NEC
in a murine model, where pups reared by IgA deficient mothers are susceptible to disease
despite exposure to maternal milk. Our findings show that maternal IgA shapes the
host-microbiota relationship of preterm neonates and that IgA is a critical and necessary
factor in maternal milk for the prevention of NEC.
Main
NEC is associated with an intestinal microbiota of decreased diversity and increased
Enterobacteriaceae, but this association is not sufficient for disease
7,8
. Bioactive components of maternal milk, including IgA antibodies, shape the neonatal
microbiota
9-12
. It is not known how the anti-bacterial IgA repertoire of maternal milk varies between
women, but mammary gland IgA-producing B cells traffic from the intestine and thus
may differ between mothers as a result of individualized microbiomes and infectious
histories
13-15
. We hypothesized that differential binding of the preterm microbiota by maternal
IgA is a central feature of NEC pathogenesis.
To analyze immunoglobulin (Ig) binding of gut bacteria in preterm infants we stained
fecal samples (Table 1a) with anti-human IgA, IgM and IgG antibodies and measured
the Ig-bound populations with flow cytometry
16,17
. This initial sample set contained 30 samples collected at the time of NEC diagnosis
and 39 samples from age-matched controls. Surveyed across all samples, the percentage
of IgA-bound bacteria was far greater than the percentages of IgM- and IgG-bound bacteria
and samples from maternal milk-fed infants contained a far greater abundance of IgA
positive bacteria compared to formula-fed infants (Figure 1a,b and Extended Data 1a,b).
Although a majority (11/19) of formula-fed infants had <1% of their intestinal bacteria
bound by IgA,some samples from formula-fed infants contained high amounts of IgA positive
bacteria (Figure 1b). Because B cells generally do not populate the intestine until
about 4 weeks of age
18
, we hypothesized that fecal samples from formula-fed infants collected before this
time point would not contain IgA-bound bacteria. Indeed, we found a significant temporal
relationship between age and IgA binding in formula-fed infants that was not observed
among maternal milk-fed infants (Figure 1c). A dedicated analysis of samples from
a single formula-fed preterm infant revealed no IgA positive bacteria in the first
4 weeks of life, strongly implicating maternal milk as the primary source of perinatal
IgA (Extended Data 1c). Limiting our analysis of this data set to 4 weeks post-delivery,
we found that samples from infants with NEC contained less IgA-bound bacteria than
samples from age-matched controls (Figure 1d). However, NEC infants in this cohort
were more likely to be formula fed; additionally, their fecal samples were collected
after NEC was diagnosed and treatment had been initiated with antibiotics and cessation
of feeding. To eliminate the impact of these confounding variables, we selected and
analyzed a prospectively collected longitudinal series of samples from 23 milk-fed
preterm infants, of which 43.4% subsequently developed NEC (Table 1b). Critically,
we found that the fraction of IgA positive bacteria decreased with time among infants
that developed NEC, whereas IgA binding of fecal bacteria showed no relationship in
controls. (Figure 1e and Extended Data 2a,b). Thus, it appears that in infants that
will develop NEC, a change occurs in either the intestinal microbiota or the maternal
IgA repertoire that leads to the ‘escape’ of intestinal bacteria from binding.
We next sought to identify shifts in the microbiota that might explain this drop in
IgA positive bacteria. Analyzing all samples together, we observed a modest enrichment
for Enterobacteriaceae and reduction in Gram positive anaerobes such as Lachnospiraceae
in infants who will go on to develop NEC (Figure 2a and Extended Data 3). This result
aligns well with published reports describing dysbiosis prior to onset of NEC
7
. We therefore hypothesized that loss of IgA binding of Enterobacteriaceae may be
important to the decrease in IgA-bound bacteria in NEC. In support of this hypothesis,
we found that the abundance of IgA-bound bacteria varied inversely with the abundance
of Enterobacteriaceae among infants that would later develop NEC, but not controls
(Figure 2b).
To directly identify which bacterial taxa might be more or less bound by maternal
antibodies in our longitudinal prospective cohort (Table 1b), we physically separated
the two fractions and measured microbial diversity by sequencing the V4 region of
bacterial 16S rRNA genes (IgSeq)
19
. In accord with published reports
16,20
, we did not achieve purity (>99%) in our IgA separations, complicating our ability
to discriminate IgA positive and negative bacteria (Extended Data 4a). Post-sort flow
cytometric analysis of each sample allowed us to deconvolve the contamination, significantly
increasing the correlation between the percent IgA positive bacteria in the unsorted
sample and the ratio of IgA positive and negative bacteria reads from IgSeq (Extended
Data 4b-d). Deconvolution is particularly important in samples with low levels of
IgA positive bacteria, which both validates that our technique is correcting for contamination
and improves analysis of the samples most critical to explaining our observation of
reduced IgA positive bacteria in NEC (Extended Data 4c and d).
Longitudinal IgSeq analysis of deconvolved samples revealed that control infants showed
a significant increase in bacterial diversity amongst IgA negative bacteria over time,
while no significant changes were detected in the IgA positive sample (Figure 2c).
Conversely, among infants that progressed to NEC, the diversity of both IgA positive
and negative fecal bacteria significantly decreased over time (Figure 2c)
21
. In accord with our analysis of the total microbiome (Figure 2b), IgSeq also revealed
that over time the IgA negative intestinal microbiota of NEC infants became dominated
by Enterobacteriaceae, while anaerobes (notably Clostridiales and Bifidobacteriales)
were virtually undetectable (Figure 2d,e and Extended Data 5). Conversely, amongst
controls, the relative abundance of IgA negative anaerobes increased and the relative
abundance of IgA negative Enterobacteriaceae decreased, confirming that Enterobacteriaceae
are bound at relatively higher frequencies over time in infants that do not develop
NEC (Figure 2b,d,e and Extended Data 5). While the IgA positive and negative fractions
from controls differed in both diversity and the relative abundance of Enterobacteriaceae,
in NEC patients these fractions were not discernibly different (Figure 2c and d).
The lack of differences between IgA positive and negative fractions of NEC infants
may be explained by very low diversity and domination by Enterobacteriaceae, so essentially
there are few taxa available for IgA to bind (Extended Data 5).
An advantage of IgSeq deconvolution is that it allows for a meaningful comparison
of taxon abundance between IgA positive and IgA negative samples, as their ratio corresponds
to the abundance of IgA-bound bacteria in the unsorted sample (Extended Data 4). When
we calculated the ratio of the paired IgA negative and positive reads, we observe
a unique increase over time in IgA negative total and Enterobacteriaceae reads in
infants who will go on to develop NEC, that correlates well with IgA binding data
from flow cytometry (Figure 1e and 2f). In contrast, control infants show no significant
shifts in the relative abundances of IgA positive and negative total bacteria, Enterobacteriaceae
or anaerobes and only show significant shifts in less abundant OTUs (Figure 2f and
Extended Data 6). As Enterobacteriaceae is the most abundant OTU in preterm infants
and the only taxa uniquely increasing in the IgA negative fraction of the microbiota
prior to disease, the increase in IgA negative Enterobactericeae is the most likely
driver of the reduced IgA bound bacteria seen preceding NEC (Figure 1e, 2a, 2f and
Extended Data 6). Thus, we have identified that NEC infants uniquely fail to diversify
their intestinal microbiota with anaerobic bacteria and instead remain dominated by
IgA-unbound Enterobacteriaceae.
To test the possibility that the increase in Enterobacteriaceae was driven by a rapid
bloom we measured the number of each bacterial taxa per unit mass in fecal samples
22,23
. Although we saw increased Enterobacteriaceae amongst some infants developing NEC,
inter-individual variation was high and there were no statistically significant differences
between cases and controls (Extended Data 7). However, our analysis may not adequately
represent bacteria in the small intestine so we cannot rule out the possibility of
focal expansions of Enterobacteriaceae associated with NEC
24
. Nonetheless, we favor a model where loss of IgA binding of the microbiota is induced
either by mutation or by transcriptional modifications that allow sub-populations
of Enterobacteriaceae to escape maternal IgA but do not lead to increases in their
total number.
To further define the contribution of maternal IgA to disease pathogenesis, we turned
to an experimental murine model of NEC
25,26
. We bred mice so that heterozygote wild-type pups were fed by mothers that either
can (C57BL/6) or cannot produce IgA (Rag1−/− or Igha−/−), and compared them to formula-fed
positive controls (Figure 3a). We confirmed that mice, like humans, produce little
IgA during their first two weeks of life and that dams are the primary source of neonatal
IgA
18,27,28
(Figure 3b). We also determined that Enterobacter spp. gavaged into pups (C57BL/6
dams) to induce NEC was enriched in the IgA positive fraction, indicating that murine
dams may produce protective IgA without being vaccinated (Extended Data 8). Strikingly,
pups undergoing the NEC protocol that were breast-fed by mothers lacking IgA (Rag1−/−
or Igha−/−) showed a phenotype that was indistinguishable from formula-fed controls.
Specifically, they exhibited increased mortality, and severe intestinal damage characterized
by shortened necrotic villi and mucosal sloughing (Figure 3c-e). Furthermore, pups
fed by Igha−/− mothers exhibited a significant reduction in weight gain compared to
pups fed by wild-type mothers (Figure 3f). We have thus shown, using an experimental
model of NEC, that maternal milk only protects against NEC when it contains IgA.
Discussion
Previous studies have shown associations between the abundance of Enterobacteriaceae
and NEC
7
. We now show that IgA-unbound Enterobacteriaceae is more closely linked to NEC development
than total Enterobacteriaceae abundance. Animal studies have indicated that both host
and maternal IgA is important in controlling Enterobacteriaceae and establishing a
mature microbiota characterized by fastidious anaerobic bacteria
29,30
. Our results indicate that binding of bacteria by maternally-derived IgA may promote
diversity in the microbiome and the acquisition of anaerobic bacteria during the critical
window when infants make little or no IgA of their own, perhaps by limiting inflammation
driven by Enterobacteriaceae
31-33
.
Future studies will be required to elucidate the mechanism by which IgA controls and
modifies gut bacterial colonization in newborns. IgA has been shown to modify bacterial
surface protein expression and motility which may limit the ability of bacteria to
gain access to the intestinal epithelium
17,34
. IgA may accomplish these tasks by ‘enchaining’ bacterial cells, allowing for easier
expulsion and preventing gene transfer
35
. Importantly, the current study did not discriminate between bacteria at the strain
level. Thus, it remains to be determined whether the loss of IgA binding results from
the appearance of new organisms not constrained by the existing IgA repertoire, or
alternatively from changes to bacterial genomes and/or gene expression that allow
early colonizers to escape IgA binding
36,37
. Temporal changes in bacterial binding could also result from shifts in the maternal
IgA repertoire.
Previous attempts to prevent NEC with intravenous immunoglobulins have largely failed
to show efficacy
38,39
. However, the repertoire of intravenous antibodies may differ from that of secretory
IgA and bacterial specificity was not accounted for in these studies. Future efforts
might allow for precision microbiome-informed strategies that enable augmentation
of milk or preterm infant formulas with rationally selected protective antibodies.
Methods
Mice
C57BL/6 mice were purchased from Taconic. Rag1−/− mice were obtained from Jackson
Laboratories. Igha−/− mice were obtained from Dr. Yasmine Belkaid (NIH/NIAID). All
mice were maintained at and all experiments were performed in an American Association
for the Accreditation of Laboratory Animal Care-accredited animal facility at the
University of Pittsburgh and housed in accordance with the procedures outlined in
the Guide for the Care and Use of Laboratory Animals under an animal study proposal
approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh.
Mice were housed in specific pathogen-free (SPF) conditions.
Human Fecal Samples
The human study protocol was approved by the Institutional Review Board (Protocol
Nos. PRO16030078, PRO09110437) of the University of Pittsburgh. Fecal samples were
collected fresh or from the diaper of preterm infants at UPMC Magee-Womens Hospital
and frozen immediately at −80°C. The samples were later divided into age-matched controls
and NEC depending on the incidence of NEC.
Fecal IgA Flow Cytometry and Magnetic Sorting of IgA+ and IgA− Bacteria
Either fecal pellets collected from mice after sacrifice or ~50 mg of frozen human
fecal material was placed in 1.5ml Eppendorf tubes and 1ml Phosphate Buffered Saline
(PBS) was added. The fecal material was disrupted by a combination of vortexing and
pipetting and passed through a 40μm filter to remove food/fibrous material. The fecal
material is diluted with PBS to obtain a bacterial OD of ~0.4 to maintain equality
between samples and to prevent the magnetic columns from clogging. A volume of 200μl
of the suspended bacterial material was then frozen as an ‘unsorted’ control. An additional
200μl of the suspended material was divided equally on a 96-well plate for IgA staining
and isotype control for each sample to eliminate non-specific binding. The fractions
were washed with twice with staining buffer (1% Bovine Serum Albumin (Sigma) in PBS-filtered
through a 2.2μm filter). The bacteria were stained with Syto BC (Green Fluorescent
nuclear acid stain, Invitrogen-1:400), APC Anti-Human IgA (Miltenyi Biotec clone IS11-8E10)
(1:10)/ Anti-Human IgA APC (Miltenyi Biotec clone REA1014) (1:50), Anti-Human IgM
BV421 (BD Biosciences clone G20-127) (1:30)/ BV421 Mouse Anti-Human IgG (BD Biosciences
clone G18-145) (1:10) or PE-conjugated Anti-Mouse IgA (eBioscience clone mA-6E1) (1:500),
Anti-Mouse Rat IgM BV421 (BD Biosciences clone R6-60.2) (1:30)/ Anti-Mouse Rat IgG2a
isotype (BD Biosciences clone R35-95), Anti-Mouse IgG FITC (BioLegend clone Poly4060)
(1:30) and blocking buffer of 20% Normal Mouse Serum for human or 20% Normal Rat Serum
for mouse samples (ThermoFisher). The isotype control was stained similarly using
APC Mouse IgG1 isotype control (Miltenyi Biotec clone-IS5-21F5) (1:10) or PE-conjugated
Rat Anti-Mouse IFNγ (eBioscience clone XMG1.2). The stained samples were incubated
in the dark for an hour at 4°C. Samples were then washed three times with 200μl of
staining buffer before flow-cytometric analysis (LSRFortessa-BD Biosciences).
For magnetic activated cell sorting (MACS), we used 500μl of the suspended fecal material
to compensate for the loss of material during sorting and scaled our staining volume
accordingly. Anti-IgA stained fecal bacterial pellets were incubated in 1ml per sample
of staining buffer containing 45μl of anti-APC or anti-PE MACS Microbeads (Miltenyi
Biotec) (20 min at 4°C in the dark), washed twice with 1 ml Staining Buffer (8000
x rpm, 5 min, 4°C), and then sorted using MS columns (Miltenyi Biotec). The flow-through
was collected as IgA-unbound (IgA-negative) fraction. Columns were washed with 70%
ethanol and sterile PBS between separations. The IgA-bound fraction was added in the
column and the steps mentioned above were repeated four times for maximum enrichment.
100μl each of the IgA-bound and IgA-unbound fraction was used for post-sort flow cytometric
analysis (along with unsorted sample). Absolute bacterial counts were determined by
adding a known number of AccuCheck Counting beads (Life Technologies) to antibody
stained fecal samples of a given mass, which allows for the calculation of the total
number of SYTO (DNA)+ events in any given sample. This can then be multiplied by the
measured abundance of any OTU to represent the number of bacteria of that taxon/mass
in any sample.
DNA Extraction
All microbial DNA was extracted using the MO BIO PowerSoil DNA Isolation kit (single
tube extractions). The unsorted, IgA-bound and IgA-unbound pellets were resuspended
in Solution TD1 by pipetting and vortexing and ~200μl of 0.1mm diameter Zirconia/Silica
beads (Biospec) were added and shaken horizontally on a lab mixer for 12-18 min at
maximum speed using a MO BIO vortex adaptor. All remaining steps followed the manufacturer’s
protocol. The DNA extracted was stored at −20°C for further 16S amplicon PCR and sequencing.
16S Amplicon PCR, Sequencing and Analysis
PCR amplification of the small subunit ribosomal RNA gene (16S rRNA) was performed
in triplicate 25μl reactions. Reactions were held at 94°C for 3 min to denature the
DNA, with amplification performed for 30 cycles at 94°C for 45 s, 50°C for 60 s, and
72°C for 90 s; followed by a final extension of 10 min at 72°C. Amplicons were produced
utilizing primers adapted for the Illumina MiSeq. Amplicons target the V4 region and
primers utilized either the Illumina adaptor, primer pad and linker (forward primer)
or Illumina adaptor, Golay barcode, primer pad and linker (reverse primer) followed
by a sequence targeting a conserved region of the bacterial 16S rRNA gene as described
40,41
. The only deviation from the protocol was that PCR was run for 30 cycles. Amplicons
were cleaned using the Qiagen UltraClean 96 PCR Cleanup Kit. Quantification of individual
amplicons was performed with the Invitrogen Quant-iT dsDNA High Sensitivity Assay
Kit. Amplicons were then pooled in equimolar ratio. Agarose gel purification was performed
to further purify the amplicon pool and remove undesired PCR products prior to submission
for paired-end sequencing on the Illumina MiSeq. Read pairing, clustering and core
diversity statistics were generated through PEAR, UPARSE and QIIME and R
42,43
. LEfSe was used to compare family level relative abundances between NEC and control
groups
44
. Raw 16S rRNA data has been uploaded to NCBI BioSample/SRA and is available under
accession number PRJNA526906.
Deconvolution and microbiome data analysis:
Flow cytometry was used to determine the percentage of IgA positive and IgA negative
bacteria in each sample (unsorted, IgA positive, IgA negative) post magnetic separation.
We assumed that contamination affected each OTU equally and that the IgA positive
and IgA negative samples are reciprocal (fractions of the same whole). The raw reads
from sequencing of 16S rRNA genes were then deconvolved by summing the proportion
of IgA bound or IgA unbound (as measured by flow cytometry) across the paired (infant
and time point [Day of life]) IgA positive and IgA negative samples for each OTU (Extended
Data 3c).
Example to solve for IgA positive from one time point and one OTU ‘X’:
Total IgA positive
‘
X
’
=
‘
X
’
correctly bound to IgA positive +
‘
X
’
contaminating IgA
negative
Total IgA positive
‘
X
’
=
(
%
IgA+ in bound
)
∗
(
#
‘
X
’
reads in bound
)
+
(
%
IgA+ in
unbound
)
∗
(
#
‘
X
’
reads in unbound
)
The deconvoluted data was processed through the QIIME2 workflow to create alpha diversity
metrics with sampling depth chosen based on alpha rarefaction plotting. Abundance
of individual or ‘pooled’ (Anaerobes) OTUs was then calculated using the deconvolved
values. The algorithm for deconvolution is available on GitHub (https://github.com/handlab/IgA_Seq_Deconvolution).
If IgSeq is accurate, the ratio of the read numbers between IgA positive and negative
(IgA+/IgA++IgA−) samples should roughly correspond to the percent IgA+ bacteria in
the unsorted sample from where they were derived. This relationship does not hold
for samples with low levels of IgA positive bacteria (Extended Data 4c) prior to deconvolution
but is much improved after our method has been applied (Extended Data 4d)
We categorized all of the following OTUs as ‘Anaerobes’: Bifidobacteriaceae, Prevotellaceae,
Bacteroidiales_S24-7, Clostridiaceae, Lachnospiraceae, Peptostreptococcaceae, Ruminococcaceae,
Veillonellaceae, Tissierellaceae (Figure 2e and Extended Data 6b).
Quantitative PCR for 16S rRNA.
PCR amplification of the small subunit ribosomal RNA gene (16S rRNA) was performed
in triplicate 10μl reactions. Reactions were held at 95°C for 3 min to denature the
DNA, with amplification performed for 35 cycles (95°C for 10 s and 60°C for 30 s).
The forward primer sequence of 16S is ACTCCTACGGGAGGCAGCAGT and the reverse primer
sequence of 16S ATTACCGCGGCTGCTGGC.
Quantitative PCR for Enterobacter spp.
PCR amplification of the small subunit ribosomal RNA gene (23S rRNA) was performed
in triplicate 10μl reactions. Reactions were held at 95°C for 3 min to denature the
DNA, with amplification performed for 35 cycles (95°C for 10 s and 60°C for 30 s).
The forward primer sequence of Enterobacter 23S is AGTGGAACGGTCTGGAAAGG and the reverse
primer sequence of Enterobacter 23S TCGGTCAGTCAGGAGTATTTAGC
45
.
Induction of NEC
NEC is induced in 7- to 8-day-old mice by hand-feeding mice formula via gavage 5 times/day
(22-gauge needle; 200μl volume; Similac Advance infant formula [Ross Pediatrics, Columbus,
Ohio]/ Esbilac canine milk replacer 2:1). The formula is supplemented with 107 CFUs
of Enterobacter spp. (99%) and Enterococcus spp. (1%) and mice are rendered hypoxic
(5%O2, 95% N2) for 10 minutes in a hypoxic chamber (Billups-Rothenberg, Del Mar, CA)
twice daily for 4 days
46,47
. We used males and females in all experiments. Disease is monitored by weighing mice
daily prior to the second feed. The severity of disease was determined on histologic
sections of the entire length of the small intestines stained with hematoxylin and
eosin by trained personnel who were blinded to the study conditions according to previously
published scoring system from 0 (normal) to 4(severe)
48
.
Statistics
Statistical tests used are indicated in the figure legends. Lines in scatter bar charts
represent the mean of that group. Group sizes were determined based on the results
of preliminary experiments. Mouse studies were performed in a non-blinded fashion.
Statistical significance was determined with the two-tailed unpaired Student’s t-test
or non-parametric Mann-Whitney test when comparing two groups and one-way ANOVA with
multiple comparisons, when comparing multiple groups. All statistical analyses were
calculated using Prism software (GraphPad). Differences were considered to be statistically
significant when p < 0.05.
Data Availability Statement
Patient-related data not included in the paper were generated as part of clinical
trials and may be subject to patient confidentiality. The human study protocol was
approved by the Institutional Review Board (Protocol Nos. PRO16030078, PRO09110437)
of the University of Pittsburgh. All raw and analyzed sequencing data can be found
at the NCBI Sequence Read Archive (accession number: PRJNA526906). Algorithm for deconvolution
of IgSeq data available on GitHub (https://github.com/handlab/IgA_Seq_Deconvolution).
Biological Materials Availability Statement
All materials used in the production of this paper are available upon request (timothy.hand@chp.edu).
Some reagents may require a Material Transfer Agreement through the University of
Pittsburgh.
Extended Data
Extended Data Fig. 1.
Maternal milk-derived antibodies binding to intestinal bacteria from preterm infants.
Extended Data Fig. 2.
Fraction of intestinal bacteria bound by IgA in preterm infants.
Extended Data Fig. 3.
Linear discriminant analysis of the microbiota of infants that will develop NEC and
controls.
Extended Data Fig. 4.
Deconvolution method to decrease the effect of contamination in IgSeq.
Extended Data Fig. 5.
Longitudinal analysis of the intestinal microbiota of preterm infants.
Extended Data Fig. 6.
Ratio of IgA− to IgA+ reads for low-abundance taxa.
Extended Data Fig. 7.
Absolute number of bacteria and number of bacteria associated with the dominant taxa
in preterm infants.
Extended Data Fig. 8.
Enterobacter spp. is enriched in the IgA+ fraction of breast-fed mouse pups.