The puzzle in the bottle
Human milk is not mostly sugar in the way a soft drink is, but it contains a lot of carbohydrate that the infant cannot use. Human milk oligosaccharides (HMOs) are the third-largest solid component of human milk, after lactose and fat, and the infant gut does not produce the glycosidases needed to cleave their linkages (Underwood et al., Pediatr Res, 2015). The mother spends energy building complex glycans; the baby’s own enzymes cannot take them apart. If you are looking for an evolutionary reason for a symptom or a biological oddity, this is one of the cleanest examples: a costly maternal investment with no direct nutritional return to the child.
The trade-off, named plainly: maternal metabolic expenditure on indigestible glycans versus direct nutritive yield to the infant. The return is indirect. HMOs arrive largely intact in the distal small bowel and colon, where they shape which microbes get to eat (Underwood et al., 2015). The infant does not digest the sugars; the infant’s microbial partners do, and the infant gets the consequences.
What the sugars actually are
HMOs are built from five monosaccharides — glucose, galactose, N-acetylglucosamine, fucose and sialic acid — in varied chains and branches. There are more than 150 structures, with 2′-fucosyllactose (2′-FL), 3′-fucosyllactose (3′-FL), 6′-sialyllactose (6′-SL), lacto-N-tetraose (LNT) and lacto-N-neotetraose (LNnT) among the most abundant (Mulakala et al., mSystems, 2026). Maternal genetics influence the mix: non-secretor mothers produce no 2′-FL or longer-chain 2-fucosylated HMOs (Mulakala et al., 2026). Milk is not one fixed formula; it is a variable substrate.
That variability matters for the trade-off. If the point of HMOs were simply bulk fibre, the exact structures would not matter much. They do matter, because the microbial capacity to consume them is structure-specific.
Who eats them, and how specifically
Among the microbial species studied, only two genera — Bifidobacterium and Bacteroides — are able to comprehensively use HMOs as a primary food source (Underwood et al., 2015). Within Bifidobacterium, the standout is Bifidobacterium longum subsp. infantis (B. infantis). Most strains sequenced to date carry a 43-kb gene cluster (HMO cluster I) encoding oligosaccharide transporters and glycosyl hydrolases that is not found in other bifidobacterial species (Underwood et al., 2015). Among the bifidobacteria tested, only B. infantis is able to digest all HMO structures (Underwood et al., 2015).
The specificity runs down to single glycosidic bonds. LNT and LNnT are both abundant HMO tetrasaccharides and are isomers differing by one linkage. When B. infantis grows on either, the ratio of secreted acetate to lactate shifts to 1.7–2.0, compared with about 1.5 on their component carbohydrates. Inefficient metabolism of LNnT prompts the bacterium to shunt carbon toward formic acid and ethanol (Özcan and Sela, Front Nutr, 2018). One bond changes the fermentation profile. That is the kind of detail a coevolutionary account has to accommodate: the substrate and the consumer are matched at a resolution finer than “sugar.”
What the infant gets — and what the evidence actually shows
The honest answer is that the infant gets a shaped gut environment, and the downstream health claims are strongest in model systems and weakest in term-infant outcomes. In vitro, B. infantis grows better than other strains on HMOs, shows anti-inflammatory activity in premature intestinal cells, and decreases intestinal permeability (Underwood et al., 2015). Bifidobacterial metabolites reduced LPS-induced inflammation in a cell-culture model (Özcan and Sela, 2018). In germ-free mice, mono-colonization with B. infantis increased monocytes, macrophages, B cells, CD4+ T cells and regulatory T cells in mesenteric lymph nodes relative to controls (Mulakala et al., 2026).
These are model-system findings. They are consistent with the trade-off story; they are not proof that a given infant will be healthier because of a given HMO profile. The retrieved literature does not report term-infant outcome trials of B. infantis or HMO supplementation, and it does not show that HMO-supplemented formula reproduces the effects of human milk. The mouse work is framed as helping improve formula composition, not as demonstrating equivalence (Mulakala et al., 2026).
One more boundary: the sources describe HMOs as having likely mediated the coevolution of the nursing infant with its gut microbiome (Özcan and Sela, 2018). That is a claim about coevolution, not about purpose. Milk was not designed to feed B. infantis; the two lineages shaped each other, and the trade-off is the accounting.
Why this is a trade-off and not a paradox to dissolve
It is tempting to say the puzzle is solved: the sugars feed the good bacteria. Resist the tidy ending. The maternal cost is real and paid whether or not the infant’s microbiome responds as expected. The benefit is contingent on which microbes are present, which depends on delivery mode, feeding, antibiotics and a hundred other things. A trade-off can be adaptive on average and still be a bad deal for a particular dyad. Naming it as a trade-off keeps the explanation honest and keeps it from sliding into teleology.
The framework also generates a prediction worth stating as reasoning, not as a finding: if milk-glycan diversity and microbial catabolic capacity coevolved, they should be matched. A strain lacking the full HMO gene cluster is not automatically equivalent to one that has it, and a single added glycan is not automatically equivalent to the native mixture. That is a way to read new studies, not a clinical recommendation.
The Quebec anchor, scoped honestly
Quebec has an extraordinary resource for population-genetics questions: the Balsac population database at UQAC, developed since 1971, contains data on more than 6 million known individuals from 1621 to the present, reconstructed from civil registry records (baptisms, marriages, burials) and linked into families and lineages (Balsac). It is used by researchers in Quebec and elsewhere in both the social sciences and the biomedical sciences (Balsac). The GenoRef-Q initiative, for example, will sequence the genome of 1,900 participants recruited in CARTaGENE, the population-based cohort of the Province of Quebec, and will work in collaboration with Balsac (Balsac, About).
What Balsac is not is a source of HMO or infant-microbiome findings. It can support questions about ancestry and genetic structure in Quebec populations; it does not contain milk-glycan data. Saying so plainly is part of keeping the local anchor useful rather than decorative. If a future study links Balsac genealogies to infant-feeding or microbiome data, that would be a different article.
A hypothetical worth labelling as such
Imagine two mothers with different secretor status, producing different HMO profiles. The framework predicts different substrate availability for B. infantis. The retrieved sources do not measure infant outcomes by secretor status, so the example stays hypothetical. It is useful for teaching the logic of the trade-off; it is not evidence of a clinical difference.
What would change the answer
Three kinds of evidence would move this from a well-supported mechanism to a well-supported clinical story. First, a term-infant cohort or trial linking HMO profiles by maternal secretor status to infant microbiome composition and health outcomes. Second, a trial of B. infantis or HMO supplementation in term infants with clinical endpoints, not just colonization or metabolite shifts. Third, a Quebec-specific dataset connecting infant feeding or microbiome measures to administrative or cohort data. Until then, the trade-off is the right frame and the health claims stay at the level the models support.
FAQ
If babies cannot digest HMOs, why are they in milk at all?
Because the infant is not the only consumer. HMOs reach the distal small bowel and colon largely intact and feed a narrow set of microbes, especially B. infantis, which carries the genetic equipment to break them down (Underwood et al., 2015). The benefit to the infant is indirect and mediated by the microbial community.
Does HMO-supplemented formula do the same thing as breast milk?
The retrieved evidence does not show equivalence. Mouse and cell models suggest that HMOs and B. infantis can shape immune and metabolic markers, and the mouse work is explicitly framed as helping improve formula composition (Mulakala et al., 2026). That is not the same as demonstrating that a supplemented formula reproduces human-milk effects in infants.
Is this an example of Quebec population genetics?
Not directly. Balsac is a population-genetics and historical-demography infrastructure, not an HMO or infant-microbiome dataset (Balsac). The connection to this topic is methodological: both fields ask how variation — in genes or in milk glycans — meets variation in the environment.
Why call it a trade-off instead of a design feature?
Because the maternal cost is paid regardless of outcome, and the benefit depends on which microbes are present. “Trade-off” describes the accounting without implying that milk was built for a particular bacterium. The sources describe coevolution, not purpose (Özcan and Sela, 2018).