Single cell profiling at the maternal–fetal interface reveals a deficiency of PD-L1+ non-immune cells in human spontaneous preterm labor

The mechanisms that underlie the timing of labor in humans are largely unknown. In most pregnancies, labor is initiated at term (≥ 37 weeks gestation), but in a signifiicant number of women spontaneous labor occurs preterm and is associated with increased perinatal mortality and morbidity. The objective of this study was to characterize the cells at the maternal–fetal interface (MFI) in term and preterm pregnancies in both the laboring and non-laboring state in Black women, who have among the highest preterm birth rates in the U.S. Using mass cytometry to obtain high-dimensional single-cell resolution, we identified 31 cell populations at the MFI, including 25 immune cell types and six non-immune cell types. Among the immune cells, maternal PD1+ CD8 T cell subsets were less abundant in term laboring compared to term non-laboring women. Among the non-immune cells, PD-L1+ maternal (stromal) and fetal (extravillous trophoblast) cells were less abundant in preterm laboring compared to term laboring women. Consistent with these observations, the expression of CD274, the gene encoding PD-L1, was significantly depressed and less responsive to fetal signaling molecules in cultured mesenchymal stromal cells from the decidua of preterm compared to term women. Overall, these results suggest that the PD1/PD-L1 pathway at the MFI may perturb the delicate balance between immune tolerance and rejection and contribute to the onset of spontaneous preterm labor.

. CyTOF panel 15 Table S2. Cell surface markers on the second filter 16 Table S3. Immune cell z-scores and P-values for pairwise comparisons 17 Table S4. Non-immune cell z-scores and P-values for pairwise comparisons 18 Table S5. Distributional statistics for immune and non-immune cells 19 Table S6. T cell and NK cell abundances by labor type and term/preterm 20 Table S6. T cell subset markers  21  Table S7. Macrophage/monocyte subset markers 21 Table S7. Software and algorithms 23 Figure S1. H-SNE maps of cell surface markers. 25 Figure S2. Percentages of CD45+ and CD45-cells 26 Figure S3. Gating strategy for two subsets of granulocytes 27 Figure S4. Correlations between gestational age and major immune cell populations 28 Figure S5. Pairwise comparisons macrophage/monocyte cell clusters 30 Figure S6. Macrophage/monocyte abundances by infant sex 31 Figure S7. Correlations between gestational age and macrophage/monocyte clusters 33 Figure S8. Correlations between gestational age and T cell populations 34 Figure S9. T cell abundances by infant sex 35 Figure S10. Non-immune cell abundances by infant sex 36 Figure S11. Correlations between gestational age and non-immune cell clusters 37 Figure S12. PD-L1 + non-immune cells by infant sex 38 Figure S13. PD-L1 + non-immune cells by infant sex and gestational age in laboring and non-laboring pregnancies 40 References 41

Supplementary Methods
Cell and Bead Staining. After washing by Maxpar Cell Staining Buffer twice, cells were incubated with Human TruStain FcX TM (BioLegend, Cat#422302) for 10 mins at room temperature.
To avoid the competition of anti-CD3 and anti-γδ TCR staining, 1 ul γδ TCR antibody/sample was added and incubated with 3x10 6 cells for 5 mins at room temperature. Then cells were stained with 100 ul of the antibody mix cocktail for 30 mins at room temperature and washed twice by Maxpar Cell Staining Buffer. Cells were fixed by 1.6% PFA/PBS (Thermo Fisher Scientific, Cat#28906) for 10 mins at room temperature and washed in Maxpar Cell Staining Buffer. Cell pellet was resuspended in 1 ml of Cell-ID TM Intercalator-Ir ( Sample Quality Control. All samples included in these analyses went through a quality control check based on Fluidigm's QC guidance. In brief, the quality of Cell-ID Ir-Intercalator staining (DNA1 signal) was checked for each run. We required at least 50% of all events to be DNA1 + , to have enough beads (>1% of total events) to perform normalizations of the data, and the medium intensity of Eu153 on EQ4 beads to be greater than or equal to 1000. Samples that failed on any of these were discarded.
Testing of Signal Spillover. Although CyTOF technology does not produce fluorescence spillover between channels, metal signal spillover can occur due to impurities of antibodies, oxidation of metal isotopes, or machine sensitivity. To avoid false discoveries caused by spillover, we applied CATALYST (Table S7) and an interactive Shiny-based web application to test for signal spillover across all the channels for our panel. Briefly, polystyrene capture beads were single stained with each antibody used in the experiment. Beads were then pooled and analyzed simultaneously in the mass cytometer. Single stained controls for each dye were analyzed to determine the percentage of interfering signal in all the channels. These values are reported into a spillover matrix (see figure below).

Testing for Batch Effects
Due to the variability of 'stickiness' between these cells, it was not feasible to pool multiple samples together and use the fluidigm barcoding/pooling system. Therefore, to assess batch effects and demonstrate consistency between runs of individual samples, we (1) we compared 4 aliquots of the same decidual cell sample run at 4 different times on the same day; (2) compared runs performed on 2 different days using 2 different aliquots (vials) of the same decidual cell sample. For these studies, one decidual sample from a term laboring pregnancy was split into 4 individual tubes and stained with the 32-antibody panel. Cells from each of the 4 tubes were sequentially run on the Helios machine over the course of one day. Analysis of CD45 + and CD45cell populations showed that all four samples performed similarly with respect to both cell frequency and signal intensity (see figure below, upper panel A). We also looked at the percentage of rare populations, such as γδ T cells and observed little variation between samples (see figure below, lower panel A). All four samples showed similar cellular distributions and clustering (see figure below, panel B). This experiment was repeated using a second sample and run on a different day, with similar results. The dotted line shows a function of the applied separation cutoff corresponding to the yield upon debarcoding generated by CATALYST. (C) A signal spillover matrix was generated by staining of control antibody-capture beads, which was performed with one decidua sample in parallel. The matrix displayed the spillover in potential affected channels including M±1 and M±16. The numbers in the cells on the diagonales were 1, while others were the percentages of spillover by channels in X-axis into channels in Y-axis.
Next, aliquots of cells from one term non-laboring sample were run on 2 different days. Again, the number of live cells and the percentage of CD45 + cells between aliquots and runs were similar. Both runs yielded a similar cellular distribution and clustering (see figure below).
Based on the results of these studies, we ran 4 samples a day for our study, one sample from each of the 4 biological groups (preterm and term laboring; preterm and term non-laboring).
The order of the categories in each run were alternated to control for any remaining batch effects due to the order in which they are loaded into the machine.

Test of batch effects between 4 samples of the same decidua run in one single day. (A)
Comparison of the signal intensity and the percentage of live cells between runs of the same sample. Scatterplots on the first row represented the total live cells in one vial of decidual sample, which was evenly split into 4 tubes for antibody staining and CyTOF ran on the same day. Scatterplots on the second row shows the percentage of γδ T cells in population of CD45+ live decidua cells (B) The overlapped (colored) and individual (red) phenograph data visualization by cytofkit (R package), showing whole population structure among these 4 time points. Cellular distribution and clustering were defined by tSNE1 and tSNE2. 5000 viable single cells from each time point were subjected to Phe-noGraph (B) in cytofkit.

Test of batch effects between 2 aliquots of the same sample run on different days (2 months apart).
Signal intensity and percentage of total live cells (A) and CD45+ live cells (B) from one term non-laboring pregnancy. Scatterplots show the number of total live cells in each sample.
(C) Rphenograph display of the distribution of the samples on each day by cytofkit (R package), defined by tSNE1 and tSNE2. 3000 viable cells from each day were subjected to PhenoGraph (C) in cytofkit.
software to perform data analysis and data visualization (see Table S7). Cytosplore (1): For Hierarchical Stochastic Neighbor Embedding (HSNE) analysis, live CD45 + /CD45cells from each sample were gated, imported and analyzed on the Cytosplore platform using negative value pruned inverse hyperbolic sine transformation (cytofAsinh) transformation and HSNE visualization. The HSNE graphs and the marker expression heatmaps were exported for making figures. The matrix, including the frequency of each cell population, was exported as an excel file for statistical analysis.
Cytofkit (3): After installation of Cytofkit package and library, the command: "cytofkit_GUI()" was used to launch the GUI. The extracted .fcs files were imported and transformed using cyto-fAsinh in cytofkit. "Rphenograph" was chosen as the major cluster method, while "PCA" and "tSNE" were chosen as the visualization methods. After analysis, .RData files were saved and then loaded to Shiny APP for further exploring the data and results in an interactive manner. Launching the Shiny APP and loading .RDatwere followed usage of Shiny APP (Table S7).

Gestational Age, Sex and Labor Effects on Cell Distributions
Immune Cell Lineages. In the combined sample of term and preterm pregnancies, the abundances of the major immune cell subsets did not differ significantly by gestational age ( The abundance of granulocytes was higher in TL compared to TNL (z= 2.38, P = 0.02).
None of the macrophage/monocyte cell clusters differed between pregnancies with male or female infants ( Supplementary Fig. 6) or by gestational age (Supplementary Fig. 7).
T Cell Subsets. T cells in clusters 1 and 3, 9, 10 were overall more abundant in pregnancies with male infants. Because TL had fewer male pregnancies (25% male) compared to TNL (46% male), it is likely that the differences between TL and TNL for cells in clusters 1, 3, 9 and 10 reflected sex ratio imbalances and not true differences between laboring and non-laboring pregnancies at term. Clusters 2 and 7, defined as Th17 CD4 T cells and HLA-DR + PD-1 + CD8 T cells, were less abundant in TL compared to TNL (P = 0.0098 and P = 7.43x10 -9 , respectively; Fig. 5) and did not differ by either gestational age or infant sex in the combined sample. The same cells in cluster 7 were also higher in TNL compared to PNL (P = 0.0012), and in PL compared to TL, but these differences did not reach significance (P = 0.069) ( PD-L1 + Non-Immune Cells. The abundances of PD-L1 + non-immune cells increased with increasing gestational age among laboring pregnancies (r=0.037, P = 0.057 ( Supplementary Fig. 13A), but did not change in non-laboring pregnancies (r = 0.13, P = 0.53 ( Supplementary Fig. 13B). The latter observation suggests that the abundances of PD-L1 + non-immune cells at the MFI are not influenced by gestational age per se but rather may be a feature of labor at term. Moreover,  Fig. 13C-D). Thus, fetal sex was not likely contributing to the differences in abundances of PD-LI + non-immune cells between TL and PL (Fig. 6C).       Table S7. Phenotype keys for 10 T-cell clusters.