General composition of the breastmilk T cell compartment throughout lactation. Live T cells were present in human breastmilk samples in a comparable proportion across all time points postpartum (Figure 1, A and B, and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.181788DS1). Across all of our time points, breastmilk contained, on average, more CD4+ than CD8+ conventional T cells (TCRγδ– and TCRVα7.2–), comparable with the ratio in PBMC (Figure 1C). Mucosal-associated invariant T cells (MAIT) (TCRVα7.2+CD161+) were numerous in a few breastmilk samples but not significantly different from PBMC (Supplemental Figure 1B). TCRγδ+ T cells were less frequent in breastmilk (Supplemental Figure 1C). The percentage of double-negative (DN) (CD4–CD8–) T cells was highly variable (range, 1%–99%; mean, 42%; Supplemental Figure 1D), and samples with many DN T cells (>50% of viable CD3+) were not enriched for either MAIT or TCRγδ+ T cells.

General breastmilk T cell composition. (A and B) Representative flow cytometry plots of the CD4+ and CD8+ naive/memory composition based on CD45RA and CD27 staining in breastmilk- and PBMC-derived T cells. (C) Ratio of CD4+ to CD8+ T cells, excluding TCRγδ+ T cells and MAIT cells (PBMC n = 7, breastmilk [BM] 1 week n = 7, BM 1 month n = 19, BM 3 months n = 20, BM 6 months n = 4). (D and E) Distribution of naive (CD45RA+CD27+), memory (CD45RA–, split up in CD27– and CD27+) and T-EMRA (CD45RA+CD27–) T cell subsets expressed as the percentage of CD4+ (D) and CD8+ (E) T cells, with pairwise comparisons between BM time points and PBMC (PBMC n = 8, BM 1 month n = 20, BM 3 months n = 16, BM 6 months n = 12). Data represent mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 using the Kruskal-Wallis test followed by Dunn’s test for multiple comparisons. For D and E, significance only denotes differences between BM timepoints compared with control PBMC. Transparent lines connect data points of different time points postpartum from the same breastmilk donor. MAIT, Musocal-Associated Invariant T cells (TCRVα7.2+CD161+); TCR, T cell receptor. FACS data of T cells in breastmilk of 1 week and 1, 3, and 6 months postpartum compared with PBMC of age-matched female control donors.

The majority of CD4+ and CD8+ breastmilk T cells had low expression of CD45RA, indicating an enriched memory population (Figure 1, D and E). For CD4+ breastmilk T cells, we observed only a significant increase in CD45RA–CD27– memory T cells, whereas in CD8+ T cells, both the CD45RA–CD27– and CD45RA–CD27+ population were significantly increased compared with PBMC (Figures 1, D and E and Supplemental Figure 2, A, B, E, and F). Although not assessed in combination with CD45RA, lower expression of CCR7 in CD4+ and CD8+ T cells further indicated a prominent effector memory T cell population in breastmilk (Supplemental Figure 2J). In contrast to CD4+, the CD8+ population also included CD45RA+CD27– terminally differentiated effector memory cells reexpressing CD45RA (T-EMRA), comparable with PBMC, and maintained a small population of CD45RA+CD27+ naive T cells (Figure 1, D and E, and Supplemental Figure 2, C, D, G, and H). Although in total reduced compared with PBMC (Figure 1E), a larger fraction of these naive CD8+ T cells were CD31– (Supplemental Figure 2I), possibly indicating increased differentiation (35) or contamination with T-EMRA.

scRNA-Seq of breastmilk T cells. To get a comprehensive and unbiased overview of the breastmilk T cell population, we performed scRNA-Seq on live CD3+ T cells from 1 month postpartum breastmilk (Supplemental Figure 3, D–F). We identified 6 distinct populations (Figure 2A), with CD8A and CD8B expression in clusters 2 and 6 and CD4 expression in clusters 3 and 4, of which cluster 4 coexpressed FOXP3, indicating Tregs (Figure 2, B and C). Cluster 1 was very distinct, with high log fold changes in the differentially expressed genes (DEG; Figure 2C), including many long noncoding RNAs, possibly indicating either contamination or low-quality cells (Supplemental Figure 3, D–F). More than half of the cells in clusters 5 and 1 expressed neither CD4 nor CD8A/B. Using the index sort surface expression data (Supplemental Figure 3, A–C), we confirmed that the highest fraction of DN cells was present in cluster 5 (24%).

Breastmilk T cell populations identified by single-cell RNA-seq. T cell populations present in breastmilk at 1 month postpartum (n = 7). (A) UMAP showing cluster annotation and cluster proportions per breastmilk donor. (B) UMAP gene expression (log-normalized) projections for CD8A, CD4, and FOXP3. (C) Heatmap showing expression (log-normalized) of the top 10 differentially expressed genes (DEG, MAST test) per cluster.

Breastmilk T cells demonstrate an activated/effector profile with cytotoxicity and high cytokine production capacity. Cluster 2 contained CD8+ T cells with differential expression of the effector chemokine CCL5 and cytotoxicity-related molecules (Figure 3a), including granzymes (GZMK/A/B/H), granulysin (GNLY), perforin (PRF1), and NK cell granule protein 7 (NKG7), as well as the antiviral enzyme APOBEC3G (Figure 2C). We confirmed with FACS that breastmilk CD8+ memory T cells across all time points produced granzyme B and K, similar to PBMC (Figure 3C and Supplemental Figure 4A). The effector chemokine CCL5 was characteristic of breastmilk, with over 5 times more CCL5-producing memory CD8+ T cells compared with PBMC (Figure 3B). In addition, a strong effector function was supported by increased coproduction of granzyme B and K, although only significantly different from PBMC at 6 months postpartum (Supplemental Figure 4B). This effector profile might partially be controlled by EOMES, an effector memory transcription factor that was differentially expressed in the CD8+ cluster 2 (Figure 3A). With FACS, we showed significantly higher protein-level expression of EOMES in breastmilk CD8+ memory T cells compared with PBMC, increasing with time postpartum (Figure 3D).

Functional profile of breastmilk cytotoxic T cells. (A) UMAP showing clustering and gene expression (log-normalized) projections for the cluster 2 characteristic genes CCL5, granzyme B, and EOMES in 1 month postpartum breastmilk (single-cell RNA-Seq, n = 7). (B–D) Frequency of CCL5+ (B), granzyme B+ (C), and EOMES+ (D) cells as the percentage of CD8+CD45RA– T cells in breastmilk of 1 week and 1, 3, and 6 months postpartum compared with PBMC of age-matched female control donors (FACS) (PBMC n = 8, BM 1 week n = 1, BM 1 month n = 11, BM 3 months n = 16, BM 6 months n = 6). *P < 0.05,**P < 0.01, with pairwise comparisons among PBMC and BM across different time points using the Kruskal-Wallis test followed by Dunn’s test for multiple comparisons. Data represent mean ± SD. Gray lines connect data points of different time points from the same breastmilk donor. BM, breastmilk.

The CD4+ T cell cluster 3 expressed genes related to cytokine/TCR signaling, including SOS1, PAG1, SOCS2/3, and CYLD (Figure 2C). An activated immune cell profile was even more prominent in the predominantly DN cluster 5, with top DEG including NFKBIA/D/Z, NR4A1/2, GADD45A/B/G, DUSP1/2, NFKB2, JUN/B/D, FOS, and IER2/3/5/5L (Figure 2C). Many of these genes are involved in immediate early response signaling pathways following immune- or stress-related activation. Cluster 6 shared many of cluster 5 DEG, including DUSP and NR4A2, but not NFKBIA (Figure 2C).

In our FACS data, both CD4+ and CD8+ breastmilk memory T cells demonstrated cytokine production upon restimulation (Supplemental Figure 4, C and E–G), with significantly increased IFN-γ production at 1 month postpartum for CD4+ breastmilk T cells compared with PBMC (Figure 4A). CD4+ memory T cells displayed a preferential type-1 immune profile with IFN-γ and TNF-α production, as well as IFN-γ–TNF-α coproduction, but with little IL-13 production (Figure 4A; Supplemental Figure 4, E–G). CD4+ memory T cells also demonstrated significantly increased proliferative capacity compared with PBMC, especially at later time points (Figure 4B). In addition, we observed a significantly increased frequency of CTLA-4+ memory CD4+ and CD8+ T cells in breastmilk, compared with PBMC (Figure 4C and Supplemental Figure 4D). The proliferation marker Ki-67 was enriched in CTLA-4+ cells (Supplemental Figure 4, H and I), suggesting that activation and regulation go hand in hand in breastmilk T cells.

Cytokine production and activated profile of CD4+ breastmilk T cells. (A) Frequency of T cells producing IFN-γ upon PMA-ionomycin stimulation, expressed as the percentage of CD4+CD45RA– live T cells (PBMC n = 8, BM 1 month n = 7, BM 3 months n = 3, BM 6 months n = 1). (B and C) Frequency of Ki67+ (B) and CTLA-4+ cells (C) as the percentage of CD4+FOXP3–CD45RA– T cells (PBMC n = 7, BM 1 week n = 5, BM 1 month n = 19, BM 3 months n = 19, BM 6 months n = 4). *P < 0.05, **P < 0.01, ***P < 0.001 with pairwise comparisons among PBMC and BM using the Kruskal-Wallis test followed by Dunn’s test for multiple comparisons. Data represent mean ± SD. Gray lines connect data points of different time points from the same breastmilk donor. BM, breastmilk.

Breastmilk is highly enriched for Tregs. Part of the CD4+ T cells in our scRNA-Seq data were Tregs (cluster 4), with differential expression of the core Treg markers FOXP3, CD4, IL2RA (CD25), and CTLA4 (Figure 2C and Figure 5A). Other Treg-related DEG included TNFRSF1B (TNFR2), CD52, IL32, FCMR, and IL10RA. FACS confirmed that breastmilk is highly enriched for FOXP3+CD25+ Tregs, with a 3-fold increased frequency compared with PBMC across all time points (Figure 5B). Moreover, expression of the regulatory marker CTLA-4 was significantly increased in breastmilk Tregs (Figure 5C). Whereas differential expression of SELL (CD62L) in cluster 4 might indicate lymphoid homing/naivety, we observed low surface expression of CCR7 and CD45RA on breastmilk Tregs (Figure 5D and Supplemental Figure 2K). SELL might here instead indicate a more suppressive Treg population (36).

Treg profile in breastmilk. (A) UMAP showing gene expression (log-normalized) projections for the cluster 4 characteristic genes FOXP3, CTLA4, and SELL in 1 month postpartum breastmilk (single-cell RNA-Seq). (B) Frequency of FOXP3+CD25+ Tregs expressed as the percentage of CD4+ T cells (FACS) (PBMC n = 11, BM 1 week n = 2, BM 1 month n = 28, BM 3 months n = 18, BM 6 months n = 5). (C and D) Frequency of CTLA-4+ cells as the percentage of CD4+FOXP3+CD45RA– (PBMC n = 7, BM 1 week n = 2, BM 1 month n = 13, BM 3 months n = 7, BM 6 months n = 2) and CCR7+ cells as the percentage of FOXP3+CD25+ Tregs (PBMC n = 7, BM 1 week n = 1, BM 1 month n = 23, BM 3 months n = 4, BM 6 months n = 3) (FACS). FACS data include cells from breastmilk of 1 week and 1, 3, and 6 months postpartum compared with PBMC of age-matched female control donors. *P < 0.05, **P < 0.01, ***P < 0.001 with pairwise comparisons among PBMC and BM across different time points using the Kruskal-Wallis test followed by Dunn’s test for multiple comparisons. Data represent mean ± SD. Gray lines connect data points of different time points from the same breastmilk donor. BM, breastmilk.

Breastmilk T cells show signs of recent tissue residence. Our scRNA-Seq data reveal signs of tissue adaptation in breastmilk T cells, with high expression of the tissue-resident memory (TRM) markers CD69 and RUNX3 throughout all clusters, but especially in clusters 5 and 6 (Figure 6C). In addition, the TRM-related marker CXCR6 was differentially expressed in the CD8+ cluster 2 (Figure 6C). FACS confirmed that high surface expression of TRM markers (CD69/CD103) distinguished breastmilk from circulating CD4+ and CD8+ memory T cells, on which these markers are mostly absent (Figures 6, A and B). For CD8+, almost half of the CD69+ cells coexpressed CD103, with a small population of CD103+CD69– cells (Figure 6B). Surface expression of CXCR6 was enriched in the CD69+CD103+ TRM population and was significantly increased compared with PBMC (Figure 6D). We found that granzyme K production was significantly lower in the CD69+CD103+ TRM compared with the CD69–CD103– population (Supplemental Figure 6C), whereas there was a trend toward higher CCL5 in the TRM population (Supplemental Figure 6D). CD69, but not CD103, expression was also increased in the breastmilk DN and Treg populations, comparable with CD4+ memory T cells (Supplemental Figure 6, A and B). The DEG of the CD4+ cluster 3 included a set of genes (CCR6, MAF, RGS1/2, and CREM) (Supplemental Figure 5A), which have been associated before with skin and intestinal CD4+ T cells (37). Gene set enrichment testing of a core human TRM signature (38) showed overlap with the afore-mentioned tissue-related genes in CD4-expressing cluster 3 as well as with CD8A/B-expressing clusters 2 and 6 (Supplemental Figure 5B). Together, these results indicate a tissue origin of both CD4+ and CD8+ breastmilk memory T cells, possibly originating from the breast tissue itself.

Tissue-residency and tissue-homing profiles in breastmilk T cells. (A and B) Distribution of tissue-resident memory T cell subsets based on protein expression of the TRM markers CD69 and CD103 (FACS) expressed as the percentage of CD4+CD45RA– (PBMC n = 8, BM 1 week n = 2, BM 1 month n = 11, BM 3 months n = 17, BM 6 months n = 12) (A) and CD8+CD45RA– T cells (PBMC n = 8, BM 1 week n = 2, BM 1 month n = 11, BM 3 months n = 16, BM 6 months n = 6) (B) with pairwise comparisons between BM time points and PBMC. (C) UMAP showing clustering and gene expression (log-normalized) projections for the TRM-related genes CD69, RUNX3, and CXCR6 and the homing receptors integrin α4 (ITGA4), integrin β1 (ITGB1), and integrin β7 (ITGB7) in 1 month postpartum breastmilk (single-cell RNA-Seq, n = 7). (D) Frequency of CXCR6+ cells (FACS) as the percentage of CD8+CD45RA– T cells compared between the 4 different CD8+ TRM subsets within breastmilk (colored dots) and with the total population in control PBMC (gray dots) (PBMC n = 8, BM n = 18). (E–G) Surface expression of different homing receptors (FACS) with pairwise comparisons among PBMC and BM across different time points. Frequencies of α4β1+ cells as the percentage of CD4+FOXP3–CD45RA– T cells (PBMC n = 7, BM 1 week n = 5, BM 1 month n = 19, BM 3 months n = 19, BM 6 months n = 4) (E), α4β7+ cells as the percentage of CD8+CD45RA– T cells (F), and cells double-positive for α4β7 and CLA as the percentage of CD8+CD45RA– T cells (G). FACS data include cells from breastmilk of 1 week and 1, 3, and 6 months postpartum compared with PBMC of age-matched female control donors (PBMC n = 7, BM 1 week n = 6, BM 1 month n = 17, BM 3 months n = 15, BM 6 months n = 3). (A, B, and E–G) Pairwise comparisons among PBMC and BM time points were tested using the Kruskal-Wallis test followed by Dunn’s test for multiple comparisons. (D) For comparisons between breastmilk T cell subsets, the Friedman test was used followed by Bonferroni-corrected pairwise Wilcoxon ranked-sum post hoc testing. *P < 0.05, **P < 0.01, ***P < 0.001. Data represent mean ± SD. Gray lines connect data points of different time points from the same breastmilk donor. TRM, tissue-resident memory T cell; BM, breastmilk.

We further explored the potential breast tissue origin of breastmilk T cells by comparing our data with 2 public scRNA-Seq datasets of (immune) cells from healthy breast tissue biopsies (39, 40). After integration, we found that breastmilk T cells clustered mostly together with breast tissue T cells (Figure 7, A, B and D, and Supplemental Figure 7, A–C), with separation into CD4-expressing and CD8-expressing clusters (Figure 7C). To compare breastmilk and breast tissue T cells, we then performed pseudobulk differential gene expression, yielding 3,907 genes upregulated in milk versus tissue and 1,347 in tissue versus milk (logFC > |1| and adjusted P [Padj] < 0.05; Supplemental Table 3). Even after subsetting for the clusters enriched for T cells, we found that the comparison of breast tissue versus milk was confounded by the presence of many non–T cell–related genes. However, we found some indications that CD8-expressing cells were more similar across milk and tissue than CD4-expressing T cells. Especially many of the Treg-related genes from the above described breastmilk cluster 4 (FOXP3, CTLA4) appeared to be mostly absent in T cells from breast tissue (Figure 7E). Furthermore, we found significantly higher expression of genes related to homing (ITGA4, ITGB1, CCR6) and cytokine/TCR signaling (e.g., SOS1) in T cells from breastmilk compared with breast tissue (Figure 7E and Supplemental Figure 8). Together, these data suggest difference in regulation between breast tissue versus milk residing CD4+ T cells.

Partial overlap of breastmilk with breast tissue T cells. Clustering of breastmilk T cells with the immune cell subsets from healthy breast tissue extracted from 2 publicly available single-cell RNA-Seq datasets, with breastmilk T cells in orange (n = 7), breast tissue cells from Pal et al. (39) in blue (n = 12) and breast tissue CD45+ immune cells from Azizi et al. (40) in green (n = 10). (A) UMAP showing cluster annotation. (B) UMAP showing the original cluster annotation of the breastmilk T cells (colors) projected onto the combined clustering space of breastmilk and breast tissue (black). (C) UMAP gene expression (log-normalized) projections for CD3D, CD4, and CD8A. (D) Frequency of each of the 3 datasets for the main cell types identified in A and C. (E) Violin plots comparing gene expression (log-normalized) of CD4, CD8A, FOXP3, CTLA-4, ITGB1 (integrin β1), and CCL5 between the breastmilk and breast tissue datasets for each of the major T cell subsets.

In contrast, most of the genes related to cytotoxic T cells were similarly expressed between breastmilk and breast tissue T cells, including CD8A, GZMB, and CCL5 (Figure 7E and Supplemental Figure 8). The only differences that stood out in CD8-expressing T cells from breastmilk versus tissue were higher expression of GZMK, EOMES, CCR5, and CXCR6 (Supplemental Figure 8). We found that part of the breastmilk T cells ended up in non–T cell clusters that expressed high levels of stress-related genes such as heat-shock proteins. In general, many of the genes related to activation and/or stress, including NFKBIA and JUN, were enriched in breast tissue and abundantly expressed across breast tissue T cell and non–T cell clusters, potentially indicating a relation with tissue digestion. In summary, we found partial overlap of breastmilk with breast tissue T cells, indicating a potential tissue-origin of breastmilk T cells, especially for cytotoxic CD8+ T cells. However, differences in the expression of genes related to homing and regulation might indicate that these characteristics are more specific for breastmilk T cells.

Tissue-homing capacities of breastmilk T cells. Expression of homing receptors would be a prerequisite for a tissue role of breastmilk T cells in either the breast tissue environment or in neonatal immune defense, as indicated by animal studies. Our FACS data show expression of skin-homing (cutaneous leucocyte-associated antigen [CLA]), intestinal-homing (integrin α4β7, CCR9) and general inflammatory homing (integrin α4β1) receptors in breastmilk T cells from all time points (Figure 6, E–G, and Supplemental Figure 6, E–G). Compared with PBMC, breastmilk contained significantly more CLA-, α4β7-, and CCR9-expressing CD8+ memory T cells and more CLA- and α4β1-expressing CD4+ memory and Tregs (Figure 6, E and F, and Supplemental Figure 6, E–G). Breastmilk also contained significantly more T cells with a broad homing profile, such as α4β7+CLA+ and α4β1+CLA+, which are rare in PBMC (Figure 6G). FACS data partially confirm scRNA-Seq data, where ITGA4 and ITGB1 were enriched in both CD4+ clusters, but ITGB7 only in the Treg cluster (Figure 6C). Both proliferation (Ki-67) and regulation (CTLA-4) were significantly increased in T cells positive for homing receptors, especially for CLA+, but not for CCR9+ cells (Supplemental Figures 9 and 10). Together, these data suggest that breastmilk contains T cells with tissue migration potential and that part of these cells appear to be more activated.

Variation in part of the breastmilk T cell profile correlates with clinical parameters. Throughout our data, we observed substantial donor variation in the breastmilk T cell profile. Using stepwise multiple regression, we aimed to discriminate the independent effects of a selection of clinical parameters on the breastmilk T cell profile, with a special interest in respiratory infections of the infant (medically attended acute respiratory infections [MARI]) (Figure 8). We found that allergy of the mother and the presence of 1 or more siblings had most influence on the breastmilk T cell profile (Figure 8A), with a significant positive relationship with expression of most homing receptors (CLA, α4β7, and α4β1), activation markers, and cytokine production but a significant negative relationship to CCR9 expression and a cytotoxic T cell profile (Figure 8B). Contrary to maternal allergy and siblings, the presence of pets had a significant negative correlation with homing receptor expression (Figure 8A). Additionally, we found that nursing a boy was most strongly related to a Treg profile (Figure 8, A and B). When we then considered independently the T cell profile in infants who experienced 1 or more MARI, we found a significantly lower percentage of CD8+ T cells expressing the homing receptors CLA and α4β7 (Figure 8B).In addition, MARI were associated with a significantly lower expression of the TRM-related marker CXCR6 but a higher percentage of CD69+CD103– T cells. In summary, we found the strongest influence of maternal/household factors on the breastmilk T cell profile but also identified a couple of breastmilk T cell characteristics with a potential relationship to neonatal health.

Variation in part of the breastmilk T cell profile correlates to clinical parameters. n = 217 BM samples (see Tables 1 and 2 for missing data on clinical parameters). (A) Heatmap visualizes a selection of the T cell parameters that were significantly correlated to 1 or multiple clinical parameters, selected by stepwise multiple linear regression on log-transformed T cell parameters. Color intensity indicates the level of significance for the contribution of each clinical parameter on each T cell parameter, with yellow to red indicating a positive relationship, light blue to dark blue a negative relationship, and white indicating that the clinical parameter did not display explanatory power during stepwise model selection. (B) Effect size (regression coefficient) for the relation between T cell and clinical parameters, showing a selection of the T cell parameters from A that displayed a highly significant correlation to 1 or multiple clinical parameters. Effect size is only shown for those clinical parameters that displayed explanatory power during stepwise model selection. Color intensity indicates the level of significance. All clinical parameters were converted to a binary outcome. MARI, medically attended respiratory infections.