Characterization of the Diversity and Temporal Stability of Bacterial Communities in Human Milk

Katherine M Hunt, James A. Foster, Larry J. Forney, Ursel M. E. Schütte, Daniel Beck, Zaid Abdo, Lawrence K. Fox, Janet E. Williams, Michelle K. McGuire, Michelle K. McGuire, Mark A. McGuire, Mark A. McGuireView original
OverviewBalancedalloy voice
For a long time, people talked about human milk as if it were sterile — pure nutrition, no passengers. That picture has cracked. When labs stopped relying on petri dishes and started reading DNA, they found something surprising: milk carries its own bacterial community. Not a smear of contamination, but a living set of microbes with patterns, personalities, and potentially, consequences. So here's the setup. Hunt and colleagues followed sixteen healthy lactating women who said they were free of mastitis. Each woman picked one breast to sample, and over roughly four weeks, they gave three morning milk samples. The team took skin cleaning seriously — iodine swabs, sterile collection, full expression — because if you're going to claim there's a milk microbiome, you have to show it's not just skin rubbing off. Then, they bypassed culture altogether and sequenced a bacterial barcode — the V1 to V2 region of the 16S ribosomal RNA gene. They used barcoded pyrosequencing on the old but still useful 454 platform. Out of about three hundred thousand raw reads, around one hundred sixty thousand high-quality sequences made the cut after conservative filtering. That worked out to roughly three and a half thousand sequences per sample. They grouped sequences into operational taxonomic units, or OTUs — think of them as proxy species — at a 3 percent similarity threshold and assigned names with a standard Bayesian classifier. What came back was not a monoculture. It was a crowd. Across women and time points, the most abundant genera were familiar names in human-associated microbiology: Streptococcus, Staphylococcus, Serratia, and Corynebacterium. Those weren't the whole story, though. There was a long supporting cast — Propionibacterium and others — that often gets missed when you only grow what's easy to grow. Here's a number to anchor it: averaged across all samples, Staphylococcus came in at about 16 percent of reads, with Streptococcus around 8 percent. And the darlings of probiotic yogurt ads, Lactobacillus and Bifidobacterium? They were there, but they were a small slice — on the order of 2 to 3 percent — which contrasts with some European reports that found more of them. Diversity also wasn't a one-size-fits-all situation. When the team counted how many OTUs each woman harbored — again, those species proxies at a 3 percent cutoff — they found anywhere from about one hundred to six hundred. That's a six-fold swing in richness across healthy individuals. And beneath the variety, a pattern emerged. There was a tiny set of nine OTUs that showed up in every single sample from every single woman. Only nine. But collectively, they accounted for about half of all the bacterial reads. In other words, there's a core — a set of regulars — and around it, a rotating cast that changes from person to person and, sometimes, from week to week. How steady was this community within a mother over a month? It depended on the mother. Some women were strikingly consistent. If Staphylococcus was a major player in week one, it stayed a major player in weeks two and three. Subject five is a nice example: Staphylococcus held between roughly 22 and 59 percent across her three samples. For others, the deck shuffled. Subject one had Staphylococcus under about 5 percent every time. And a few women showed real swings in who dominated from one visit to the next. When the team used a standard ecological similarity measure to cluster samples, some mothers' three time points tucked together neatly, and others scattered — a sign that their milk communities were shifting. Here's a nuance that matters for how you think about stability. If you look at presence and absence — which OTUs are there at all — only a minority persisted across all time points within a woman. On average, between 4 and 20 percent of her OTUs were present every time. But those persistent OTUs weren't bit players. They accounted for the bulk of the community's abundance — anywhere from 60 to 99 percent of all reads in that woman's samples. So milk communities can change at the edges — the rare stuff comes and goes — while the core that dominates the biomass holds on. That's a very human microbiome kind of pattern. Now, it's one thing to list names and percentages. It's another to connect this to health. The study wasn't designed to diagnose disease, but one data point jumped out. In Subject thirteen, the first sample was almost a monoculture: about 95 percent Streptococcus. That same sample had a roughly five-fold increase in somatic cell count — that's a measure of immune cells in milk and a classic marker of inflammation in the mammary gland. She didn't report mastitis, but that spike suggests the microbiome might flag subclinical inflammation, or even contribute to it. It's a single case inside a small study, so take it as a clue, not a verdict. Still, it's the kind of clue clinicians care about. If you're thinking, "But could this just be skin?" the team worried about that too. They did skin cleansing. They collected deep into a feed. And they paired sequencing with old-school culture using mastitis-screening protocols. Those plates told a narrower, sometimes contradictory story. Many samples grew colonies of the usual suspects — coagulase-positive and coagulase-negative Staphylococcus, some Streptococcus, some Corynebacterium, a few coliforms. But roughly one in five samples didn't grow anything under those conditions. No colonies after two days at 37 degrees Celsius. Meanwhile, sequencing of the exact same milk told you there were dozens to hundreds of taxa. Culture is great at telling you what thrives on the media you chose. It's not great at telling you what's present. Earning trust in sequencing meant being picky. Hunt's group ran a harsh quality pipeline: they tossed reads with low average quality, excessive homopolymers, barcode or primer mismatches, and they used chimera detection to remove Frankenstein sequences created by the polymerase. They aligned everything to a curated reference database and analyzed communities in mothur, a standard toolkit in microbial ecology. That rigor matters because when you say "milk has six hundred OTUs", you want to know you're not just counting sequencing error. And still, the authors are clear about potential biases. Primer choice can tilt the playing field toward or away from certain groups. Geography and diet differ between populations, which could explain why some European studies reported more Lactobacillus and Bifidobacterium. And even with iodine, you can't fully rule out skin-derived bacteria. The safest takeaway is this: the signal is real, but the exact composition you see will depend on both biology and method. The other methodological point that shapes interpretation is how you define "stability." In this dataset, stability wasn't an all-or-nothing label. Some women's three time points clustered tightly by community composition; others didn't. Some had a single dominant genus that stayed dominant; others saw handoffs between Streptococcus, Staphylococcus, and Serratia across weeks. Across the cohort, the common structure was a small core of OTUs that almost always persisted, plus a much larger halo that flickered. That means one baby's daily exposure could be fairly consistent, while another's might vary. Same feeding, different microbial background. Put that in the context of infant health, and the stakes get obvious. A breastfed infant is gulping not just calories and antibodies, but hundreds of milk-derived bacterial types. Those microbes are one input into the early gut ecosystem, which in turn is tied to protection against diarrheal disease, respiratory infections, and even long-term metabolic outcomes like obesity risk. How much of that protection comes from the bugs versus the bioactive molecules is an open question. But the presence of a personal, somewhat stable milk microbiome makes a testable path: if the baseline community changes, do infant gut communities and health outcomes shift too? There's also a maternal angle. Clinicians already use somatic cell counts to watch for inflammation in dairy animals, and to a lesser extent in people. This study shows you can sometimes see a microbial signature alongside those immune changes. A community suddenly dominated by one genus — like that 95 percent Streptococcus snapshot — might be a red flag for a system under stress. And because the persistent OTUs account for so much of the biomass, interventions that nudge that core — through antibiotics, probiotics, or even pumping hygiene — could have outsized effects. That's speculation until it's tested, but it's the kind you can design trials around. One lingering question is source. Are these milk microbes just hitchhikers from skin, or do they have a pipeline into the mammary gland? Hunt's group notes two plausible routes. First, skin is always in contact with the nipple and areola, so some transfer is unavoidable. Second, during nursing, there's retrograde flow — milk and saliva can move back into the ducts — which would allow exchange between the infant mouth and the mammary environment. Neither route rules out the other, and both would make milk a hub in a shared mother and infant microbial network. That's a very different mental model than "sterile milk plus one-way exposure." Where does that leave us? With a reframed baseline. In healthy, mastitis-free women, milk hosts diverse bacterial communities. These communities vary wildly in richness — roughly one to six hundred OTUs per person — yet they share a tiny core of nine OTUs that together make up about half the total abundance. Within a mother, only a small fraction of OTUs show up every week, but that persistent minority accounts for the lion's share of cells. And when you compare petri dishes to DNA, culture misses a lot: about 20 percent of samples had no growth on standard media even though sequencing found rich communities. That's the map as of today. Two brief looks ahead. First, the obvious next step is to follow mother and infant pairs with denser sampling and higher-resolution tools. Amplicon sequencing gets you to genera and OTUs; shotgun metagenomics can give you strain-level resolution and functional genes. Pair that with immune readouts — somatic cell count, cytokines — and with careful metadata on delivery mode, antibiotics, diet, and geography, and you can start making causal connections. Second, method matters. Harmonizing primers and pipelines across cohorts, or at least calibrating them against shared mock communities, would help explain why some studies see more of certain groups than others. But the headline doesn't wait for perfect methods. Milk isn't sterile. It carries a personal, partially stable microbiome. In sixteen women over four weeks, Hunt and colleagues saw the same core nine OTUs every time. They also saw a handful of dominant genera, led by Staphylococcus around 16 percent and Streptococcus about 8 percent on average. And there was a crowd of rarer taxa that flickered in and out. One subject's near-monoculture aligned with a five-fold spike in inflammation. Culture alone missed many of these signals. Put that all together, and you get a richer, messier, and more interesting picture of what an infant actually drinks — and a new set of questions about how those microbes matter for two intertwined bodies.

For a long time, people talked about human milk as if it were sterile — pure nutrition, no passengers. That picture has cracked. When labs stopped relying on petri dishes and started reading DNA, they found something surprising: milk carries its own bacterial community.

Not a smear of contamination, but a living set of microbes with patterns, personalities, and potentially, consequences.

So here's the setup. Hunt and colleagues followed sixteen healthy lactating women who said they were free of mastitis. Each woman picked one breast to sample, and over roughly four weeks, they gave three morning milk samples.

The team took skin cleaning seriously — iodine swabs, sterile collection, full expression — because if you're going to claim there's a milk microbiome, you have to show it's not just skin rubbing off. Then, they bypassed culture altogether and sequenced a bacterial barcode — the V1 to V2 region of the 16S ribosomal RNA gene. They used barcoded pyrosequencing on the old but still useful 454 platform.

Out of about three hundred thousand raw reads, around one hundred sixty thousand high-quality sequences made the cut after conservative filtering. That worked out to roughly three and a half thousand sequences per sample. They grouped sequences into operational taxonomic units, or OTUs — think of them as proxy species — at a 3 percent similarity threshold and assigned names with a standard Bayesian classifier.

What came back was not a monoculture. It was a crowd. Across women and time points, the most abundant genera were familiar names in human-associated microbiology: Streptococcus, Staphylococcus, Serratia, and Corynebacterium.

Those weren't the whole story, though. There was a long supporting cast — Propionibacterium and others — that often gets missed when you only grow what's easy to grow. Here's a number to anchor it: averaged across all samples, Staphylococcus came in at about 16 percent of reads, with Streptococcus around 8 percent.

And the darlings of probiotic yogurt ads, Lactobacillus and Bifidobacterium? They were there, but they were a small slice — on the order of 2 to 3 percent — which contrasts with some European reports that found more of them.

Diversity also wasn't a one-size-fits-all situation. When the team counted how many OTUs each woman harbored — again, those species proxies at a 3 percent cutoff — they found anywhere from about one hundred to six hundred. That's a six-fold swing in richness across healthy individuals.

And beneath the variety, a pattern emerged. There was a tiny set of nine OTUs that showed up in every single sample from every single woman. Only nine.

But collectively, they accounted for about half of all the bacterial reads. In other words, there's a core — a set of regulars — and around it, a rotating cast that changes from person to person and, sometimes, from week to week.

How steady was this community within a mother over a month? It depended on the mother. Some women were strikingly consistent.

If Staphylococcus was a major player in week one, it stayed a major player in weeks two and three. Subject five is a nice example: Staphylococcus held between roughly 22 and 59 percent across her three samples. For others, the deck shuffled.

Subject one had Staphylococcus under about 5 percent every time. And a few women showed real swings in who dominated from one visit to the next. When the team used a standard ecological similarity measure to cluster samples, some mothers' three time points tucked together neatly, and others scattered — a sign that their milk communities were shifting.

Here's a nuance that matters for how you think about stability. If you look at presence and absence — which OTUs are there at all — only a minority persisted across all time points within a woman. On average, between 4 and 20 percent of her OTUs were present every time.

But those persistent OTUs weren't bit players. They accounted for the bulk of the community's abundance — anywhere from 60 to 99 percent of all reads in that woman's samples. So milk communities can change at the edges — the rare stuff comes and goes — while the core that dominates the biomass holds on. That's a very human microbiome kind of pattern.

Now, it's one thing to list names and percentages. It's another to connect this to health. The study wasn't designed to diagnose disease, but one data point jumped out.

In Subject thirteen, the first sample was almost a monoculture: about 95 percent Streptococcus. That same sample had a roughly five-fold increase in somatic cell count — that's a measure of immune cells in milk and a classic marker of inflammation in the mammary gland. She didn't report mastitis, but that spike suggests the microbiome might flag subclinical inflammation, or even contribute to it.

It's a single case inside a small study, so take it as a clue, not a verdict. Still, it's the kind of clue clinicians care about.

If you're thinking, "But could this just be skin?" the team worried about that too. They did skin cleansing. They collected deep into a feed.

And they paired sequencing with old-school culture using mastitis-screening protocols. Those plates told a narrower, sometimes contradictory story. Many samples grew colonies of the usual suspects — coagulase-positive and coagulase-negative Staphylococcus, some Streptococcus, some Corynebacterium, a few coliforms.

But roughly one in five samples didn't grow anything under those conditions. No colonies after two days at 37 degrees Celsius. Meanwhile, sequencing of the exact same milk told you there were dozens to hundreds of taxa.

Culture is great at telling you what thrives on the media you chose. It's not great at telling you what's present.

Earning trust in sequencing meant being picky. Hunt's group ran a harsh quality pipeline: they tossed reads with low average quality, excessive homopolymers, barcode or primer mismatches, and they used chimera detection to remove Frankenstein sequences created by the polymerase. They aligned everything to a curated reference database and analyzed communities in mothur, a standard toolkit in microbial ecology.

That rigor matters because when you say "milk has six hundred OTUs", you want to know you're not just counting sequencing error. And still, the authors are clear about potential biases. Primer choice can tilt the playing field toward or away from certain groups.

Geography and diet differ between populations, which could explain why some European studies reported more Lactobacillus and Bifidobacterium. And even with iodine, you can't fully rule out skin-derived bacteria. The safest takeaway is this: the signal is real, but the exact composition you see will depend on both biology and method.

The other methodological point that shapes interpretation is how you define "stability." In this dataset, stability wasn't an all-or-nothing label. Some women's three time points clustered tightly by community composition; others didn't. Some had a single dominant genus that stayed dominant; others saw handoffs between Streptococcus, Staphylococcus, and Serratia across weeks.

Across the cohort, the common structure was a small core of OTUs that almost always persisted, plus a much larger halo that flickered. That means one baby's daily exposure could be fairly consistent, while another's might vary. Same feeding, different microbial background.

Put that in the context of infant health, and the stakes get obvious. A breastfed infant is gulping not just calories and antibodies, but hundreds of milk-derived bacterial types. Those microbes are one input into the early gut ecosystem, which in turn is tied to protection against diarrheal disease, respiratory infections, and even long-term metabolic outcomes like obesity risk.

How much of that protection comes from the bugs versus the bioactive molecules is an open question. But the presence of a personal, somewhat stable milk microbiome makes a testable path: if the baseline community changes, do infant gut communities and health outcomes shift too?

There's also a maternal angle. Clinicians already use somatic cell counts to watch for inflammation in dairy animals, and to a lesser extent in people. This study shows you can sometimes see a microbial signature alongside those immune changes.

A community suddenly dominated by one genus — like that 95 percent Streptococcus snapshot — might be a red flag for a system under stress. And because the persistent OTUs account for so much of the biomass, interventions that nudge that core — through antibiotics, probiotics, or even pumping hygiene — could have outsized effects. That's speculation until it's tested, but it's the kind you can design trials around.

One lingering question is source. Are these milk microbes just hitchhikers from skin, or do they have a pipeline into the mammary gland? Hunt's group notes two plausible routes.

First, skin is always in contact with the nipple and areola, so some transfer is unavoidable. Second, during nursing, there's retrograde flow — milk and saliva can move back into the ducts — which would allow exchange between the infant mouth and the mammary environment. Neither route rules out the other, and both would make milk a hub in a shared mother and infant microbial network. That's a very different mental model than "sterile milk plus one-way exposure."

Where does that leave us? With a reframed baseline. In healthy, mastitis-free women, milk hosts diverse bacterial communities.

These communities vary wildly in richness — roughly one to six hundred OTUs per person — yet they share a tiny core of nine OTUs that together make up about half the total abundance. Within a mother, only a small fraction of OTUs show up every week, but that persistent minority accounts for the lion's share of cells. And when you compare petri dishes to DNA, culture misses a lot: about 20 percent of samples had no growth on standard media even though sequencing found rich communities. That's the map as of today.

Two brief looks ahead. First, the obvious next step is to follow mother and infant pairs with denser sampling and higher-resolution tools. Amplicon sequencing gets you to genera and OTUs; shotgun metagenomics can give you strain-level resolution and functional genes.

Pair that with immune readouts — somatic cell count, cytokines — and with careful metadata on delivery mode, antibiotics, diet, and geography, and you can start making causal connections. Second, method matters. Harmonizing primers and pipelines across cohorts, or at least calibrating them against shared mock communities, would help explain why some studies see more of certain groups than others.

But the headline doesn't wait for perfect methods. Milk isn't sterile. It carries a personal, partially stable microbiome.

In sixteen women over four weeks, Hunt and colleagues saw the same core nine OTUs every time. They also saw a handful of dominant genera, led by Staphylococcus around 16 percent and Streptococcus about 8 percent on average. And there was a crowd of rarer taxa that flickered in and out.

One subject's near-monoculture aligned with a five-fold spike in inflammation. Culture alone missed many of these signals. Put that all together, and you get a richer, messier, and more interesting picture of what an infant actually drinks — and a new set of questions about how those microbes matter for two intertwined bodies.

More in Nursing