Mass cytometry reveals systemic and local immune signatures that distinguish inflammatory bowel diseases

Abstract

Inflammatory bowel disease (IBD) includes Crohn's disease and ulcerative colitis. Each disease is characterized by a diverse set of potential manifestations, which determine patients' disease phenotype. Current understanding of phenotype determinants is limited, despite increasing prevalence and healthcare costs. Diagnosis and monitoring of disease requires invasive procedures, such as endoscopy and tissue biopsy. Here we report signatures of heterogeneity between disease diagnoses and phenotypes. Using mass cytometry, we analyze leukocyte subsets, characterize their function(s), and examine gut-homing molecule expression in blood and intestinal tissue from healthy and/or IBD subjects. Some signatures persist in IBD despite remission, and many signatures are highly represented by leukocytes that express gut trafficking molecules. Moreover, distinct systemic and local immune signatures suggest patterns of cell localization in disease. Our findings highlight the importance of gut tropic leukocytes in circulation and reveal that blood-based immune signatures differentiate clinically relevant subsets of IBD.

Fig. 1

Circulating leukocytes distinguish CD and UC. a Schematic of the study conducted on cohort 1. Created with BioRender. b viSNE using CD11c, CD56, CD16, CD8, CD3, CD123, CD27, CD24, CD14, CD19, CD4, CD20, and CD45RO for clustering with samples from cohort 1. Coefficient of variation was calculated for 2208 parameters per sample by disease group. Statistics: unpaired two-tailed Student’s T test (CD remission vs. HC, t = 12.43, df = 4412; CD remission vs. UC remission, t = 14.12, df = 4406; UC flare vs. HC, t = 6.994, df = 4403; UC flare vs. UC remission, t = 8.621, df = 4397). Sample sizes: CD flare = 13; CD remission = 11; UC flare = 10; UC remission = 10; HC = 12. c Features distinguished all CD and UC. Statistics: BH FDR-corrected unpaired two-tailed Student’s T test using Morpheus (see the Methods section; CCR9⁺GPR15⁺CD56⁺ B cells, t = 2.58; α4β7⁺CCR1⁺CD56⁺ plasmablasts, t = 2.74). Sample sizes: CD = 23, UC = 18. d Features differentiating CD and UC identified by hypothesis-driven tests. Statistics: unpaired two-tailed Student’s T test (Basophils [% of live singlets]: all CD vs. UC, t = 2.57, df = 42; CD vs. UC flare, t = 3.34, df = 21; CD flare vs. HC, t = 2.79, df = 23; CD flare vs. remission, t = 2.87, df = 22; all UC vs. HC, t = 3.88, df = 30; UC flare vs. HC, t = 4.02, df = 20; UC flare vs. remission, t = 6.91; df = 18. Basophils [median pCREB]: all CD vs. UC, t = 2.53, df = 42; CD vs. UC flare, t = 3.17; df = 21. pDCs [% of DCs]: all CD vs. UC, t = 2.61, df = 42; CD vs. UC flare, t = 2.97, df = 21; UC flare vs. remission, t = 4.03; df = 18. α4β7⁺ mDCs [% of mDCs]: all CD vs. UC, t = 2.07, df = 39; CD vs. UC flare, t = 3.30, df = 19; CD flare vs. remission, t = 2.33, df = 21. Effector memory CD4 T cells [median pCREB]: all CD vs. UC, t = 2.27, df = 42; CD vs. UC flare, t = 3.13, df = 21; CD flare vs. remission, t = 2.92; df = 22. IgD⁻CD27⁻ B cells [% of CD19⁺CD20⁺]: all CD vs. UC, t = 2.15, df = 42; CD vs. UC flare, t = 2.77, df = 21; UC flare vs. remission, t = 3.47, df = 18; UC flare vs. HC, t = 5.05, df = 20). Sample sizes: all CD = 24; CD flare = 13; CD remission = 11; all UC = 20; UC flare = 10; UC remission = 10; HC = 12 (23, 13, 10, 18, 8, 10, and 12, respectively, for α4β7⁺ mDCs). Center lines = mean; whiskers = standard deviation. Source data are provided as a Source Data file

Fig. 2

Some circulating leukocyte signatures differentiate disease flare and remission while others persist in remission. Significant differences between CD remission (N = 11) and HC (N = 12) samples (a), CD flare (N = 13) and CD remission (N = 11) samples (b), UC remission (N = 10) and HC (N = 12) samples (c), and UC flare (N = 10) and UC remission (N = 10) samples (d) with p ≤ 0.05 after correction for multiple testing. On the right of (b, d) are Pearson correlations based on vectors that represent all significant feature values for individual patients shown in the heatmaps immediately to the left. Statistics: BH FDR-corrected unpaired two-tailed Student’s T test using Morpheus (see the Methods section; t-statistics shown in heatmap legends). Source data are provided as a Source Data file

Fig. 3

Circulating leukocytes reflect disease heterogeneity. Features significantly different between inflammatory (N = 7) and fistulizing (N = 7) CD (a), inflammatory (N = 8) and stricturing (N = 7) CD (b), fistulizing (N = 7) and stricturing (N = 7) CD (c), and the presence (N = 7) and absence (N = 16) of perianal disease in CD patients (d). Statistics: BH FDR-corrected unpaired two-tailed Student’s T test using Morpheus (see the Methods section; IgD⁻IgA⁻ CD19⁺CD20⁺ B cells [median α4β7], t = 3.69; CD14⁺ cells, t = −4.36; IgD⁻CD27⁻ B cells, t = 3.91; IgD⁻IgA⁻ CD19⁺CD20⁺ B cells [% of CD19⁺CD20⁺], t = 4.38; α4β7⁺CCR9⁺ HLA-DR^lo DCs, t = 3.87; α4β7⁺GPR15⁺ Tregs, t = 2.29). Features that distinguish ileal (N = 5) and colonic (N = 4) CD (e), left-sided (N = 6) and pan-colonic (N = 11) UC (f), and colonic CD (N = 4) and UC (N = 18, or 20 for naïve IgD⁺ B cells) (g) identified by hypothesis-driven testing. Statistics: unpaired two-tailed Student’s T test (α4β7⁺CCR9⁺ mature NK cells, t = 3.18, df = 7; α4β7⁺CCR9⁺ CD45RO⁺ NKT cells, t = 7.15, df = 7; CCR9⁺GPR15⁺ CD38⁺HLA-DR⁺ CD4 T cells, t = 2.20, df = 15; GPR15⁺ Tregs, t = 2.20; df = 20; naive IgD⁺ B cells, t = 2.18, df = 22; GPR15⁺ naive IgD⁺ B cells, t = 2.23; df = 20; CCR9⁺GPR15⁺ plasmablasts, t = 2.28; df = 20). Center lines = mean; whiskers = standard deviation. Source data are provided as a Source Data file

Fig. 4

Tissue contains immune responses distinct from blood. a Schematic of the study conducted on cohort 2. Blood was drawn and biopsies were collected from study subjects, peripheral blood mononuclear cells (PBMCs) and tissue leukocytes were isolated and cryopreserved, and samples were analyzed in batches by CyTOF. Created with BioRender. b viSNE based on 15 core lineage antigens (CD11c, CD11b, CD56, CD16, CD8, CD3, CD123, CD27, CD24, CD14, CD19, CD4, CD20, TCRγδ, and CD45RO) for samples from cohort 2. Sample sizes: CD blood = 6; UC blood = 6; CD inflamed tissue = 11; CD uninflamed tissue = 12; UC inflamed tissue = 5; UC uninflamed tissue = 13. c Significant differences between disease tissues consistent with trends observed in the blood from cohort 1. Statistics: unpaired two-tailed Student’s T test (Basophils: all CD vs. all UC tissue, t = 1.83, df = 39; UC inflamed vs. uninflamed tissue, t = 2.43, df = 16. Plasmablasts: all CD vs. all UC tissue, t = 2.51, df = 39). Sample sizes: CD blood = 6 (5 for α4β7⁺); UC blood = 6; all CD tissue = 23; CD inflamed tissue = 11; CD uninflamed tissue = 12; all UC tissue = 18; UC inflamed tissue = 5; UC uninflamed tissue = 13. Center lines = mean; whiskers = standard deviation. d T peripheral helper (Tph) cells, defined as CD3⁺CD4⁺CD45RO⁺CXCR5⁻PD-1⁺, in paired blood and tissue samples. Statistics: unpaired two-tailed Student’s T test (all blood vs. tissue, t = 2.65, df = 51; all CD vs. UC tissue, t = 2.25, df = 39; CD inflamed vs. uninflamed tissue, t = 0.75, df = 21). Sample sizes: all blood = 12; all tissue = 41; all CD tissue = 23; all UC tissue = 18; CD inflamed tissue = 11; CD uninflamed tissue = 12. Center lines = mean; whiskers = standard deviation. Source data are provided as a Source Data file

Fig. 5

Mapping paired samples reveals blood correlates of tissue immunity. a Of 2145 manually gated cell frequencies and median expression levels, 795 were significantly different (adjusted p ≤ 0.05; data not shown) between blood (N = 12) and tissue (N = 41). Of significantly different parameters, 55.35% were higher in blood (red) and 44.65% were higher in the tissue (purple). Significantly different parameters were used to construct the Pearson correlation map. Statistics: BH FDR-corrected unpaired two-tailed Student’s T test using Morpheus (see the Methods section). b Of 2145 parameters, three were correlated (green box and green wedge in pie chart) between blood and ileum tissue samples. Statistics: Pearson correlation coefficient (r) and p-values from Pearson correlation tests are shown. P-values for these three correlations were not corrected for multiple testing because hypothesis-driven tests were conducted based on biological insight after independent preliminary analysis. Sample sizes: ten blood/tissue pairs for CD3⁺ cells and central memory CD4 T cell median pCREB expression; nine blood/tissue pairs for CCR9⁺ switched memory B cells. Solid line = linear regression; dotted lines = 95% confidence interval. Source data are provided as a Source Data file

Fig. 6

Tissue leukocytes distinguish diseases and inflammation states. a Schematic of tissue comparisons. Created with BioRender. b Paired CD tissue from inflamed and uninflamed areas of the same region (N = 4 pairs consisting of samples 239_1 and 239_4, 249_3 and 249_5, 252_1 and 252_3, and 255_1 and 255_3; see Supplementary Table 2). Statistics: ratio paired T test (df = 3; CD45RO⁺ CD4 T cells, t = 5.48; CD27⁺CD45RO⁻ CD4 T cells, t = 5.74; CD43⁺ T cells, t = 3.76; CD25⁺CD19⁺CD20⁺ B cells, t = 3.51). P-values in b were not corrected for multiple testing because hypothesis-driven tests were conducted based on biological insight after independent preliminary analysis. Lines connect paired samples from the same subject. Source data are provided as a Source Data file

Fig. 7

Signatures in the blood classify IBD patients by clinical subsets. a Schematic of the approach for non-invasive classification of CD vs. UC based on blood. Blood was drawn from study subjects, peripheral blood mononuclear cells (PBMCs) were isolated and cryopreserved, and samples were analyzed in batches by CyTOF. Created with BioRender. b Generalized linear models (GLMs) were created for eight parameters significantly different between all CD and UC samples. Corresponding receiver-operating characteristic (ROC) curves are shown for single feature and eight feature models. All CD and UC samples were used to plot ROC curves. UC was used as baseline for the purposes of the GLMs, such that a true-positive indicates correct classification of a CD sample. Statistics: generalized linear models were constructed using glm in R (see the Methods section). Intercepts and parameter coefficients for each model are provided in Supplementary Table 4. Cutoffs and associated performance characteristics are discussed in the text. Source data are provided as a Source Data file