Ultrafiltered Milk Rewrites the Microbial and Flavor Chemistry of Cheese

Cheese is one of humanity’s oldest fermentation technologies, yet its industrial production still carries an inconvenient secret: for every kilogram of cheese made, vast quantities of whey are drained away, taking valuable proteins with it and leaving behind effluent loaded with biochemical and chemical oxygen demand. A new study published in Food Chemistry: X suggests that a membrane-based technique called ultrafiltration could solve this waste problem while simultaneously reshaping the very chemistry of flavor inside the cheese itself. By combining untargeted metabolomics, volatile compound profiling, and 16S rRNA sequencing, researchers led by Xin Cai and Zhenmin Liu have mapped, in unprecedented detail, how concentrating milk before cheesemaking changes what microbes do and what the final cheese tastes like.

The logic behind ultrafiltration is straightforward. Instead of forming a curd and then expelling whey, the technique passes pasteurized skim milk through a pressure-driven membrane with a nominal molecular weight cut-off of 10 kilodaltons. Water, lactose, and small solutes cross the membrane as permeate, while proteins and fats are retained and concentrated. In this study, skim milk was concentrated five-fold to a retentate containing 17.5 percent total solids and 15.5 percent protein, all while being held below 10 degrees Celsius and processed at a transmembrane pressure of 1.5 bar. Because whey proteins are too large to pass through the membrane, they are trapped alongside caseins and can be incorporated directly into the cheese matrix, boosting yield and nutrient density while eliminating the whey stream that burdens treatment plants.

But concentrating milk is not chemically neutral. The ultrafiltration process raises the buffering capacity of the milk, lowers its water activity, and strips out much of the lactose, creating a substrate environment that can impose osmotic stress and metabolic constraints on the lactic acid bacteria responsible for fermentation. Previous work on ultrafiltered cheese focused mostly on texture, yield, and proteolysis. What remained poorly understood was how this altered matrix reshapes the simultaneous interplay between bacterial communities, non-volatile metabolites, and volatile aroma compounds. The new research set out to answer exactly that question using a controlled 2 by 2 factorial design.

The team manufactured four cheese types: a traditional control cheese and an ultrafiltered cheese, each made with two different commercial starter cultures, one combining Lactococcus lactis subspecies with Streptococcus salivarius subsp. thermophilus and the other based on Lactococcus lactis alone. Both cheese types were formulated to comparable protein levels of 13 to 15 percent and ripened under matched conditions of heat treatment, salting at 0.8 percent, and ripening at 15 degrees Celsius for seven days. Fermentation was stopped once the pH reached 4.6 to 4.8. This design allowed the researchers to separate the effect of the cheese matrix from the effect of the starter culture, a distinction that proved crucial to interpreting the results.

The free amino acid data delivered some of the most striking findings. Twenty-one amino acids were quantified across the groups, with total contents ranging from 433 to 1005 milligrams per kilogram. The ultrafiltered cheeses accumulated significantly more free amino acids than their conventional counterparts, and the composition shifted toward positive taste attributes. Umami amino acids were enriched 3.5- to 4.9-fold and sweet amino acids 1.0- to 2.5-fold in the ultrafiltered matrix. Glutamate, the signature umami compound, soared to 167 milligrams per kilogram in the ultrafiltered cheese made with the Lactococcus-based starter, a 10.2-fold increase over its control. Two-way analysis of variance confirmed a significant matrix effect for every quantified amino acid, indicating that the ultrafiltered environment broadly reprograms amino acid accumulation, with the starter culture modulating the magnitude of the response.

Untargeted metabolomics expanded the picture dramatically. Using high-resolution Orbitrap mass spectrometry coupled to both hydrophilic interaction and reversed-phase chromatography, the researchers annotated 3952 metabolites, dominated by lipids and amino acid derivatives. Principal component analysis cleanly separated control and ultrafiltered cheeses along the first axis, and rigorous screening with fold-change, variable importance, and false-discovery-rate criteria identified roughly 800 differential metabolites in each matrix comparison. Carbohydrate abundance plummeted by around 63 percent in the ultrafiltered cheeses, a direct chemical echo of lactose removal during ultrafiltration, while glycerophospholipids such as phosphatidylethanolamine, phosphatidylserine, and phosphatidylinositol accumulated, likely reflecting retention of endogenous milk fat components. Pathway enrichment pointed to galactose metabolism, starch and sucrose metabolism, and branched-chain amino acid biosynthesis as the biochemical routes most affected by the matrix change.

The volatile fraction told a complementary story. Headspace solid-phase microextraction followed by gas chromatography-mass spectrometry detected 100 volatile compounds, and the ultrafiltered cheeses contained markedly fewer acids, with concentrations dropping by 59 to 69 percent, but higher levels of alcohols, ketones, lactones, and aromatic hydrocarbons. Odor activity values, which compare each compound’s concentration to its odor threshold, revealed that methyl ketones such as 2-heptanone and 2-nonanone, creamy lactones such as delta-octalactone, and fruity alcohols such as benzyl alcohol were significantly elevated in the ultrafiltered matrix. In total, 11 key odor-active compounds distinguished the ultrafiltered cheeses, skewing their aroma profile toward fruity and creamy notes rather than the sharp, sour character contributed by free fatty acids in conventional cheese.

Microbial sequencing added the ecological dimension. Firmicutes dominated all samples at more than 99 percent abundance, with Lactococcus and Streptococcus as the principal genera. Intriguingly, the starter culture, not the matrix, appeared to be the primary driver of community composition: cheeses sharing a starter clustered together regardless of whether they were ultrafiltered. Diversity indices, however, responded strongly to the matrix, rising sharply in one ultrafiltered group and collapsing in another. Spearman correlation analysis then linked the pieces together. Lactococcus abundance was positively associated with methyl ketones, delta-octalactone, and most taste-active amino acids, consistent with its proteolytic activity, whereas Streptococcus correlated with specific alcohols such as 1-octen-3-ol and 3-methyl-1-butanol and showed the opposite lipid-metabolite pattern.

The authors are careful to frame these associations as hypothesis-generating rather than proof of causation, and they acknowledge real limitations. Seven days is an early ripening stage, the aroma results were not validated by sensory panels or gas chromatography-olfactometry, and odor activity values were computed using aqueous thresholds that may not reflect release in a fatty cheese matrix. Because only a single ripening time point was examined, whether these matrix-driven differences persist, fade, or intensify during extended maturation remains an open question. Still, the study provides the most integrated view to date of how a membrane-imposed matrix sculpts the metabolic landscape of cheese.

The implications reach well beyond the laboratory. Whey discharge is a genuine environmental and economic burden for the dairy industry, and a whey-free process that retains high-value proteins while yielding cheese with enhanced umami and creamy-fruity character could transform sustainable cheesemaking. If future longitudinal and sensory studies confirm these findings, ultrafiltration may evolve from a yield-boosting trick into a precision tool: by tuning the milk matrix, cheesemakers could steer microbial metabolism toward desired flavor destinations before a single starter cell is added. In an era when food science increasingly treats fermentation as an engineered ecosystem, this work shows that the environment itself, not just the microbes within it, is a powerful lever for designing what we taste.

Subject of Research: Effects of ultrafiltration on microbial communities, metabolism, and flavor profiles in natural cheese

Article Title: Ultrafiltration matrix modulates metabolism and flavor profiles in natural cheese: Insights from microbial and metabolomic profiling

Article References: Cai, X., Zheng, Y., Su, M., Ma, S., Zhang, X., & Liu, Z. (2026). Ultrafiltration matrix modulates metabolism and flavor profiles in natural cheese: Insights from microbial and metabolomic profiling. Food Chemistry: X, Article 104609. https://doi.org/10.1016/j.fochx.2026.104609

Image Credits: AI Generated

DOI: 10.1016/j.fochx.2026.104609

Keywords: ultrafiltration, cheese, metabolomics, flavor chemistry, lactic acid bacteria, 16S rRNA sequencing, whey, volatile compounds, free amino acids, dairy sustainability, Lactococcus, Streptococcus