The 10,000 Immunomes Project: Building a Resource for Human Immunology

Abstract

There is increasing appreciation that the immune system plays critical roles not only in the traditional domains of infection and inflammation but also in many areas of biology, including tumorigenesis, metabolism, and even neurobiology. However, one of the major barriers for understanding human immunological mechanisms is that immune assays have not been reproducibly characterized for a sufficiently large and diverse healthy human cohort. Here, we present the 10,000 Immunomes Project (10KIP), a framework for growing a diverse human immunology reference, from ImmPort, a publicly available resource of subject-level immunology data. Although some measurement types are sparse in the presently deposited ImmPort database, the extant data allow for a diversity of robust comparisons. Using 10KIP, we describe variations in serum cytokines and leukocytes by age, race, and sex; define a baseline cell-cytokine network; and describe immunologic changes in pregnancy. All data in the resource are available for visualization and download at http://10kimmunomes.org/.

Figure 1.. Resource Development and Selected Applications

Data from 242 studies and 44,775 subjects (including flow cytometry and CyTOF, mRNA expression, secreted protein levels [including cytokines, chemokines, and growth factors], clinical lab tests, HAI titers, HLA type, and others) were collected from the NIAID Immunology Data and Analysis Portal, ImmPort (http://www.immport.org/). We hand curated the entire contents of ImmPort to filter for normal healthy control human subjects. Each of the 10 data types was systematically processed and harmonized. These data constitute the largest compendium to date of cellular and molecular immune measurements on healthy normal human subjects. Both the normalized data and their raw counterparts are openly available for visualization and download at http://10kimmunomes.org/.

Figure 2.

High-Throughput Secreted Protein Data: Characterizing the Range of Unperturbed Secreted Protein Levels in a Diverse Population (A) t-distributed stochastic neighbor embedding (tSNE) visualization of high-throughput secreted protein data, colored by study accession, reveals that much of the variance across the data is explained by batch. (B) After batch correction with an empirical Bayes algorithm, which accounts for both mean and variance difference across studies while maintaining effects of covariates such as age, sex, and race, the data no longer cluster by batch. (C) Secreted protein data as measured by multiplex ELISA across 17 studies captures known effects, such as elevated levels of serum leptin in female relative to male subjects (analysis of covariance [ANCOVA], n = 906, p = 9 × 10⁻²⁸). Each point represents an individual subject. Ribbons indicatethe mean and standard error ofeach group. (D) Analysis of the reference population reveals demographic associations, including elevated CXCL5 in African American subjects as compared to other races. (ANCOVA, n = 917, p values: *p < 0.05, **p < 0.01). Each dash represents an individual subject. The width of the violin represents the relative density of subjects at each value. The length of theviolin represents the range ofvalues. (E) We characterize the distribution of secreted protein levels from serum across the reference population (n = 1,286). Each dash represents an individual subject. The width of the violin represents the relative density of subjects at each value. The length of the violin represents the range of values.

Figure 3.

Mass Cytometry: Characterizing the Range of Cell-Subset Percentages in a Diverse Population (A) Distribution of cell-subset percentages across the 10KIP. The width of the violin represents the relative density of subjects at each value. The length of the violin represents the range of values. (B) Analysis of mass cytometry data reveals significant effects of age on cell-subset percentages while accounting for sex and race. Only cell-subset associations with Benjamini-Hochberg-corrected p values < 0.05 are shown. (ANCOVA, n = 578, *p < 0.05, **p < 0.01, ***p < 0.001). Effect sizes are displayed as Pearson’s r ± 95% confidence intervals. Each point represents the effect size of age. Error bars represent the standard error of that estimate. (C) Naive CD8+ T cells decrease significantly with age (ANCOVA, n = 565, p = 1.1 × 10~²¹), and central memory CD4 T cells increase significantly with age (ANCOVA, n = 578, p = 5.3 × 10⁻⁶), while accounting for sex and race. Each point represents an individual subject. Ribbons indicate the mean and standard error of each group. (D) Analysis of mass cytometry data reveals significant effects of sex on cell-subset percentages, while accounting for age and race. Only cell-subset associations with Benjamini-Hochberg-corrected p values < 0.05 are shown. (ANCOVA, n = 578, *p < 0.05, **p < 0.01, ***p < 0.001). Effect sizes are displayed as Cohen’s d ± 95% confidence intervals. Each point represents the effect size age. Errorbars representthestandard errorofthat estimate. (E) T cells (ANCOVA, n = 565, p = 7.4 × 10⁻⁶) and naive CD4+ T cells (ANCOVA, n = 578, p = 3.3 × 10⁻⁸) are significantly elevated in women as compared to men, accounting for age and race. Each dash represents an individual subject. The width of the violin represents the relative density of subjects at each value. The length of the violin represents the range of values.

Figure 4.. Immune Cell and Serum Cytokine Bipartite Graph

Immune cell percentages and serum protein concentrations, as measured by CyTOF and multiplex ELISA, were processed as described in STAR Methods, and the cell-cytokine relationship was described as partial correlations accounting for age, sex, and race. Only relationships significant at a BH-corrected p < 0.01 are shown.

Figure 5.. Comparing Pregnancy Data to the Common Control Reveals Cell-Subset and Immune Protein Modulation in Pregnancy

(A) PCA plot depicting the variation in serum proteins, as measured by multiplex ELISA, over the course of pregnancy, taken from ImmPort Study SDY36, as compared to multiplex ELISA measurements from women between the ages of18–40 from the reference population. The variance in measurements is dominated by a deviation in serum cytokine measurements during the first trimester(teal) relative to all other time points during pregnancy and relative to the 10KIP controls (green). These differences are driven primarily by changes in CCL2, CCL3, CCL4, CCL5, CCL11, IL6, and CXCL10. (B) As an example of cytokine modulation in pregnancy, serum CCL5 levels are significantly increased in the first and second trimester relative to the 10KIP controls, decrease during the third trimester and remain low for at least 6 weeks postpartum. CCL5 levels return to baseline levels by6months postpartum (ANOVA with Tukey HSD, n = 142 controls, n = 57 pregnancy, *p < 0.05, **p < 0.01, ***p < 0.001). (C) In contrast, serum IL15 levels make no significant deviations from normal over the course of pregnancy (ANOVA with Tukey HSD, n = 142 controls, n = 57 pregnancy). (D) PCA plot depicting the variation in immune cell subsets, as measured by flow cytometry, over the course of pregnancy, taken from ImmPort Study SDY36, as compared to cytometry measurements from women between the ages of18and 40yearsfromthe 10KIPcontrols.Asopposed to cytokine measurements(A), the preponderance of variation in cell-subset measurements is not due to changes over the course of pregnancy. All time points during and following gestation substantially overlap with the controls (green). (E) The percentage ofCD4+ T cells, as a fraction of lymphocytes, is significantly elevated over the duration of pregnancy but returns to baseline in the postpartum period (ANOVA with Tukey HSD, n = 94 controls, n = 57 pregnancy, *p < 0.05, ***p < 0.001). (F) The percentage of B cells, as a fraction of lymphocytes, exhibits a small but significant dip in the second and third trimesters (ANOVA with Tukey HSD, n = 94 controls, n = 57 pregnancy, *p < 0.05). (B, C, E, and F) The central line indicates the mean. Edges of the box represent the 25th and 75th percentile of the data. The upper whisker represents the smaller of the maximum value and the 75th percentile + 1.5 × the interquartile range. If the latter, outliers are represented by individual points. The lower whisker represents the larger of the minimum value and the 25th percentile — 1.5 × the interquartile range. If the latter, outliers are represented by individual points