Multimodal analysis of dysregulated heme metabolism, hypoxic signaling, and stress erythropoiesis in Down syndrome

SUMMARY

Down syndrome (DS), the genetic condition caused by trisomy 21 (T21), is characterized by delayed neurodevelopment, accelerated aging, and increased risk of many co-occurring conditions. Hypoxemia and dysregulated hematopoiesis have been documented in DS, but the underlying mechanisms and clinical consequences remain ill defined. We report an integrative multi-omic analysis of ~400 research participants showing that people with DS display transcriptomic signatures indicative of elevated heme metabolism and increased hypoxic signaling across the lifespan, along with chronic overproduction of erythropoietin, elevated biomarkers of tissue-specific hypoxia, and hallmarks of stress erythropoiesis. Elevated heme metabolism, transcriptional signatures of hypoxia, and stress erythropoiesis are conserved in a mouse model of DS and associated with overexpression of select triplicated genes. These alterations are independent of the hyperactive interferon signaling characteristic of DS. These results reveal lifelong dysregulation of key oxygen-related processes that could contribute to the developmental and clinical hallmarks of DS.

In brief

Donovan et al. report that individuals with Down syndrome display multi-omic signatures indicative of hypoxic signaling and altered heme metabolism, along with signs of stress erythropoiesis. These signatures associate with key proteomic and immune cell changes in Down syndrome and are independent of hyperactive interferon signaling.

Graphical Abstract

INTRODUCTION

Down syndrome (DS) is caused by trisomy 21 (T21), the most prevalent chromosomal abnormality and a leading cause of intellectual and developmental disability.¹ Over the last century, thanks to advances in medical care and inclusion in most aspects of society, people with DS have experienced a remarkable increase in life expectancy and improved health outcomes.¹ Epidemiological studies of this growing population reveal a unique clinical profile characterized by varying risks of co-occurring conditions throughout their lifespan, including decreased risk of most solid malignancies and hypertension,^2,3 and increased risk of congenital heart defects (CHDs), autism spectrum disorders, seizure disorders, obstructive sleep apnea (OSA), autoimmune disorders, severe complications from respiratory infections, diverse leukemias, and Alzheimer’s disease.^2,4–8 Despite many efforts, with a few exceptions, the mechanisms by which T21 causes these developmental and clinical phenotypes await elucidation. Therefore, research on the mechanisms driving the pathophysiology of DS could benefit both the population with DS and the general population affected by these conditions.

Humans experience hypoxia, a state of reduced oxygen availability, both as part of normal physiological processes and during pathological states. Inadequate responses to hypoxia have been linked to the etiology of various cardiovascular diseases,⁹ chronic lung conditions,¹⁰ certain cancers,¹¹ cognitive decline,¹² and immune dysfunction.¹³ In DS, OSA, associated with episodes of intermittent hypoxia, has been proposed to exacerbate cognitive and behavioral phenotypes.^14–18 Low oxygen saturation (below 85%) is the most common cause of neonatal intensive care unit admissions for neonates with DS,¹⁹ who experience high rates of hypoxemic events even in the absence of OSA and CHD.²⁰ Furthermore, T21 has been described to induce a pseudo-hypoxic state at the cellular level, as evidenced by metabolic reprogramming reminiscent of that observed in euploid cells upon oxygen deprivation.²¹ These observations indicate that T21 may impart an inherent predisposition to hypoxic signaling, which may contribute to the complex clinical manifestations of DS.

The production of new red blood cells (RBCs), or erythropoiesis, plays a critical role in the systemic response to hypoxia. Erythropoiesis is stimulated by the hormone erythropoietin (EPO), which in turn is induced by hypoxia-inducible factors (HIFs) in the kidneys.²² EPO stimulates other hematological changes including increases in RBC size, known as macrocytosis.²³ Interestingly, individuals with DS display distinct hematological changes, including decreased RBC counts and markers of macrocytosis such as increased mean corpuscular volume (MCV) and elevated mean corpuscular hemoglobin (MCH).^24,25 Additionally, children with DS have been shown to have elevated hemoglobin (Hgb) and hematocrit (Hct) levels.²⁶ These observations highlight the importance of unraveling the interplay between hypoxia, erythropoiesis, and the variable pathophysiology of DS.

Within this context, we report here a multidimensional analysis of hypoxic signaling and associated processes in a large cohort study of individuals with DS, the Human Trisome Project (HTP, NCT02864108). Matched analysis of whole-blood transcriptome, plasma proteome, and immune cell profiles demonstrates that individuals with DS show signs of elevated, yet variable, hypoxic signaling and dysregulated erythropoiesis across the lifespan. Transcriptome signatures indicative of increased heme metabolism and HIF activation are associated with elevated plasma levels of EPO and other biomarkers of hypoxia. Dysregulation of EPO in DS is accompanied by an altered hematological profile indicative of stress erythropoiesis and lymphopenia. HIF1A transcriptome signatures associate with an immune profile marked by elevated neutrophils and inflammatory monocytes, as well as depletion and remodeling of T and B cell lineages. Many of these phenomena are reproduced in the Dp16 mouse model of DS and associated with overexpression of specific chromosome 21 (chr21) genes. Furthermore, these events are independent of the elevated interferon (IFN) signaling characteristic of DS. Altogether, these results point to dysregulated heme metabolism and hypoxic signaling as potential contributors to the pathophysiology of DS and justify further investigations.

RESULTS

Signatures of dysregulated heme metabolism and hypoxic signaling in Down syndrome

Gene set enrichment analysis (GSEA) of whole-blood transcriptome data from the HTP²⁷ reveals that individuals with T21 not only show enhanced IFN responses and other inflammatory signatures, but they also display elevated expression of genes involved in heme metabolism (Figures 1A and S1A).²⁷ Ingenuity Pathway Analysis (IPA) of upstream regulator networks indicates that, beyond factors driving inflammatory signaling (e.g., IRFs [interferon regulatory factors]), HIF1A ranks 5^th among upstream regulators predicted to drive gene expression changes in T21 (Figure 1B; Table S1). Also predicted to be activated are GATA1 (ranked 13^th), a master regulator of erythropoiesis and RBC maturation,²⁸ and its downstream target NFE2 (ranked 18^th), which regulates expression of myriad erythroid-specific genes²⁹ (Table S1). Genes in the heme metabolism and HIF1A signatures upregulated in T21 display minimal overlap (Figures 1D–1G and S1B). The heme metabolism and HIF1A signatures also show little overlap with chr21-encoded genes or IFN-stimulated genes (ISGs) from the IFN gamma/alpha response signatures (Figures 1D and S1C).

Figure 1. Transcriptomic signatures of dysregulated heme metabolism and hypoxic signaling in Down syndrome.

(A and B) Heatmaps showing (A) normalized enrichment scores (NES) from GSEA and (B) Z scores from upstream regulator IPA for whole-blood transcriptome changes in T21 (n = 304) relative to D21 (n = 96). Asterisks represent significance (q < 0.1) after Benjamini-Hochberg adjustment for (A) and absolute Z score ≥2 for (B).

(C) Volcano plots summarizing whole-blood transcriptome changes in T21 relative to D21, highlighting chromosome 21 (chr21, left), heme metabolism (middle), and HIF1A (right) genes.

(D) Overlap between chr21 (n = 179), heme metabolism (n = 107), and HIF1A (n = 76) genes upregulated in T21 (fold change [FC] > 1, q < 0.1).

(E) Heatmap displaying expression (log₂FC) of genes in both the heme metabolism and HIF1A signatures in T21 relative to D21. Asterisks denote significance (q < 0.1) after Benjamini-Hochberg adjustment of p values.

(F) Sina plots displaying mRNA levels (log₂RPKM) for GATA1 and SLC2A1 in T21 and D21. Benjamini-Hochberg adjusted p values (q-values) are indicated.

(G) Heatmap displaying expression (log₂FC) of top 10 differentially expressed genes (DEGs) in the heme metabolism and HIF1A signatures in T21 relative to D21. Asterisks denote significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values.

(H and I) Sina plots displaying mRNA levels (log₂RPKM) for (H) heme metabolism and (I) HIF1A genes in T21 relative to D21. Benjamini-Hochberg adjusted p values (q-values) are indicated.

(J) Sina plots showing Heme (left) and HIF1A (right) scores in T21 relative to D21. p values from linear regressions are denoted.

(K) Scatterplot depicting relationship between Heme and HIF1A scores in T21. Spearman rho and p value are denoted. Points are colored by density; blue lines represent linear fit, with 95% confidence intervals in gray.

(L) Sina plots showing mRNA levels (median Z scores) of genes in Heme (left) and HIF1A (right) signatures in white blood cells (WBCs) and whole blood in a subgroup of participants (n = 12), with p values denoted.

(M) Scatterplot depicting expression (median Z scores) of genes from the Heme score in WBCs compared to whole blood.

For sina plots, boxes represent interquartile ranges and medians, with notches approximating 95% confidence intervals.

The heme metabolism signature encompasses genes that regulate hemoglobin oxygen affinity (e.g., CA1 and BPGM) and those essential for hemoglobin assembly and stabilization (e.g., HBD and AHSP) (Figure 1H; Table S1). It also includes critical erythrocyte glycoproteins (e.g., BCAM and GYPB), key factors for erythropoiesis and erythrocyte differentiation (e.g., TSPO2 and KLF1), as well as genes involved in heme biosynthesis (e.g., ALAS2 and FECH) (Figures 1H and S1D; Table S1).

In contrast, the HIF1A signature includes genes involved in extracellular matrix (ECM) remodeling and tissue repair (e.g., MMP1 and ADAMTS1) and those that contribute to inflammatory and immune responses (e.g., IFNG and CXCL8) (Figure 1I; Table S1). This signature also encompasses genes that regulate cell survival and apoptosis (e.g., BCL2L1 and BIRC5), modulate fibrinolysis and clotting (e.g., SERPINE1 and PLAUR), and contribute to cell adhesion and vascular function (e.g., ITGB2 and ICAM1) (Figures 1I and S1E; Table S1).

Despite the elevated expression of genes from both signatures in DS, there is notable inter-individual variability. To investigate this phenomenon, we generated composite expression scores for both signatures, referred to henceforth as Heme and HIF1A scores (see STAR Methods). Expectedly, both scores are significantly elevated in DS (Figure 1J). Despite having minimal gene overlap, the scores are highly correlated in individuals with T21 (Figure 1K). Interestingly, whereas the Heme score is slightly higher in euploid females compared to males, this difference is not observed in individuals with T21 (Figure S1F), nor are there sex disparities in HIF1A scores for either karyotype group (Figure S1G). Finally, both scores are significantly associated with age and body mass index (BMI) in both euploid controls and individuals with T21, with the Heme score revealing a particularly pronounced association with BMI (Figures S1H–S1K).

To explore potential sources driving these expression signatures in the whole blood, we conducted a separate transcriptomic analysis on a subset of HTP participants (n = 12) comparing the expression of genes from both the Heme and HIF1A scores between RNA-seq data derived from whole blood versus isolated white blood cells (WBCs). These participants were chosen via a random selection process to provide a representative cross-section of the overall cohort. This analysis revealed that Heme score genes are expressed at lower levels in WBCs compared to whole blood, with some genes not expressed at all (Figures 1L and 1M). In contrast, HIF1A genes showed more similar expression distributions across both specimen types (Figures 1L and S1L). This indicates that the Heme signature is most likely not driven by gene expression changes in WBCs but rather residual RNA from enucleated blood cells (e.g., reticulocytes and platelets).

Taken together, these results demonstrate individuals with T21 display robust transcriptomic signatures of dysregulated heme metabolism and hypoxic signaling captured by composite mRNA expression scores.

Elevated heme metabolism and hypoxic signaling are associated with multi-omic markers of stress erythropoiesis in trisomy 21

We previously reported that T21 causes notable dysregulation of the plasma proteome.^27,30 To define the interplay between these proteomic changes and elevated heme metabolism and HIF1A signaling, we used Spearman correlation analysis to identify associations between the Heme and HIF1A scores and ~4,600 proteins in the matching HTP SomaScan plasma proteomics dataset (Figures 2A and 2B; Table S2). Comparing correlations across the two scores reveals remarkably similar associations (Figure 2C). As an example, the transferrin receptor (TFRC), which is required for cellular iron uptake and erythropoiesis,³¹ is strongly correlated with both scores, albeit with a higher correlation against the Heme score (Figures 2C and 2D). Both scores also positively correlate with EPO, general inflammatory markers (e.g., CRP and SAA1), and complement subunits (e.g., CFB and C9) in addition to several other factors (Figures 2E, S2A, and S2B; Table S2).

Figure 2. Elevated heme metabolism and HIF1A signaling are associated with multi-omic markers of stress erythropoiesis in trisomy 21.

(A and B) Volcano plots depicting Spearman correlations for (A) Heme and (B) HIF1A scores against plasma proteins (n = 4,601) in T21 (n = 304, one technical replicate). Significant correlations (q < 0.1) are colored red.

(C) Scatterplot comparing correlations against plasma proteins for the Heme score (x axis) and HIF1A score (y axis). Spearman rho and p value are denoted.

(D and E) Sina plots (left) displaying plasma levels (relative abundance) of TFRC (D) and EPO (E) in T21 (n = 316) and euploid controls (D21, n = 103); Benjamini-Hochberg adjusted p values (q-values) are denoted. Scatterplots (right) displaying relationship with Heme and HIF1A scores in T21; Spearman rho and Benjamini-Hochberg adjusted p values (q-values) are denoted.

(F) Volcano plot depicting Spearman correlations for EPO against plasma proteins. Significant correlations (q < 0.1) are colored red.

(G) Scatterplots comparing relationship between plasma levels (relative abundance) of EPO (x axis) and TFRC and GDF15 (y axis) in T21, with Spearman rho and Benjamini-Hochberg adjusted p values (q-values) denoted.

(H) Sina plot showing plasma levels (relative abundance) of GDF15 in T21 and D21, with Benjamini-Hochberg adjusted p values (q-values) denoted.

(I) Sina plots displaying whole-blood mRNA levels of HBG1, HBG2, and BCL11A in T21 (n = 304) and D21 (n = 96), with Benjamini-Hochberg adjusted p values (q-values) denoted.

(J) Heatmap comparing levels (median Z score) of complete blood count (CBC) parameters in T21 (n = 147, one technical replicate) and D21 (n = 171). Asterisks denote significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values.

(K) Sina plots showing indicated CBC parameters in T21 and D21, with units and Benjamini-Hochberg adjusted p values (q-values) denoted.

For scatterplots, points are colored by density; blue lines represent linear fit, with 95% confidence intervals in gray. Spearman rho and p value comparing sets of correlations are denoted.

For sina plots, boxes represent interquartile ranges and medians, with notches approximating 95% confidence intervals.

Considering the critical role of EPO in facilitating erythropoiesis and RBC production in response to hypoxia, driven primarily by HIF2A activation,^32,33 we further investigated the implications of elevated EPO in T21. Individuals with T21 consistently exhibit higher EPO levels compared to euploid controls across the lifespan, with no significant associations with age in either group (Figure S2C). When defining Spearman correlations against all other plasma proteins (Figure 2F; Table S2), TFRC emerges as the protein most strongly correlated with increased EPO production in individuals with T21 (Figures 2F and 2G; Table S2). Additionally, EPO levels correlate with GDF15, which is notably higher in individuals with T21 (Figures 2F–2H). GDF15, a pleiotropic cytokine involved in cellular stress responses to multiple disease states such as hypoxia and inflammation,³⁴ is a master regulator of stress erythropoiesis,³⁵ a compensatory mechanism leading to accelerated extramedullary RBC production during conditions of severe anemia or hypoxia.³⁶ Elevated levels of TFRC, EPO, and GDF15 are hallmarks of stress erythropoiesis,³⁶ and elevated GDF15 levels are observed during hypoxic conditions³⁵ and in patients with thalassemia syndromes, congenital dyserythropoiesis, and acquired sideroblastic anemias.³⁷ Stress erythropoiesis is characterized by a shift toward expression of fetal hemoglobin, where beta globin subunits are replaced by fetal gamma globin subunits.³⁶ Analysis of whole-blood mRNA revealed that individuals with T21 overexpress the gamma globin genes HBG1 and HBG2 and express lower levels of BCL11A, the major repressor of fetal globin expression (Figure 2I).

Sudden stimulation of erythropoiesis by EPO induces increases in RBC size, known as macrocytosis, in both humans and mice.²³ Additionally, it is well established that individuals with DS display altered hematological parameters, including macrocytosis and reduced RBC counts.^24,25 Furthermore, elevated MCV, a marker of macrocytosis, has been reported in patients with hypoxia-associated conditions, such as hypoxic chronic obstructive pulmonary disease.³⁸ In line with these previous reports, analysis of complete blood count (CBC) data revealed significant differences in multiple blood parameters in research participants with DS, including elevated MCV, MCH, Hgb, and Hct alongside decreased RBC, WBC, and lymphocyte counts (Figures 2J, 2K, and S2D). Plotting levels of these parameters across all ages reveals dysregulation in individuals with T21 across the lifespan, with clear age trajectories (Figure S2E). Both MCV and MCH are elevated in T21 across the lifespan and show a significant increase with age in both populations. Conversely, lymphocytes are consistently lower in T21 and decrease with age. Interestingly, however, the age trajectory for RBC counts in T21 is opposite of that for euploid controls. Whereas RBCs decline with age in T21, they increase in euploid individuals.

Collectively, these results indicate that the Heme and HIF1A signatures are associated with hallmarks of stress erythropoiesis in DS.

Hypoxic signaling associates with immune remodeling in trisomy 21

We recently completed immune cell profiling of 388 research participants (284 with T21 versus 90 euploid controls) using mass cytometry, which revealed global immune remodeling across all major myeloid and lymphoid lineages in DS.²⁷ To investigate which of these changes could be associated with elevated heme metabolism and HIF1A signaling, we carried out unsupervised clustering using the FlowSOM algorithm³⁹ to identify major immune cell lineages among all live cells, CD45⁺ CD66^lo non-granulocytes, T cells, and B cells (see STAR Methods), and we defined their relative abundances by T21 status and association with the Heme or HIF1A scores. Although no associations were found with the Heme score, likely due to the fact that these genes are mostly expressed in non-WBCs (Figures 1L and 1M), we found many interesting associations with the HIF1A score (Figures 3A and S3A–S3D; Table S3).

Figure 3. Hypoxic signaling associates with immune remodeling in trisomy 21.

(A) Overview of major immune cell lineages with heatmaps showing (left) effect of T21 and (right) association with HIF1A score. Columns on the left depict differential abundance comparison between individuals with T21 (n = 284, one technical replicate) and euploid controls (D21, n = 90, one technical replicate), shown as log₂(fold change) from beta regression analysis. Asterisks denote significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values. Columns on the right depict association with HIF1A score, shown as log₂(fold change) associated with each unit increase in HIF1A score. Asterisks denote significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values.

(B–J) Sina plots (left) displaying levels (% of indicated cell population) of specified immune cell lineages in T21 (n = 284) and D21 (n = 90), and scatterplots (right) displaying relationship with HIF1A scores in T21. Benjamini-Hochberg adjusted p values (q-values) from beta regressions are denoted. For sina plots, boxes represent interquartile ranges and medians, with notches approximating 95% confidence intervals. For scatterplots, points are colored by density; blue lines represent linear fit, with 95% confidence intervals in gray.

For example, among granulocytes, neutrophils are both mildly elevated in T21 and positively associated with the HIF1A score (Figures 3A and 3B). However, the striking elevation of basophils observed in DS is not associated with increased HIF1A signaling (Figures 3A and S3A). Pro-inflammatory monocytes (i.e., intermediate and non-classical), which are increased in frequency in DS, also associate positively with HIF1A signaling (Figures 3A and 3C). Conversely, within the CD45⁺CD66^lo population, double-negative (CD56⁻CD16⁻) natural killer cells are reduced in DS and inversely associated with the HIF1A score (Figures 3A and 3D). A notable finding is the depletion of total T cells (among all live cells) in T21, indicative of T cell lymphopenia, which associates with elevated HIF1A scores (Figures 3A and 3E). This depletion coincides with an increase in CD8⁺ terminally differentiated effector memory T cells, which are positively associated with the HIF1A score, and a corresponding decrease in CD8⁺ naive T cells, which negatively associate with the HIF1A score (Figures 3A, 3F, and 3G). Furthermore, B cell lymphopenia, a hallmark of DS, is also associated with the HIF1A score (Figures 3A and 3H). This reduction in total B cells among live cells is accompanied by relative enrichment of plasmablasts and age-associated B cells, within the B cell lineage, which also associate with elevated HIF1A signaling (Figures 3A, 3I, and 3J).

Together, these analyses reveal that whereas some immune hallmarks of DS are not associated with elevated HIF1A signaling (e.g., elevated basophils), many other important changes are, including increased frequencies in key lymphoid subsets (e.g., neutrophils, inflammatory monocytes), T cell differentiation, and B cell lymphopenia.

A mouse model of Down syndrome displays dysregulated heme metabolism, hypoxia across multiple tissues, and stress erythropoiesis

Next, we investigated whether dysregulation of heme metabolism, HIF1A signaling, and associated phenomena were conserved in mouse models of DS. Toward this end, we first analyzed CBC data from the Dp10 (B6; 129S7-Dp(10Prmt2-Pdxk)2Yey/J), Dp16 (B6.129S7-Dp(16Lipi-Zbtb21)1Yey/J), and Dp17 (B6; 129S7-Dp(17Abcg1-Rrp1b)3Yey/J) mouse models, which harbor segmental duplications of regions of mouse chromosomes 10, 16, and 17 that are syntenic to human chr21.⁴⁰ Notably, whereas the Dp16 model demonstrates hallmarks of macrocytosis (increased MCV and MCH and decreased RBCs), this is not the case for Dp10 or Dp17 mice (Figures 4A, 4B, S4A, and S4B). As part of our extensive ongoing characterization of Dp16 mice, we have generated transcriptome data from multiple tissues/organs and time points.^27,41 GSEA of these data reveals that hallmark heme metabolism and hypoxia signatures are enriched across multiple tissues in the Dp16 mice (Figures 4C, 4D, and S4C). IPA also predicted activation of the HIF1A signature across multiple tissues (Figure 4E).

Figure 4. A mouse model of Down syndrome displays dysregulated heme metabolism, hypoxia across multiple tissues, and stress erythropoiesis.

(A) Heatmap displaying differential abundance (log₂ fold change) of complete blood count (CBC) parameters in Dp10 (n = 9, one technical replicate), Dp16 (n = 6, one technical replicate), and Dp17 (n = 6, one technical replicate) mice relative to wild-type (WT, equal sample sizes) mice. Asterisks denote significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values.

(B) Sina plots showing transformed (log₂) CBC parameters in Dp16 and WT mice, with units and Benjamini-Hochberg adjusted p values (q-values) denoted.

(C and D) GSEA of transcriptome data comparing Dp16 to WT mice, depicting normalized enrichment scores (NES) and transformed (−log₁₀) q-values for the hallmark heme metabolism (D) and hypoxia (E) signatures across tissues. Vertical dashed line indicates significance (q < 0.1) threshold.

(E) Upstream regulator IPA of transcriptome data comparing Dp16 to WT mice, depicting activation Z scores. Asterisks indicate absolute Z score ≥2.

(F–H) Sina plots showing mRNA levels (log₂RPKM) of select genes from the (F) hallmark heme metabolism signature in the spleen, (G) hallmark hypoxia signature in the adult heart, and (H) IPA HIF1A signature in embryonic day 18.5 (E18.5) hearts of Dp16 and WT mice, with Benjamini-Hochberg adjusted p values (q-values) denoted.

(I) Sina plots showing mRNA levels (log₂RPKM) of Epo in the kidneys of Dp16 and WT mice, with Benjamini-Hochberg adjusted p values (q-values) denoted.

(J) Overlap of differentially expressed genes (DEGs) between specified gene signatures in whole blood from individuals with T21 and tissues showing the highest enrichment of each signature in Dp16 mice.

For sina plots, boxes represent interquartile ranges and medians, with notches approximating 95% confidence intervals.

Heme metabolism shows the most prominent enrichment in the spleen, as evidenced by the overexpression of genes such as Tspo2, Klf1, Alas2, Fech, Gata1, Nfe2, Tal1, and Tfrc (Figure 4F), all of which are elevated in the bloodstream of individuals with DS (Figures 1 and S1; Table S1). Although we could not assess gamma globin expression due to absence of these genes in mice, we observed a decrease in the expression of the Bcl11a repressor in the spleen of Dp16 mice (Figure 4F), indicating transcriptional changes associated with stress erythropoiesis. Despite the limitation that these measurements were not made with mouse whole blood, the spleen, as a critical secondary lymphoid organ, contains a variety of hematopoietic and immune cells involved in blood filtration and storage⁴² and is a key site of stress erythropoiesis.³⁶

Conversely, the hypoxia signature is most prominent in embryonic (E12.5 and E18.5) and adult hearts, followed by embryonic facial mesenchyme, brain cortex, liver, and kidney (Figures 4D and S4C). Notably, both Gdf15 and Cdkn1a are upregulated in the heart of Dp16 mice across all time points analyzed (Figures 4G and S4D). Additional overexpressed hypoxia-related genes in the heart include Plac8 and Plaur in adults, Maff and Pygm at E12.5, and Eno2 and Hmox1 at E18.5 (Figures 4G and S4D).

Examples of genes from the HIF1A signature overexpressed in the heart at E18.5 include Icam1 and Plod2, which were observed as upregulated in the whole blood of individuals with DS, in addition to Agt and Epor (Figure 4H). Concurrent with these results, the kidneys of Dp16 mice also show overexpression of Epo (Figure 4I).

Overlap analyses demonstrate statistically significant commonalities between the dysregulated signatures in human whole blood and heme signatures in the murine spleen, hypoxia signatures in the embryonic and adult murine heart, and HIF1A signatures across multiple tissues (Figures 4J, S4E, and S4F), underscoring that the Dp16 strain is a suitable model for investigating dysregulation of heme metabolism, hypoxic signaling, and stress erythropoiesis in DS.

Heme metabolism and HIF1A signatures associate with expression of distinct chromosome 21 genes but not with interferon hyperactivity

Next, we investigated the association between expression of chr21 genes and elevated heme metabolism and HIF1A signaling in DS. To this end, we conducted Spearman correlation analyses between whole-blood expression of chr21 genes (n = 179) and the Heme and HIF1A scores. This revealed numerous positive and inverse correlations between both scores and chr21 genes (Figures 5A, 5B, S5A, and S5B; Table S4), with overall similar trends (Figure 5C). Interestingly, the chr21 gene showing the strongest correlation with both scores is ABCC13, which has been described as a pseudo-gene incapable of producing a functional protein⁴³ and is highly overexpressed in individuals with T21 (Figures 5D–5F). This observation suggests that ABCC13 transcription may be regulated by similar mechanisms driving the enrichment of the heme metabolism and HIF1A signatures in T21, indicative of a possible interconnected regulatory network. Beyond ABCC13, both scores correlate with several other notable chr21 genes, including ERG, MAP3K7CL, ITSN1, and TFF3 (Figures 5A–5F, S5A, and S5B). These scores also exhibit unique associations with distinct chr21 genes. For example, the Heme, but not HIF1A, score correlates with expression of APP and BACE2 (Figures S5A and S5B). Conversely, the HIF1A, but not Heme, score correlates with expression of the IFN receptors IL10RB and IFNGR2 (Figures S5A and S5B).

Figure 5. Heme metabolism and HIF1A signatures associate with expression of distinct chromosome 21 genes but not with interferon hyperactivity.

(A and B) Volcano plots depicting Spearman correlations for (A) Heme and (B) HIF1A scores against whole-blood mRNA levels of chromosome 21 (chr21) genes in T21 (n = 304, one technical replicate). Significant correlations (q < 0.1) are colored red.

(C) Scatterplot comparing correlations against plasma proteins for the Heme score (x axis) and HIF1A score (y axis). Spearman rho and p value are denoted.

(D) Heatmaps depicting for chr21 genes: (left) differential expression in T21 (n = 304) vs. D21 (n = 96), (middle) correlations with Heme and HIF1A scores, and (right) which mouse models harbor triplicated copies. Asterisks indicate significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values.

(E and F) Scatterplots displaying relationship between mRNA levels (log₂RPKM) of chr21 genes and the Heme (E) and HIF1A (F) scores in individuals with T21 (n = 304). Spearman rho and Benjamini-Hochberg adjusted p values (q-values) are denoted.

(G) Sina plot displaying expression changes while a research participant was taking a JAK inhibitor (JAKi) for interferon-stimulated genes (ISGs), hallmark heme metabolism, hallmark hypoxia, and HIF1A score signatures. Red points indicate significance (q-value < 0.1) after Benjamini-Hochberg adjustment of p values. One technical replicate was used for sequencing.

(H) Sina plots displaying expression changes of (left) hallmark heme metabolism genes in the spleen, (middle) hallmark hypoxia genes in the adult heart, and (right) HIF1A genes in E18.5 hearts when comparing Dp16 (n = 6) to wild-type (WT, n = 6) mice (blue) and Dp16 mice with normalized interferon receptor copy number (Dp16^2xIfnrs, n = 6) to Dp16 (n = 6) (yellow).

(I and J) Sina plots displaying mRNA levels (log₂RPKM) of select genes from spleen heme metabolism signature (I) and adult heart hypoxia signature (J) in WT, Dp16, and Dp16^2xIfnrs mice, with Benjamini-Hochberg adjusted p values (q-values) denoted.

(K) Scatterplot depicting mRNA levels (median Z scores) for chr21 in white blood cells (WBCs) compared to whole blood.

For scatterplots, points are colored by density; blue lines represent linear fit, with 95% confidence intervals in gray.

For sina plots, boxes represent interquartile ranges and medians, with notches approximating 95% confidence intervals.

Given the well-documented impact of hyperactive IFN signaling on DS phenotypes^{27,30,41,44–49} and the complex interplay between inflammation, hypoxia, and stress erythopoiesis,^36,50 we next investigated how attenuating IFN signaling impacts expression of genes in the heme metabolism and HIF1A/hypoxia signatures. Toward this end, we first analyzed whole-blood transcriptome data from a research participant with DS while taking the JAK inhibitor tofacitinib for treatment of alopecia areata.²⁷ As previously reported, this participant provided blood samples while off and on the medicine.²⁷ This analysis confirmed downregulation of many ISGs (e.g., RSAD2 and IFIT1) when the participant was on the medicine but no significant changes in the heme metabolism, hypoxia, or HIF1A signatures (Figure 5G). When analyzing CBCs from this participant, we find that while WBCs increase with JAK inhibition, there are no changes in markers of macrocytosis or dyserythropoiesis (Figures S5C and S5D).

Next, we analyzed transcriptome data generated from a previously published modified Dp16 mouse model (Dp16^2xIfnrs), in which the copy number of the Ifnr locus has been normalized from three to two via CRISPR-mediated genome editing.⁴¹ Whereas the genetic correction does indeed reverse Ifnr overexpression (Figure S5E), it does not attenuate overexpression of genes in the heme metabolism, hypoxia, or HIF1A signatures (Figure 5H), as illustrated by Klf1 and Alas2 in the spleen or Plac8 and Gdf15 in the adult heart (Figures 5I and 5J).

Lastly, we analyzed the relative expression of chr21 genes in the whole blood versus WBC transcriptome (Figure 5K), which revealed genes more highly expressed in the whole-blood dataset, such as ABCC13 and TFF3, versus genes preferentially expressed in the WBC dataset, such as AGPAT3 and RRP1B (Figures S5F and S5G). This indicates that some chr21 genes such as ABCC13 and TFF3 are expressed preferentially in non-WBCs, with potential roles in affecting RBC formation and function, which would require further investigations.

Collectively, although these results do not identify a single chr21 gene as potential driver of dysregulated heme metabolism and hypoxic signaling in DS, they narrow down the potential list of triplicated genes that could be driving these effects.

DISCUSSION

Over the last century, people with DS have experienced significant increases in life expectancy and well-being.⁵¹ Today, thanks to appropriate medical care and greater inclusion in most aspects of society, children with DS have opportunities that would have been unthinkable a few decades ago. Furthermore, for the first time in history, there is now a large population of adults with DS who are thriving members of society and challenging outdated preconceptions about what is possible when living with an extra chromosome. However, despite these advances, individuals with DS have a significantly shorter lifespan and increased medical care needs relative to the general population, as they are more likely to develop a wide range of co-occurring conditions.^6,8

Respiratory complications and lung diseases stand as leading causes of mortality and morbidity among individuals with DS.^5,8 These complications include not only respiratory conditions such as pulmonary hypertension, recurrent lung infections, and OSA but also inherent morphological abnormalities in the respiratory system, affecting the entirety of the lung in addition to vascular and lymphatic defects.⁵² The prevalence of these respiratory issues places individuals with DS at elevated risk of chronic hypoxia, which may exacerbate existing co-occurring conditions and potentially trigger a range of secondary complications from cognitive decline to cardiovascular stress.^14–18,53 Even though hypoxia may have a significant influence on clinical outcomes for individuals with DS, its etiology is complex and understudied.

While previous studies have documented abnormal hematological parameters⁵⁴ and confirmed the presence of chronic hypoxia in individuals with DS,^20,55 our analysis of whole-blood transcriptomes revealed robust signatures of dysregulated heme metabolism and increased HIF activity relative to euploid controls. Interestingly, these transcriptomic changes were captured in composite scores, composed of distinct genes, yet exhibiting a high degree of correlation. These findings raise intriguing questions regarding the interplay between these two signatures in individuals with T21. Dysregulation in heme metabolism can potentially contribute to inadequate oxygen supply, leading to or exacerbating a hypoxic state. Indeed, elevated heme biosynthesis has been shown to increase the severity of OSA, as demonstrated by a higher percentage of sleeping time in hypoxemia.⁵⁶ Conversely, hypoxic conditions could lead to altered heme metabolism.⁵⁷ Therefore, future studies should aim to investigate both the origins and ramifications of these signatures in DS in addition to the complex cause-effect relationships between them.

Our multimodal analysis, integrating Heme and HIF1A scores with high-throughput plasma proteomic profiling and CBC analysis, demonstrated clear signs of stress erythropoiesis in individuals with DS. This is particularly underscored by elevated levels of TFRC, EPO, GDF15, and fetal hemoglobins. TFRC, essential for the cellular uptake of iron and erythropoiesis,³¹ and EPO, the principal stimulator of erythropoiesis, are both crucial for healthy RBC genesis and turnover. Remarkably, despite elevated EPO levels, typically indicative of a compensatory response to hypoxia,⁵⁸ individuals with T21 exhibit decreased RBCs. However, this is concurrent with elevated MCV and MCH, hallmarks of macrocytosis,⁵⁹ which have been shown to be stimulated by ectopic EPO supplementation in both humans and mice.²³ This hematological profile, which also coincides with elevated Hgb and Hct, indicates RBCs in T21 are fewer in number but larger and possibly carrying more Hgb per cell. These findings, along with elevated GDF15 levels and increased expression of fetal globin genes, are consistent with stress erythropoiesis, a compensatory mechanism for extramedullary RBC production.³⁶

DS is characterized by distinct immunological manifestations, and our study provides robust evidence of a connection between hypoxic signaling and key immunological changes in T21. HIF1A scores are associated with pro-inflammatory shifts in myeloid lineages, such as neutrophilia and increased frequencies of pro-inflammatory monocytes, coupled with T and B cell lymphopenia and a shift toward more differentiated T and B cell lineages. Previous studies have documented that hypoxia and HIF1A signaling can profoundly influence immune cell landscapes. For example, hypoxia enhances neutrophil survival through an HIF1A-dependent mechanism⁶⁰ and also influences T cell differentiation, survival, and function.⁶¹

By utilizing three different mouse models of DS, each harboring triplications of distinct genes syntenic to human chr21, we were able to determine that the Dp16 model alone displays hematological indicators of hypoxia-induced macrocytosis. Moreover, aside from ABCC13, which lacks a murine paralog, chr21 genes triplicated in the Dp16 model, namely ERG and MAP3K7CL, show the strongest correlations with the Heme and HIF1A scores. Intriguingly, ERG, a transcription factor, has been shown to be part of a regulatory network involving TAL1 and GATA2, crucial for the transition from hematopoietic stem cells to erythroid cells.⁶² While future mechanistic studies are needed to define the exact mechanisms by which T21 dysregulates heme metabolism and hypoxic signaling, our results narrow down the number of potential driver genes and indicate that these phenomena are largely independent of the well-documented elevation in IFN signaling in DS.^{27,30,41,44–49}

Limitations of the study

First, results from our cross-omics correlation analyses should be interpreted with caution and without inferring cause-effect relationships. For example, whereas the strong correlation between HIF scores and EPO levels agrees with the that fact that EPO transcription is stimulated by HIFs,^22,63 EPO levels can also be induced by other stimuli, such as chronic kidney disease.⁶⁴ Second, at the current sample size, we could not define strong associations between clinical variables other than age and BMI and the observed dysregulation of heme metabolism and hypoxic signaling. Defining whether these signatures are associated with OSA, cardiovascular, or pulmonary conditions common in DS will require additional investigations in larger cohort studies. Third, our dataset does not include measurements of oxygen tension in the research participants at the time of (or days prior to) the blood draws used for multi-omics data generation. Additional studies will be needed to define the degree to which our results are associated with lower oxygen availability in DS.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Joaquin M. Espinosa (joaquin.espinosa@cuanschutz.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• All data needed to evaluate the conclusions in this paper are are publicly available as of the date of publication. Sample/participant metadata has been deposited at the Synapse data sharing platform (syn31488784: https://doi.org/10.7303/syn31488784). Whole blood RNA-seq data from HTP participants have been deposited at Synapse (syn31488780: https://doi.org/10.7303/syn31488780), the INCLUDE Data Hub, and Gene Expression Omnibus (GEO: GSE190125). White blood cell RNA-seq data from HTP participants have been deposited at Gene Expression Omnibus (GEO: GSE251967). Plasma SomaScan proteomics data have been deposited at Synapse (syn31488781: https://doi.org/10.7303/syn31488781). Immune cell mass cytometry data from HTP participants have been deposited at Synapse (syn31488783: https://doi.org/10.7303/syn31488783). Murine RNA-seq data have been deposited at Gene Expression Omnibus (adult heart, GEO: GSE218883; adult whole brain, GEO: GSE218885; adult hippocampus, GEO: GSE272567; adult cortex, GEO: GSE272568; adult kidney, GEO: GSE272569; adult liver, GEO: GSE272570, adult spleen, GEO: GSE272572; embryonic (E10.5) facial mesenchyme, GEO: GSE218887, embryonic (E12.5) heart, GEO: GSE218888; embryonic (E18.5) heart, GEO: GSE218889; adult mesenteric lymph nodes, GEO: GSE218890; adult lung, GEO: GSE229762.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Human Trisome Project participants

The human subjects data, results, and analyses presented in this report come from a nested study within the Crnic Institute’s Human Trisome Project cohort study (HTP, NCT02864108). The protocol for the HTP was approved by the Colorado Multiple Institutional Review Board (COMIRB 15–2170, see also www.trisome.org). Written informed consent was provided by all study participants or their legal guardians. Medical records and participant reports were used to curate detailed clinical histories for each participant. Participant meta-data including trisomy status, age, sex, body mass index (BMI) and sample source location were also collected and are publicly available (see key resources table). In cases where appropriate, these data were used to inform on the influence of these covariates on the results. Participants include both males and females aged 1–61 years. For data in Figure 5G, a research participant enrolled in the HTP biobank received intermittent treatment with tofacitinib (Xeljanz, 5 mg doses, once to twice a day) for alopecia areata with remarkable hair regrowth while on the medicine.⁶⁵ Over the course of ~3 years, the participant provided research blood draws when on the medicine and during periods of voluntary treatment interruption.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Fc Receptor Binding Inhibitor polyclonal	eBioscience/ThermoFisher Scientific	Cat # 14-9161-73; RRID: AB_468582
Mouse monoclonal Anti-Human CD45 (Clone HI30)	Fluidigm	Cat # 3089003B; RRID: AB_2661851
Mouse monoclonal anti-human CD66a/c/d/e (clone B1.1/CD66)	BD Biosciences	Cat # 551354; RRID: AB_394166
Mouse monoclonal anti-human Cleaved PARP (Clone F21-852)	BD Biosciences	Cat # 552596, RRID: AB_394437
Mouse monoclonal anti-human RORgt (Clone RORg2)	BioLegend	Cat # 646502, RRID: AB_2238503
Mouse monoclonal anti-human IgM (clone MHM-88)	BioLegend	Cat # 314502; RRID: AB_493003
Mouse monoclonal anti-human PICP (Clone PCIDG10)	Millipore	Cat # MAB1913, RRID: AB_94406
Mouse monoclonal anti-human CD19 (clone HIP19)	Fluidigm	Cat # 3142001; RRID: AB_2651155
Mouse monoclonal anti-human CD123 (clone 6H6)	Fluidigm	Cat # 3143014B, RRID: AB_2811081
Mouse monoclonal anti-human CD11b (Clone ICRF44)	Fluidigm	Cat # 3144001, RRID: AB_2714152
Mouse monoclonal anti-human CD4 (clone RPA-T4)	Fluidigm	Cat # 3145001; RRID: AB_2661789
Mouse monoclonal anti-human IgD (clone IA6-2)	Fluidigm	Cat # 3146005B; RRID: AB_2811082
Mouse monoclonal anti-human CD11c (clone Bu15)	Fluidigm	Cat # 3147008; RRID: AB_2687850
Mouse monoclonal anti-human TCR Va7.2 (Clone 3C10)	Biolegend	Cat # 351702, RRID: AB_10900258
Mouse monoclonal anti-human CD127 (clone A019D5)	Fluidigm	Cat # 3149011; RRID: AB_2661792
Mouse monoclonal anti-human CD1c (clone L161)	Biolegend	Cat # 331501; RRID: AB_1088996
Mouse monoclonal anti-human CD14 (clone M5E2)	Fluidigm	Cat # 3151009B, RRID: AB_2810244
Mouse monoclonal anti-human TCRgd (Clone 11F2)	Fluidigm	Cat # 3152008B, RRID: AB_2687643
Mouse monoclonal anti-human CD7 (clone CD7-6B7)	DVS Sciences	Cat # 3153014B; RRID: AB_2811084
Mouse monoclonal anti-human CD3 (clone UCHT1)	DVS Sciences	Cat # 3154003B; RRID: AB_2811086
Mouse monoclonal anti-human CD279/PD1 (clone EH12.2H7)	Fluidigm	Cat # 3155009B; RRID: AB_2811087
Mouse monoclonal anti-human PD-L1 (clone 29E.2A3)	Fluidigm	Cat # 3156026, RRID: AB_2687855
Mouse monoclonal anti-human CD45RA (clone HI100)	BioLegend	Cat # 304102; RRID: AB_314406
Mouse monoclonal anti-human EMR1 (Clone BM8)	Biolegend	Cat # 123102, RRID: AB_893506
Mouse monoclonal anti-human FOXP3 (clone 259D/C7)	Fluidigm	Cat # 3159028A; RRID: AB_2811088
Mouse monoclonal anti-human T-bet (Clone 4B10)	Fluidigm	Cat # 3160010B, RRID: AB_2810251
Mouse monoclonal anti-human CD16 (clone B73.1)	Biolegend	Cat # 360702; RRID: AB_2562693
Mouse monoclonal anti-human CD8a (clone RPA-T8)	Fluidigm	Cat # 3162015; RRID: AB_2661802
Mouse monoclonal anti-human CD34 (clone 581)	Fluidigm	Cat # 3163014B; RRID: AB_2811091
Mouse monoclonal anti-human CD45RO (clone UCHL1)	Fluidigm	Cat # 3164007B; RRID: AB_2811092
Mouse monoclonal anti-human CD15 (Clone W6D3)	Biolegend	Cat # 323002, RRID: AB_756008
Rabbit monoclonal anti-human phospho-4E-BP1 (Thr37/Thr46) (clone 236B4)	Cell Signaling Technology	Cat # 2855; RRID: AB_560835
Mouse monoclonal anti-human CD27 (clone L128)	Fluidigm	Cat # 3167006B; RRID: AB_2811093
Rabbit monoclonal anti-human phospho-STAT1 (Tyr701) (clone 58D6)	Cell Signaling Technology	Cat # 9167; RRID: AB_561284
Mouse monoclonal anti-human CD25 (clone 2A3)	Fluidigm	Cat # 3169003; RRID: AB_2661806
Mouse monoclonal anti-human CD33 (clone WM53)	Biolegend	Cat # 303402, RRID: AB_314346
Mouse monoclonal anti-human CD95 (clone DX2)	Biolegend	Cat # 305602, RRID: AB_314540
Mouse monoclonal anti-human CD38 (clone HIT2)	Fluidigm	Cat # 3172007B; RRID: AB_2756288
Mouse monoclonal anti-human GZMB (clone GB11)	Fluidigm	Cat # 3173006B; RRID: AB_2811095
Mouse monoclonal anti-human HLA-DR (clone L243)	Fluidigm	Cat # 3174001; RRID: AB_2665397
Mouse monoclonal anti-human CD161 (clone DX12)	BD Biosciences	Cat # 556079, RRID: AB_396346
Mouse monoclonal anti-human CD56 (clone N901)	Fluidigm	Cat # 3176009B; RRID: AB_2811096
Critical commercial assays
Vacutainer EDTA tubes	BD	Cat # 36643
PAXgene Blood RNA tubes	Qiagen	Cat # 762165
PAXgene Blood RNA Kit	Qiagen	Cat # 762164
Allprep DNA/RNA/miRNA Universal Kit	Qiagen	Cat # 80224
GlobinClear kit	ThermoFisher Scientific	Cat # AM1980
NEBNext Poly(A) mRNA Magnetic Isolation Module	New England Biolabs	Cat #E7490
NEBNext Ultra II Directional RNA Library Prep Kit for Illumina	New England Biolabs	Cat #E7760;
V-PLEX Human Biomarker 54-Plex	MesoScale Discovery	Cat #K15248D
Transcription Factor Phospho Buffer Set	BD Pharmingen	Cat # 563239
10X PBS, pH 7.2	Rockland	Cat # MB-008
Cell Staining Buffer	Fluidigm	Cat # 201068
Cell-IDTM 20- Plex Pd Barcoding Kit	Fluidigm	Cat # PRD023
Cell-ID Intercalator-Ir	Fluidigm	Cat # 201192A
Maxpar Antibody Labeling Kit	Fluidigm	Cat # 201160B
Deposited data
HTP Sample/Participant metadata and co-occurring conditions	Galbraith et al.27	Synapse:https://doi.org/10.7303/syn31488784
HTP whole blood RNA-seq data	Galbraith et al.27	GEO: GSE190125, Synapse:https://doi.org/10.7303/syn31488780
HTP plasma proteomics data	Galbraith et al.27	Synapse:https://doi.org/10.7303/syn31488781
HTP mass cytometry data	Galbraith et al.27	Flow Repository: FR-FCM-Z5GE, Synapse:https://doi.org/10.7303/syn31488783
HTP white blood cell RNA-seq data	Powers et al.47	GEO: GSE251967
Mouse adult heart RNA-seq data	Waugh et al.41	GEO: GSE218883
Mouse embryonic (E12.5) heart RNA-seq data	Waugh et al.41	GEO: GSE218888
Mouse embyronic (E18.5) heart RNAseq data	Waugh et al.41	GEO: GSE218889
Mouse adult whole brain RNA-seq data	Waugh et al.41	GEO: GSE218885
Mouse adult brain hippocampus RNA-seq data	This study	GEO: GSE272567
Mouse adult brain cortex RNA-seq data	This study	GEO: GSE272568
Mouse adult kidney RNA-seq data	This study	GEO: GSE272569
Mouse adult liver RNA-seq data	This study	GEO: GSE272570
Mouse adult spleen RNA-seq data	This study	GEO: GSE272572
Mouse embryonic (E10.5) facial mesenchyme RNA-seq data	Waugh et al.41	GEO: GSE218887
Mouse adult mesenteric lymph node RNA-seq data	Waugh et al.41	GEO: GSE218890
Mouse adult lung RNA-seq data	This study	GEO: GSE229762
Software and algorithms
R	R Foundation for Statistical Computing	v4.3.1; RRID:SCR_001905
R Studio	R Studio, Inc.	v2023.09.1 + 494; RRID:SCR_000432
Bioconductor	Bioconductor	v3.17; RRID:SCR_006442
Tidyverse collection of packages for R	CRAN	RRID:SCR_019186
limma package for R	Bioconductor	v3.56.2; RRID:SCR_010943
FASTQC	Babraham Institute	v0.11.5; RRID:SCR_014583,https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
FastQ Screen	Babraham Institute	v0.11.0; RRID:SCR_000141, https://www.bioinformatics.babraham.ac.uk/projects/fastq_screen/
bbduk/BBTools	Bushnell et al.71	v37.99; RRID:SCR_016968,https://jgi.doe.gov/data-and-tools/bbtools/
fastq-mcf/ea-utils	N/A	v1.05; RRID:SCR_005553,https://expressionanalysis.github.io/ea-utils/
HISAT2	Kim et al.72	v2.1.0; RRID:SCR_015530,https://daehwankimlab.github.io/hisat2/
Human genome sequence primary assembly fasta	Gencode	GRCh38; RRID: SCR_014966,https://www.gencodegenes.org/human/release_33.html
Human genome basic annotation GTF file	Gencode	v33; RRID:SCR_014966,https://www.gencodegenes.org/human/release_33.html
Samtools	N/A	v1.5; RRID: SCR_002105, http://www.htslib.org/
HTSeq-count	N/A	v0.6.1; RRID:SCR_005514,https://htseq.readthedocs.io/en/master/
DESeq2 package for R	Bioconductor	v1.28.1; RRID:SCR_015687
fgsea package for R	Bioconductor	v1.26.0; RRID:SCR_020938
Hmisc package for R	CRAN	v5.1.1; RRID: SCR_022497
ggplot2 package for R	CRAN	v3.4.4; RRID:SCR_014601
rstatix package for R	CRAN	v0.7.2; RRID: SCR_021240
ComplexHeatmap package for R	CRAN	v2.4.2; RRID:SCR_017270
tidyheatmap package for R	CRAN	v1.8.1
ggforce package for R	CRAN	v0.4.1
SomaDataIO package for R	GitHub, SomaLogic, Inc.	v3.1.0; RRID: SCR_022198https://github.com/SomaLogic/SomaDataIO
Normalizer	GitHub, Finck et al.77	https://github.com/nolanlab/bead-normalization
Single Cell Debarcoder	GitHub, Zunder et al.78	https://github.com/zunderlab/single-cell-debarcoder/wiki/Installing-the-Single-Cell-Debarcoder
CyTOF Batch Adjust	GitHub, Schuyler et al.79	https://github.com/CUHIMSR/CytofBatchAdjust
CellEngine	CellCarta, Montreal, Canada	RRID: SCR_022484, https://docs.cellengine.com/
flowCore package for R	Hahne et al.80; Bioconductor	v2.0.1; RRID: SCR_002205
CATALYST package for R	Chevrier et al.81; Bioconductor	v1.12.2; RRID: SCR_017127,https://github.com/HelenaLC/CATALYST
FlowSOM package for R	GitHub, Van Gassen et al.39	v1.20.0; RRID: SCR_016899https://github.com/SofieVG/FlowSOM
ConsensusClusterPlus package for R	Bioconductor; Wilkerson and Hayes83	v1.52.0; RRID: SCR_016954
tidySingleCellExperiment package for R	Bioconductor	v1.3.3; RRID: SCR_022493,https://github.com/stemangiola/tidySingleCellExperiment
MEM package for R	GitHub, Diggins et al.84	v3; RRID: SCR_022495,https://github.com/JonathanIrish/MEMv3
Betareg package for R	CRAN; Cribari-Neto and Zeileis87	v3.1-4; RRID: SCR_022494
ggeffects package for R	CRAN; Lüdecke85	v1.1.0; RRID: SCR_022496
cluster package for R	CRAN	v2.1.0
janitor package for R	CRAN	v2.0.1

Mouse models of Down syndrome

Experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Colorado Anschutz Medical Campus under Protocol #00111 and performed in accordance with National Institutes of Health (NIH) guidelines. Dp10 (B6; 129S7-Dp(10Prmt2-Pdxk)2Yey/J), Dp16 (B6.129S7-Dp(16Lipi-Zbtb21)1Yey/J), and Dp17 (B6; 129S7-Dp(17Abcg1-Rrp1b) 3Yey/J) mice were originally purchased from The Jackson Laboratory and gifted by Dr. Diana Bianchi’s laboratory (NIH) then intermixed and maintained on the C57BL/6J background. For generation of the Dp16^2xIfnrs mice,⁴¹ female WT^1xIfnrs were crossed with male Dp16. To power the comparison arms representing each genotype in this study, multiple litters were combined, where each litter contributed randomly to the sum of each cohort, thus minimizing the impact of any potential shift in genetic background. Mice were housed separately by sex in groups of 1–5 mice/cage under a 14:10 light:dark cycle with controlled temperature and 35% humidity with ad libitum access to food (6% fat diet) and water. All mice were at least 87.5% C57BL/6J with the remaining background inferred as C57BL/6NTac (Taconic). Ages of the mice employed varied for each experiment as indicated in each figure from embryonic stage E12.5 to adulthood.

METHOD DETAILS

Blood sample collection and processing

Peripheral blood was collected using PAXgene RNA tubes (Qiagen) BD Vacutainer K2 EDTA tubes (BD). PAXgene RNA tubes containing whole blood were frozen within 4 h of collection and are stable at −20°C for up to 11 years. For EDTA samples, two 0.5 mL aliquots were removed and processed for mass cytometry as described below. One 200 μL was aliquoted and used for complete blood counts (CBC) as described below. The remainder of EDTA whole blood was centrifuged at 700 × g for 15 min in order to separate plasma, buffy coat containing white blood cells (WBC), and red blood cells (RBCs), which were then aliquoted, flash frozen and stored at −80°C until subsequent processing and analysis. Plasma was centrifuged an additional time at 2200 × g for 15 min, 4°C, then aliquoted and stored at −80°C until subsequent processing and analysis. The buffy coat containing white blood cells (WBC) was further processed into WBCs, aliquoted in freezing media, and cryogenically stored until subsequent processing and analysis. Processing of EDTA samples took place within 2 h of collection. As described below, these samples were further processed and employed for specific assays from the exact same sample to enable cross-omic analyses with mostly overlapping datasets. Due to a variety of factors (sample quality/quantity, etc.), not all blood samples were used for all assays, limiting our sample size for cross-omic comparisons.

Whole Blood RNA sequencing

Whole blood RNA was purified from blood samples collected in PAXgene RNA tubes using the PAXgene Blood RNA Kit (PreAnalytiX). RNA was purified from WBC isolated from whole blood collected in EDTA extracts using the Allprep DNA/RNA/miRNA Universal Kit (Qiagen). The quality of purified RNA was assessed using a 2200 TapeStation system (Agilent) and concentration was determined using a Qubit fluorometer (Invitrogen). Globin depletion, Poly-A(+) RNA enrichment, and strand-specific library preparation was carried out using the GlobinClear kit (ThermoFisher Scientific), NEBNext Poly(A) mRNA Magnetic Isolation Module, and NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs), respectively. Paired-end sequencing with 150 bp read length was carried out by Novogene Co., Ltd on an Illumina NovaSeq 6000 instrument and data was provided in FASTQ format.

Plasma SomaScan proteomics

Profiling of the plasma proteome in HTP participants was carried out by SomaLogic using the SomaScan assay.⁶⁶ This assay relies on SOMAmer reagents, which are short hydrophobic single-stranded DNA sequences synthesized with fluorophores and bound to biotin via photocleavable linkers. Briefly, SOMAmer reagents are immobilized on streptavidin beads and incubated with 125 μL of EDTA plasma. After incubation, unbound proteins are washed, and remaining complexes are biotinylated. UV light is then used to break the photocleavable linkers and release the complexes into solution, which are then treated with a polyanionic competitor to enhance binding specificity. Complexes are then re-captured to new streptavidin beads. After washing, proteins are denatured to release SOMAmers. SOMAmers are then hybridized to complementary sequences on microarrays and quantified by fluorescence intensity of their corresponding fluorophores.

Complete blood counts (CBC)

Complete blood counts (CBC) for HTP participants were determined from whole blood samples collected in EDTA tubes. Prior to analysis, the 200 μL aliquot (described above) was thoroughly mixed to ensure a homogeneous sample. A sample volume of 12 μL was then removed and analyzed using an AcT 10 hematology analyzer (Beckman Coulter). Results are reported as white blood cell count (WBC, 10³/μL), red blood cell count (RBC, 10³/μL), hemoglobin (Hgb, g/dL), hematocrit (Hct, %), mean corpuscular volume (MCV, fL), mean corpuscular hemoglobin (MCH, pg), mean corpuscular hemoglobin concentration (MCHC, g/dL), platelet count (Plt, 10³/μL), absolute lymphocyte count (10³/μL) and percentage of lymphocytes (%). For murine CBC data, blood was collected in EDTA or lithium heparin tubes and analyzed on Heska Hematrue instrument to produce a 3-part leukocyte differential, along with erythrocyte and platelet counts and indices.

White blood cell mass cytometry (CyTOF)

For mass cytometry, two 0.5 mL aliquots of EDTA whole blood samples underwent red blood cell lysis and white blood cell fixation using Lyse/Fix Buffer (BD Phosflow Lyse/Fix Buffer 5X, BD Biosciences). White blood cells were then washed 1 × in PBS (Rockland), resuspended in Cell Staining Buffer (Fluidigm) and stored at −80°C. For antibody staining, samples were thawed at room temperature, washed in Cell Staining Buffer, barcoded using a Cell-ID 20-Plex Pd Barcoding Kit (Fluidigm), and combined per batch. Each batch was able to accommodate 19 samples with a common reference sample. Antibodies were either purchased pre-conjugated to metal isotopes or conjugation was performed in-house using a Maxpar Antibody Labeling Kit (Fluidigm). See key resources table and Table S5 for antibody details. Working dilutions for antibody staining were titrated and validated using the common reference sample and comparison to relative frequencies obtained by independent flow cytometry analysis. Surface marker staining was carried out for 30 min at 4°C in Cell Staining Buffer with added Fc Receptor Binding Inhibitor (eBioscience/ThermoFisher Scientific). Staining was followed by a wash in Cell Staining Buffer. Next, cells were permeabilized in Buffer III (Transcription Factor Phospho Buffer Set, BD Pharmingen) for 20 min at 4°C followed by washing with perm/wash buffer (Transcription Factor Phospho Buffer Set, BD Pharmingen). Intracellular transcription factor and phospho-epitope staining was carried out for 1 h at 4°C in perm/wash buffer (Transcription Factor Phospho Buffer Set, BD Pharmingen), followed by a wash with Cell Staining Buffer. Cell-ID Intercalator-Ir (Fluidigm) was used to label barcoded and stained cells. Labeled cells were analyzed on a Helios instrument (Fluidigm). Mass cytometry data were exported as v3.0 FCS files for pre-processing and analysis.

Mouse multi-tissue RNA sequencing

RNA was purified using TRIzol reagent (Invitrogen) following the manufacturer’s instructions. Whole hearts from E12.5 and E18.5 mice were homogenized in Lysing Matrix D tubes (MP Biomedicals) with 500 μL of TRIzol reagent (Invitrogen) for 30 s using a Mini-Beadbeater-24 (BioSpec Products). For E10.5 facial mesenchyme, a water bath at 37°C was used for quick freeze/thaw, and total RNA was isolated using the QIAshreder (QIAGEN) and AllPrep DNA/RNA/Protein Mini Kit (QIAGEN). For adult tissues, mice were killed by CO₂ asphyxiation and cervical dislocation then immediately perfused with 1 × PBS using a Perfusion Two Automated Perfusion Instrument (Leica). Whole brain samples consist of the entire right brain hemisphere from 6- to 9-month-old mice. Mesenteric lymph nodes, spleen, and heart were similarly removed from perfused 4- to 5-month-old mice. The 3–5 mesenteric lymph nodes per animal were flash-frozen together while brain and adult heart samples were first placed in Lysing Matrix D tubes (MP Biomedicals) containing 594 μL of lysis buffer RLT Plus (QIAGEN) and 6 μL of 2-mercaptoethanol (Sigma-Aldrich) before storage at −80°C. Upon quick freeze/thaw at 37°C, mesenteric lymph nodes were placed in Lysing Matrix D tubes also containing RLT with 2-ME. Adult tissues were then homogenized for 30 s using a Mini-Beadbeater-24 (BioSpec Products) and then frozen at −80°C. Total RNA was isolated using the AllPrep Kit (QIAGEN) following the manufacturer’s instructions. Library preparation was carried out using a Universal Plus mRNA Kit Poly(A) (Tecan). Paired-end, 150 bp sequencing was carried out on an Illumina NovaSeq 6000. Subsequent analysis was carried out as for human whole-blood RNA-seq, except alignment and gene-level count summarization used a mouse GRCm38 reference genome index and Gencode M24 annotation GTF.

QUANTIFICATION AND STATISTICAL ANALYSIS

Data analysis and visualization

Pre-processing, statistical analysis, and visualization off datasets was carried out using R (R 4.0+/RStudio 2022.12.0+/Bioconductor 3.16+)^67,68 as detailed below. Comparisons of distributions between categorical groups were displayed with sina plots using ggplot2 and the geom_sina() function from the ggforce R package.^69,70 Points were jittered horizontally by local density and overlayed with boxplots representing medians and interquartile ranges. Comparisons for continuous data were displayed with scatterplots generated using ggplot2.⁷⁰ Points were colored by local density using a custom function. Heatmaps were generated using the ComplexHeatmap, tidyheatmap, and CosensusClusterPlus R packages.

RNA-seq differential expression

For human whole blood RNA-seq, data yield was ~33–103 × 10⁶ raw reads and ~21–69 × 10⁶ final mapped reads per sample. Data quality was assessed using FASTQC (v0.11.5) and FastQ Screen (v0.11.0). Low-quality reads were trimmed and filtered using bbduk from BBTools (v37.99)⁷¹ and fastq-mcf from ea-utils (v1.05). Alignment to the human reference genome (GRCh38) was carried out using HISAT2 (v2.1.0)⁷² in paired, spliced-alignment mode against a GRCh38 index and Gencode v33 basic annotation GTF, and alignments were sorted and filtered for mapping quality (MAPQ >10) using Samtools (v1.5).⁷³ Gene-level count data were quantified using HTSeq-count (v0.6.1)⁷⁴ with the following options (–stranded = reverse –minaqual = 10 –type = exon –mode = intersection-nonempty) using a Gencode v33 GTF annotation file. Differential gene expression in T21 versus D21 was evaluated using DESeq2 (version 1.28.1)⁷⁵ with source, sex and age used as covariates. Benjamini-Hochberg adjustment of p values was used to adjust for multiple hypotheses, with q < 0.1 (10% FDR) as the threshold for differential expression. Prior to visualization, RPKMs were adjusted for age, sex, and sample source using the removeBatchEffect() function from the limma package (v3.44.3).⁷⁶

For mouse RNA-seq, data yield was a minimum of ~30 million raw reads. Differential gene expression was evaluated using DESeq2 v1.28.1 with sex and batch as covariates, and q < 0.1 as the threshold for DEGs. GSEA was carried out in R v4.0.1 as described above. Number of animals per group were as follows: E10.5 facial mesenchyme WT (n = 3, 2 male and 1 female), Dp16 (n = 3, 2 male and 1 female) and Dp16^2xIfnrs (n = 3, 2 male and 1 female); E12.5 hearts WT (n = 6, 2 male and 4 female), Dp16 (n = 6, 3 male and 3 female) and Dp16^2xIfnrs (n = 6, 5 male and 1 female); E18.5 hearts WT (n = 6, 4 male and 2 female), Dp16 (n = 6, 3 male and 3 female) and Dp16^2xIfnrs (n = 6, 3 male, 3 female); adult mesenteric lymph nodes WT (n = 5, 2 male, 3 female), Dp16 (n = 6, 3 male and 3 female) and Dp16^2xIfnrs (n = 6, 3 male and 3 female); adult brains WT (n = 6, 2 male and 4 female), Dp16 (n = 5, 2 male and 3 female) and Dp16^2xIfnrs (n = 7, 4 male and 3 female); adult hearts WT (n = 5, 2 male and 3 female), Dp16 (n = 6, 3 male and 3 female) and Dp16^2xIfnrs (n = 5, 3 male and 2 female); spleen WT (n = 6, 3 male and 3 female), Dp16 (n = 6, 3 male and 3 female) and Dp16^2xIfnrs (n = 6, 3 male and 3 female).

SomaScan differential abundance

Normalized data (RFU) in the SomaScan adat file format was imported to R using the SomaDataIO R package (v3.1.0). Extreme outliers were classified per-karyotype and per-analyte as values more than three times the interquartile range below or above the first and third quartiles, respectively (below Q1 − 3*IQR or above Q3 + 3*IQR) and excluded from further analysis. Differential abundance analysis was performed using linear regression in R with log₂-transformed relative abundance as the outcome/dependent variable, T21 status as the predictor/independent variable, and age, sex, and sample source as covariates. Multiple hypothesis correction was performed with the Benjamini-Hochberg method using a false discovery rate (FDR) threshold of 10% (q < 0.1). Prior to visualization or correlation analysis, SomaScan data were adjusted for age, sex, and sample source using the removeBatchEffect() function from the limma package (v3.44.3).⁷⁶

CBC differential abundance

CBC results from the AcT 10 hematology analyzer were read analyzed using R/Rstudio. Extreme outliers were classified per-karyotype and per-analyte as values more than three times the interquartile range below or above the first and third quartiles, respectively (below Q1 − 3*IQR or above Q3 + 3*IQR) and excluded from further analysis. Linear regression was performed on a per-analyte basis with the log2-transformed units as the outcome/dependent variable, T21 status as the predictor/independent variable, and age, sex, and sample source as covariates. Multiple hypothesis correction was performed with the Benjamini-Hochberg method using a false discovery rate (FDR) threshold of 10% (q < 0.1). Prior to visualization or correlation analysis, SomaScan data were adjusted for age, sex, and sample source using the removeBatchEffect() function from the limma package (v3.44.3).⁷⁶

CyTOF differential abundance

For pre-processing, bead-based normalization via polystyrene beads embedded with lanthanides, both within and between batches, followed by bead removal was carried out using the Matlab-based Normalizer tool.⁷⁷ Batched FCS files were demultiplexed using the Matlab-based Single Cell Debarcoder tool.⁷⁸ Reference-based normalization of individual samples across batches against the common reference sample was then carried out using the R script BatchAdjust() from the CyTOF Batch Adjust package.⁷⁹ For the analyses described in this manuscript, CellEngine (CellCarta) was used to gate and export per-sample FCS files at four levels: Firstly, CD3⁺CD19⁺ doublets were excluded, and remaining cells exported as ‘Live’ cells; Live cells were then gated for hematopoietic lineage (CD45-positive) non-granulocytic (CD66-low) cells and exported as CD45⁺CD66low. Lastly, CD45⁺CD66low cells were gated on CD3-positivity and CD19-positivity and exported as T- and B-cells, respectively. Per-sample FCS files were then subsampled to a maximum of 50,000 events per file for subsequent analysis.

For each of the four levels (live, non-granulocytes, T cells, and B cells), all 388 per-sample FCS files were imported into R as a flowSet object using the read.flowSet() function from the flowCore R package.⁸⁰ Next a SingleCellExperiment object was constructed from the flowSet object using the prepData() function from the CATALYST package.⁸¹ Arcsinh transformation was applied to marker expression data with cofactor values ranging from ~0.2 to ~15 to give optimal separation of positive and negative populations for each marker, using the estParamFlowVS() function from the flowVS R package⁸² and based on visual inspection of marker histograms (see Table S5). Quality control and diagnostic plots were examined with the help of functions from CATALYST and the tidySingleCellExperiment R package. Unsupervised clustering using the FlowSOM algorithm³⁹ was carried out using the cluster() function from CATALYST, with grid size set to 10 × 10 to give 100 initial clusters and a maxK value of 40 was explored for subsequent meta-clustering using the ConsensusClusterPlus algorithm.⁸³ Examination of delta area and minimal spanning tree plots indicated that 30–40 meta clusters gave a reasonable compromise between gains in cluster stability and number of clusters for each level. Each clustering level was re-run with multiple random seed values to ensure consistent results.

To aid in assignment of clusters to specific lineages and cell types, the MEM package (marker enrichment modeling)⁸⁴ was used to call positive and negative markers for each cell cluster based on marker expression distributions across clusters. Manual review and comparison to marker expression histograms, as well as minimal spanning tree plots and tSNE plots colored by marker expression, allowed for high-confidence assignment of most clusters to specific cell types. Clusters that were insufficiently distinguishable were merged into their nearest cluster based on the minimal spanning tree. Relative frequencies for each cell type/cluster were calculated for each sample as a percentage of total live cells and as a percentage of cells used for each level of clustering: total CD45⁺CD66low cells, total T cells, or total B cells.

To identify cell clusters for which relative frequencies are associated with either trisomy 21 status or polygenic HIF1A scores among individuals with trisomy 21, beta regression analysis was carried out using the betareg R package,⁷⁴ with each model using cell type cluster proportions (relative frequency) as the outcome/dependent variable and either T21 or HIF1A score as independent/predictor variables, along with adjustment for age and sex, and a logit link function. Extreme outliers were classified per-karyotype and per-cluster as measurements more than three times the interquartile range below or above the first and third quartiles, respectively (below Q1 − 3*IQR or above Q3 + 3*IQR) and excluded from beta regression analysis. Correction for multiple comparisons was performed using the Benjamini-Hochberg (FDR) approach. Effect sizes (as fold-change in T21 vs. euploid controls) for each cell type cluster were obtained by exponentiation of beta regression model coefficients. Fold-changes were visualized by overlaying on tSNE plots using ggplot2. For visualization of individual clusters, data points were adjusted for age and sex, using the adjust() function from the datawizard R package. Individual examples were as sina plots (separated by T21 status) or as XY scatterplots (for comparison to HIF1A scores), by local density using a custom function and overlaid with beta regression fit curves and 95% confidence intervals extracted from model objects using the ggemmeans() function from the ggeffects package (v1.1.0).⁸⁵

Ingenuity pathway analysis (IPA)

Results from DESeq2 analysis of whole-blood transcriptomes (RNAseq) were analyzed using QIAGEN IPA software (QIAGEN Inc., https://digitalinsights.qiagen.com/IPA).⁸⁶ The reference set was specified to the user dataset and the analysis only considered direct relationships. Cutoffs for both expression fold change (−1.5/1.5) and q-value (0.1), were employed. Results were exported and visualized using R and RStudio.

Polygenic heme and HIF1A scores

Polygenic scores representative of the Hallmark Heme Metabolism and HIF1A upstream regulator network transcriptome signatures were derived as follows. Genes contributing to the Heme score were selected from the leading-edge genes of the Hallmark Heme Metabolism signature enriched in individuals with T21 and previously published.²⁷ Genes in the HIF1A score were selected from genes contributing to IPA-based predicted activation of the HIF1A upstream regulator network. For both scores, genes were filtered to those overexpressed greater than 1.5× in individuals with T21 relative to euploid controls. Chr21 genes were also removed. Using RPKM, we calculated for each gene z scores using the mean and standard deviation from euploid controls. The z scores were then summed to derive the Heme and HIF1A composite scores per individual. Differences between individuals with T21 and euploid controls were determined by linear regression T21 status as the independent variable, scores as dependent variable and age, sex and sample source as covariates. Differences between sexes used sex as the independent variable with age and sample source as covariates. Associations with age and BMI used these variables as independent variables with sex and sample source as covariates. The Benjamini-Hochberg method was used for multiple hypothesis correction using a false discovery rate of 10% (q-value <0.1). For visualization and downstream correlation analysis, scores were derived from RPKMs data adjusted for Source; Source and Sex; Source and Age; and Source, Sex and Age.

Spearman correlations

Spearman correlation coefficients (rho values) and p-values were calculated between both scores and all proteins in the SomaScan plasma proteome dataset using the rcorr() function from the Hmisc R package (v5.1.1). Resulting rho and p-value matrices were subset to correlations for the Heme score, HIF1A score and EPO (a SomaScan protein) against all proteins from the SomaScan dataset. Multiple hypothesis correction was carried out separately for each set of correlations using the Benjamini-Hochberg adjustment with a false discovery rate of 10% (q-value <0.1).

Highlights.

• Multi-omic analysis indicates elevated hypoxic signaling and heme metabolism in Down syndrome

• These signatures track with hallmarks of stress erythropoiesis and immune remodeling

• The Dp16 mouse model of Down syndrome recapitulates these molecular phenotypes

• These events occur independently of hyperactive interferon and JAK/STAT signaling