Environmental enteric dysfunction (EED) is a paediatric disorder characterized by chronic intestinal inflammation that is associated with malnutrition, stunted growth, cognitive impairment and attenuated response to oral vaccines1,2,3,4,5. Previously described as ‘tropical enteropathy’ or ‘environmental enteropathy’, EED has gained renewed interest in recent years due to its devastating effect on millions of children in low- and middle-income countries. The intestine of EED patients commonly exhibits villous atrophy, nutrient malabsorption, barrier dysfunction and inflammation6,7. As there is currently no existing in vitro model of EED and only a limited number of animal models, mechanistic understanding of this disease is limited, which has hampered efforts to define biomarkers for diagnosis or develop therapeutics8. For example, deficiencies of micronutrients, such as zinc9 and vitamin A10, may contribute to EED pathophysiology as they are associated with abnormal lactulose:mannitol (L:M) ratios—a measure of intestinal permeability. However, attempts to treat EED-related stunting by nutritional interventions have been disappointing probably due to ongoing problems with nutrient absorption and inflammation in these patients8. Similarly, other dietary interventions, such as administering omega-3 long-chain polyunsaturated fatty acids11, optimizing amino acid profiles12, supplementing with multiple micronutrients13 or improving food digestibility through fermentation, hydrolysis or enzyme supplementation12, have also been tested with limited success.

Inadequate dietary intake not only leads to stunting but can also increase host vulnerability to environmental factors. For example, diets low in the essential amino acid tryptophan lead to decreased antimicrobial peptide secretion and increased susceptibility to chemical-induced intestinal inflammation in mice14. Low serum tryptophan levels are also linked to stunting in children suffering from EED15,16. Tryptophan is both a building block for proteins and a precursor for niacin, melatonin and neurotransmitters (such as serotonin and tryptamine)17, and supplementation with amino acids including tryptophan was recently shown to improve symptoms in EED patients18. In animal studies, symptoms of tryptophan deficiency, including anorexia and impaired growth, may occur with intakes as little as 25% below the standard requirement, which translates to 2–2.5 mg kg−1 body weight for human infants 6–24 months old17,19. Recently, niacin (nicotinic acid) deficiency has also been implicated as a contributor to EED18 and other inflammatory intestinal conditions as administration of niacin was shown to ameliorate dextran sodium sulfate-induced colitis via prostaglandin D2‐mediated D prostanoid receptor 1 activation20. In addition, niacin serves as a precursor for coenzymes, such as nicotinamide adenine dinucleotide (NAD) and nicotinamide adenine dinucleotide phosphate (NADP), which are essential for the normal function and survival of living cells. However, a mechanistic role for malnutrition in driving EED pathophysiology in humans remains to be demonstrated.

Studying a multifactorial disease such as EED raises substantial methodical and modelling challenges and, at present, there are only a few murine models and no human in vitro models that can be used to study this disease21. Thus, establishing an in vitro human EED model would help to elucidate disease pathophysiology and enable the development of new prevention and therapeutic measures. Here we describe how human organ-on-a-chip microfluidic culture technology that faithfully recapitulates the structure and function of many human organs, including the intestine22,23,24,25,26,27,28,29,30,31, can be leveraged to meet this challenge. We used human intestine chips lined with organoid-derived primary intestinal epithelium isolated from either healthy children or paediatric EED patients who were refractory to nutritional intervention. We have previously shown that these intestine chips support the formation of differentiated three-dimensional villus-like epithelial structures as well as the production of an overlying mucus layer, which require the presence of dynamic fluid flow28,32,33,34. Our studies comparing healthy versus EED intestine chips revealed that both nutritional deficiencies and genetic or epigenetic changes in the intestinal epithelium contribute to the clinically observed EED phenotype. Moreover, by comparing healthy and EED patient-derived intestine chips, we were able to study phenotypic responses to nutritional deficiencies, such as villus blunting and barrier dysfunction, which are known to be common to multiple intestinal pathologies (such as inflammatory bowel diseases and coeliac disease), and distinguish them from responses due to transcriptomic and cytokine signatures that are seen in EED.


Nutritionally deficient EED chips recapitulate EED patient transcriptional signatures

We previously described a two-channel microfluidic human intestine chip, lined with living human intestinal epithelium isolated from patient-derived organoids, that undergoes villus differentiation, accumulates mucus and exhibits many features of living human intestine when cultured on-chip under continuous flow with peristalsis-like mechanical deformations28 (Fig. 1a). Additionally, transcriptional analysis demonstrated that when lined by organoid-derived duodenal epithelium, this intestine chip more closely mimicked in vivo human duodenum than the organoids used to create the chips28. To define the contribution of the intestinal epithelium to the EED phenotype, we created intestine chips lined with intestinal epithelial cells from organoids derived from surgical biopsies of either healthy or EED patient duodenum (healthy chips and EED chips, respectively). Compared with healthy chips, EED chips showed differential expression of 287 genes (q < 0.05 and fold change ≥1.5, q = FDR-adjusted-p-value; 86 upregulated, 201 downregulated) (Fig. 1b). EED chips showed upregulation of MUC5AC (previously shown to reduce inflammation, intestinal injury and bacterial translocation in an experimental intestinal injury35), neuregulin-4 (a known survival factor for colonic epithelium that protects against experimental intestinal injury36,37) and the intestinal stem cell marker SMOC2, whereas brush border peptidase MME, oxidative stress and inflammatory response controlling ectoenzyme VNN1, tight junction protein CLDN10 and secreted goblet cell protein CLCA1 were all downregulated (Supplementary Table 1).

Fig. 1: Nutritionally deficient EED chips recapitulate EED patient transcriptional signatures.
figure 1

a, A schematic representation of small-intestine chips. b, Compared with healthy chips, the EED transcriptome consists of 287 differentially expressed genes (red; q < 0.05 and fold change ≥1.5). Exposure of healthy and EED chips to -N/-T media resulted in an increased number of differentially expressed genes with 690 (yellow) and 969 genes (blue), respectively. Of these, 307 genes were differentially expressed in both healthy and EED chips exposed to nutritional deficiency (yellow and blue). Seventy-one genes were differentially expressed in EED chips when compared with healthy chips in control medium that are further affected by the addition of nutritionally deficient medium (blue and red, respectively; 3 chips for each condition). Each chip corresponds to one biological replicate. c, A comparison of the 20 most up- and 20 most downregulated genes from the clinical EED signature, with healthy or EED chip gene expression depicted as a heat map (red, upregulation; blue, downregulation) showing that EED -N/-T has the closest hierarchical relationship to the clinical EED signature (clinical EED Z-score distance from intestine chips results was 15.2 for the EED -N/-T vs EED; 18.7 for the healthy -N/-T vs healthy control; and 19.4 for EED control vs healthy control). The comparison for EED -N/-T to healthy control is similar and shown in Extended Data Fig. 1. Ten dendrogram gene clusters were also defined and the corresponding roots enumerated; 3 chips for each condition. Each chip corresponds to one biological replicate. d, Of the top 9 upregulated genes in the clinical EED signature, 6 were also upregulated when EED chips were exposed to -N/-T media; 3 chips for each condition. Each chip corresponds to one biological replicate. e, Functional pathway analysis was performed using the contextual language-processing programme COMPBIO. The spatial map depicts themes related to differentially expressed genes as nodes, with interconnections depicted as edges whose thickness relates to the degree of interconnectedness. The themes were also ranked according to a score representing fold enrichment over random clustering and this score determines sphere size (Supplementary Table 2); 3 chips for each condition. Each chip corresponds to one biological replicate.

We compared this differential gene expression profile with a recently derived clinical EED signature, which was obtained by comparing profiles of intestinal tissue samples from EED patients who were also refractory to nutritional intervention (Study of Environmental Enteropathy and Malnutrition, SEEM)38 versus samples from healthy control patients who were investigated for gastrointestinal symptoms but had normal endoscopic and histologic findings (Cincinnati Children’s Hospital Medical Center). When we compared gene profiles from EED chips versus healthy chips cultured in control medium (that is, with all nutrients present), the differentially expressed genes had some overlap with the clinical EED signature, including most notably a shared downregulation of metallothioneins (MT1X, MT1A, MT1F, MT1H, MSMB and MT1M; gene dendrogram cluster 5) (Fig. 1c).

We then carried out the same experiment but perfused both the healthy and EED intestine chips with medium deficient in niacinamide and tryptophan (−N/−T), selected on the basis of past work implicating their role in EED15,16,19. When we compared expression profiles from the healthy chips exposed to nutritional deficiency (healthy −N/−T chip) versus the healthy control chips, we detected differential expression of 690 genes (q < 0.05 and fold change ≥1.5; 556 upregulated, 124 downregulated) (Fig. 1b and Supplementary Fig. 1), including upregulation of the amino acid starvation-related transcription factor ATF4, its downstream solute carriers (SLC34A2, SLC7A5 and SLC6A9) and the inflammation-associated gene LCN2 (Supplementary Table 1). Uniformity of transcriptome analysis of samples in each group was confirmed by principal component analysis (Supplementary Fig. 2). There also appeared to be a trend towards upregulation of several antimicrobial and immune response genes as seen in the clinical EED signature, but these changes did not reach statistical significance. While there was greater overlap between the transcriptome of the healthy −N/−T chip with the clinical EED transcriptome than observed with the EED control chip compared with healthy control chip, some genes were regulated in an opposing direction, including upregulation of metallothioneins (Fig. 1c, gene dendrogram cluster 5).

In contrast, we observed closer unsupervised hierarchical clustering with the clinical EED signature when the EED chip was exposed to nutritional deficiency (EED −N/−T chip vs EED control chip) (Fig. 1c). Culture of the EED intestine chips in −N/−T medium yielded differential expression of 969 genes (q < 0.05 and fold change ≥1.5; 522 upregulated, 447 downregulated) (Fig. 1b). This was manifested by the upregulation of antimicrobial genes (SAA1, SAA2, DUOXA2, DUOX2 and CXCL5; gene dendrogram clusters 1, 3 and 4) and downregulation of not only metallothioneins but also metabolic and digestive genes (SLC6A14 and GUC2AB; gene dendrogram clusters 4 and 9). This congruence was most noticeable among the top 10 upregulated genes of the clinical EED signature, 6 of which (60%) were also upregulated when EED chips were exposed to nutritionally deficient medium (Fig. 1d).

To identify pathways affected by exposure of EED intestine chips to −N/−T nutritional deficiency, we used a contextual language-processing programme to identify and rank functionally related clusters of genes39. This analysis revealed several pathways that were significantly upregulated when EED chips were exposed to −N/−T media, including a chemokine pathway (score 1,833.89, indicates fold enrichment over random association) and a pathway associated with amino acid starvation (score 1,988.09) (Fig. 1e and Supplementary Table 2). Within the amino acid starvation pathway, the ATF4 gene is upstream of several other pathways including tRNA metabolism (score 1,547.01), DNA damage (score 1,790.28), p53 methylation (score 1,608.44) and amino acid transporters (score 2,022.97). Many of these same genes were also upregulated in nutritionally deficient healthy −N/−T chips, including ATF4, TP53, AARS, YARS, MDM2, CCND2 and several amino acid transporters (Supplementary Table 1). Conversely, nutritional deficiency led to the downregulation of pathways related to fatty acid uptake (score 3,590.64), brush border structural integrity (score 3,777.69), mitosis (score 3,311.49), cytochrome p450 (score 2,581.54) and fatty acid utilization (score 2,056.86) in the EED chips. These results are consistent with the observation that the intestines of nutritionally deficient EED patients are characterized by having decreased brush border development and impaired cell growth40,41,42.

We also compared gene expression changes in our model with previously identified clusters of intestinal cell-type gene markers43,44,45. In control medium, enterocyte markers were expressed at lower levels in EED chips compared with healthy chips (in particular, MME, APOB and MTTP were reduced by 15-fold, 2.7-fold and 2.3-fold, respectively) and were even more broadly downregulated in EED chips that were exposed to nutritional deficiency (in particular, PCK1, MEP1B and CREB3L3 were decreased by 7.7-fold, 6.2-fold and 4.9-fold, respectively; Supplementary Fig. 3 and Table 3). Paneth cell markers were also downregulated in EED chips compared with healthy chips in control medium (for example, MT2A, CFTR and MT1H were suppressed by ~2–2.5-fold, respectively), but were differentially regulated when chips were exposed to nutritional deficiency. Healthy chips exposed to nutritional deficiency upregulated Paneth cell markers (for example, MT1H, MT1M and MT1G all increased by ~2-fold), while EED chips exposed to nutritional deficiency exhibited further Paneth cell marker downregulation (specifically, ID1, SULT1E1 and MT1X were reduced ~2–2.5-fold). These findings are consistent with the histopathological finding of Paneth cell depletion in EED patient intestinal biopsies38.

Intestinal villus atrophy and barrier compromise

As the transcriptomic analysis revealed downregulation in pathways involved in cell growth and intestinal barrier formation, we carried out differential interference contrast (DIC) and immunofluorescence microscopic analyses, which indeed confirmed that both healthy and EED intestine chips showed dramatically reduced growth of villus-like structures when cultured under nutrition deficient (−N/−T) conditions compared with healthy and EED control chips perfused with complete medium (Fig. 2a,b). Quantification of the height of the epithelium revealed that removal of these nutrients resulted in significant villus blunting in both healthy and EED chips in response to −N/−T deficiency, as indicated by a 70% and 80% reduction in epithelial height, respectively, compared with the same chips cultured in complete medium (Fig. 2c).

Fig. 2: Intestinal villus atrophy and barrier compromise.
figure 2

a, DIC imaging of intestine chips, top-down view (scale bar, 50 µm). b, Immunofluorescence cross-section micrographs showing villus-like structures in the intestine chips. Yellow, phalloidin; Blue, Hoechst; Dashed white line, upper surface of chip membrane. Scale bar, 50 µm. c, Villus-like structures height differences between control and −N/−T, healthy (mean: 211 µm for control and 60 µm for −N/−T) and EED (mean: 288 µm for control and 48 µm for −N/−T) intestine chips. Healthy, P < 0.000001, EED, P = 0.000685, multiple Student’s t-test. Each datapoint on the graph represents the average of 4 to 5 measurement points (membrane to top of the villi) measured in 2 to 3 cross-sectional images of the chip in at least 2 chips per condition. Two different healthy donors and one EED donor were used. Each chip corresponds to one biological replicate. d, Immunofluorescence microscopic views of longitudinal section through the two-channel intestine chips showing the mucus layer overlying the epithelium in the upper channel stained with Alexa 488-conjugated lectin. Scale bar, 250 µm. e, Graph showing differences in mucus thickness between control and −N/−T, healthy (mean: 448 µm for control and 243 µm for −N/−T) and EED (mean: 533 µm for control and 221 µm for −N/−T) intestine chips. Each datapoint on the graph represents the average of 2 measurement points measured in views similar to those shown in d; 2 chips per condition with chips created from cells from two different healthy donors and one EED donor. Healthy, P = 0.004092; EED, P < 0.000001, multiple Student’s t-test. Each chip corresponds to one biological replicate. f, Graph showing differences in Papp between control and −N/−T, healthy (mean: 3.82 × 10−7 for control and 3.41 × 10−6 for −N/−T) and EED (mean: 5.71 × 10−7 for control and 1.41 × 10−6 for −N/−T) intestine chips (9 for healthy chips and 8 for EED intestine chips). Healthy, P = 0.000005; EED, P = 0.001382, multiple Student’s t-test. Each chip corresponds to one biological replicate. Graphed data in c, e and f are mean ± s.e.m.

Source data

Another gene cluster identified as being preferentially sensitive to nutritional deficiency includes genes governing brush border structural integrity. These genes included myosin 1a (MYO1A), which links actin to the overlying apical membrane and whose absence results in irregularities of microvilli packing and length46; protocadherin-24 (CDHR2) that forms links between adjacent microvilli and is the target of enterohemorrhagic Escherichia coli-mediated brush border damage47; and mucin-like protocadherin (CDHR5) that forms heterophilic complexes with CDHR2 (Supplementary Fig. 4). Indeed, scanning electron microscopic imaging revealed that culturing healthy intestinal epithelium in −N/−T medium on-chip resulted in severe loss of apical microvilli (as well as links between adjacent microvilli) relative to control enterocytes that had their entire surface covered with tightly packed microvilli (Supplementary Fig. 5). Analysis of mucus accumulation by live imaging and staining with fluorescent lectin also revealed that both the healthy and EED chips exhibited a much thinner mucus layer when exposed to nutrient-deficient conditions (Fig. 2d,e).

Following these observations of structural changes due to exposure to nutritional deficiency, we next leveraged the advantage of using a two-channel microfluidic intestine chip (Fig. 1a) (that is, as opposed to intestinal organoids cultured within a solid extracellular matrix gel) to assess the effect of exposure to −N/−T medium on differences in intestinal barrier function between healthy and EED chips. We compared apparent permeability (Papp) values, which were measured by calculating the transfer of Cascade Blue fluorescent tracer (596 Da) from the epithelial lumen in the top channel to the underlying parallel channel below. Both healthy and EED control chips exhibited a tight barrier under baseline conditions (within the 10−7 Papp range) and displayed small but statistically significant reductions in barrier function when exposed to nutritional deficiency, as indicated by 8.9- and 2.5-fold increases in Papp for the healthy and EED chips, respectively (Fig. 2f).

Reduced nutrient absorption

Our transcriptomic analysis also revealed that nutritional deficiency resulted in the downregulation of multiple genes associated with the uptake and processing of important nutritional components, including fatty acids, certain amino acids and carbohydrates (Figs. 1e, 3a, 4a and Extended Data Fig. 2). This is clinically relevant because reduced absorption of nutrients is another hallmark of EED, and it affects weight and linear growth as well as cognitive development in children48,49,50,51. For example, expression levels for fatty acid translocase, cluster of differentiation 36 (CD36), microsomal triglyceride transfer protein (MTTP), apolipoprotein B (ApoB) and apolipoprotein C-III (ApoC3) were all lower in nutritionally deficient epithelium (Fig. 3a). Similarly, when we used immunofluorescence microscopy to assess expression of ApoB protein, which is a marker of chylomicron and fat metabolism in the intestine52, we found that exposure to −N/−T nutritional deficiency resulted in significant downregulation of ApoB in EED intestine chips (Fig. 3b and Supplementary Fig. 6). Furthermore, when we quantified cellular uptake of fatty acids using fluorescently labelled dodecanoic acid, we found that exposure to nutritional deficiency reduced fatty acid uptake by 1.68- and 1.69-fold in healthy −N/−T chips and EED −N/−T chips, respectively, compared with healthy and EED control chips (Fig. 3c). These findings are consistent with clinical data that similarly show impaired fatty acid metabolism in children suffering from EED16 and suggest that nutritional deficiency alone is sufficient to reduce fatty acid uptake even in healthy intestine.

Fig. 3: Reduced nutrient absorption.
figure 3

a, Transcriptional pathway analysis revealed a strong theme of downregulation for genes related to fatty acid uptake when EED chips were exposed to −N/−T media. The log2 fold changes between the average expression levels are depicted along with the associated 95% confidence interval. These changes included a 9.8-fold downregulation of the receptor CD36 (q < 0.001), a 10.2-fold downregulation of ApoB (q < .001) and an 8.2-fold downregulation of ABCC2 (q < 0.001). There was a similar trend towards downregulation when healthy chips were exposed to −N/−T media, but most gene changes did not reach statistical significance; 3 chips for each condition. Unmoderated Student’s t-test and the FDR method for multiple testing correction. Each chip corresponds to one biological replicate. b, Differences in ApoB corrected total fluorescence expression in control and −N/−T healthy and EED intestine chips. Healthy (mean: 1.7 × 105 for control and 1.2 × 105 for −N/−T), not significant; EED (mean: 2.2 × 105 for control and 1.3 × 105 for −N/−T), P = 0.012284. Each datapoint on the graph represents the average of corrected total fluorescence expression from 3 different measurement areas of intestine chip cross-sections, from at least 2 chips per condition. Each chip corresponds to one biological replicate. c, Differences in fluorescently labelled dodecanoic fatty acid uptake by healthy (left) and EED (right) intestine chips at the 5 min, 10 min, 30 min and 60 min timepoints. For each timepoint, 7 images covering the entire area of a representative chip from each condition were used to calculate the corrected total fluorescence expression (bottom). Graphed data in b and c are mean ± s.e.m. *P < 0.04, ***P ≤ 0.0008, ****P < 0.0001 by a 2-way ANOVA.

Source data

Fig. 4: Abnormal amino acid uptake and metabolism.
figure 4

a, Amino acid processing and transport was among the strongly downregulated themes when EED chips were exposed to −N/−T media. This included a 6.2-fold downregulation of MEP1B (q < 0.001), a 4.4-fold downregulation of SLC2A2 (q = 0.0074) and a 3.3-fold downregulation of XPNPEP2 (q = 0.0036), which are shown as the log2 fold changes between the average expression levels along with the associated 95% confidence interval. There was occasional downregulation of these genes when healthy chips were exposed to −N/−T media, but none that achieved statistical significance; 3 chips for each condition. Each chip corresponds to one biological replicate. b, Heat map showing 36 metabolites taken up and transferred from the luminal medium to the lower channel of the intestine chips at higher abundance in the healthy vs EED chips (DM, differentiation medium; 3 chips for each condition). Each chip corresponds to one biological replicate. c, Venn diagram showing the number of common and unique metabolites secreted by the cells in intestine chips.

Source data

Abnormal amino acid uptake and metabolism

Children suffering from EED exhibit impaired development, and protein availability from the diet is a key factor responsible for linear growth; thus, we next explored differences in uptake by healthy and EED intestine chips. In the intestine, dietary protein is broken down into short peptides and free amino acids that are taken up by enterocytes, these short peptides and free amino acids serving as building blocks and energy sources for various organs and tissues48. Transcriptomic analysis revealed absorption and metabolizing factors that were downregulated in EED control chips vs healthy control chips, including the solute carrier family 2 (facilitated glucose transporter) member 2 (SLC2A2) and solute carrier family 2 (facilitated glucose/fructose transporter) member 5 (SLC2A5). Other nutrient transporters (such as amino acid absorption and metabolizing factors) that were found to be significantly downregulated in EED chips in response to −N/−T deficiency (EED −N/−T vs EED control), include SLC36A1, which encodes the proton-coupled amino acid transporter 1, ANPEP membrane enzyme alanyl aminopeptidase that is responsible for peptide digestion at the brush border, retinol binding protein 2 (RBP2), cytochrome P450, family 27, subfamily A, and polypeptide 1 (CYP27A1) (Fig. 4a and Supplementary Fig. 7).

We then assessed differences in absorption of nutrients between the healthy and EED intestine chips grown in control medium by performing untargeted metabolomic analysis by liquid chromatography tandem mass spectrometry (LC–MS/MS) to analyse epithelial uptake of nutrients and transfer of these molecules from the lumen of the intestinal epithelium in the top channel to the underlying basal channel (Fig. 1a). We detected 36 metabolites (out of >500 identified) that exhibited lower transport in EED intestine chips compared with healthy chips grown in control medium. These included mainly amino acid metabolites, but also metabolites related to nucleotides, cofactors, lipids, carbohydrates and xenobiotic pathways (Fig. 4b; an extended list of metabolites analysed can be found in Supplementary Fig. 8). Interestingly, 9 of these metabolites were amino acids previously identified as being reduced in serum of Malawian stunted children53. These included essential amino acids (isoleucine, leucine, methionine, phenylalanine, lysine), conditionally essential amino acids (arginine, glycine) and non-essential amino acids (glutamate, serine).

Our LC–MS/MS analysis also revealed 74 metabolites that were secreted by the intestinal cells as they were absent or at extremely low levels (<5%) in the perfusion medium. Of these metabolites, 17 were unique to the healthy control chips and included products of pathways related to metabolism of amino acids (N6,N6,N6-trimethyllysine, 3-hydroxyisobutyrate, methionine sulfone, N-acetylglutamine, homoarginine, indolelactate, glutamine, alanine, betaine, imidazole lactate, 3-methylhistidine, 1-methylhistidine), nucleotides (thymine, 2’-deoxyuridine, 7-methylguanine), xenobiotic metabolism (guaiacol sulfate) or lipids (3-hydroxy-3-methylglutarate). Interestingly, we also identified 4 metabolites unique to the EED chips cultured in control medium, including products of purine nucleotide metabolism (xanthine), vitamin B6 metabolism (pyridoxal), citric acid cycle/energy metabolism (alpha-ketoglutarate) and fatty acid metabolism (acetylcarnitine (C2)) (Fig. 4c).

To assess whether the observed differential uptake and metabolism of molecules in the intestine chips were transporter dependent, we quantified uptake of the dipeptide, glycyl-sarcosine (Gly-Sar), by the epithelial cells in the top channel, and its transfer to the lower channel using LC–MS/MS. This analysis revealed reduced uptake and transport of Gly-Sar in nutritionally deficient healthy intestine chips compared with the same chips cultured in control medium (Supplementary Fig. 9). In addition, these studies confirmed that these effects were due to transport through the PEPT1 transporter, and not due to passive inter- or intracellular diffusion, as this response could be completely prevented by adding Gly-Gly dipeptide (1 mM), which is a specific inhibitor of this transporter54 (Supplementary Fig. 9).

Altered inflammatory mediators

Altered intestinal inflammation is a key component of EED and our transcriptional analysis revealed that genes encoding key inflammatory marker proteins, such as lipocalin 2 (LCN2) and regenerating islet-derived protein 3 alpha (REG3A), were upregulated when EED or healthy chips were exposed to nutritional deficiency (Fig. 1c and Extended Data Fig. 3). Indeed, when we quantified the expression of nine key intestinal cytokines using a bead-based multiplexed ELISA assay, we detected higher levels of several cytokines (IL-6, ICAM, VCAM, IL-33, MCP-1, MIP-1 alpha and IL-8) in the epithelial lumen of healthy −N/−T intestine chips compared with the same chips cultured in control medium (Fig. 5a). The antimicrobial peptide Reg3A was also downregulated by nutritional deficiency in these chips. Interestingly, EED intestine chips grown in complete medium displayed reduced levels of all of the secreted cytokines analysed, while their levels were significantly upregulated when these chips were grown in nutritionally deficient medium (Fig. 5a). Similar responses were observed when we analysed cytokine levels in the lower parenchymal or vascular channel; however, significantly higher levels of the inflammatory cytokines IL-8, IP-10 and MCP-1 were observed (Fig. 5a). Interestingly, we found that intestine chips that were not exposed to mechanical peristalsis-like deformation secreted lower levels of IL-8 and MCP-1 (Extended Data Fig. 4).

Fig. 5: Altered inflammatory mediators.
figure 5

Heat maps showing differential expression of 9 cytokines secreted into the lumen or lower channel of the control or −N/−T, healthy and EED intestine chips and quantified using a bead-based multiplexed ELISA. The colour-coded scale represents the log2 fold change in expression; 3–6 chips for each condition. Each chip corresponds to one biological replicate.

Source data


Given the importance of studying the pathophysiology and the underlying mechanisms of EED, and the current lack of human in vitro models, this study leveraged the human intestine chip technology to create an in vitro EED model using cells obtained from EED and healthy patient intestinal biopsies. The intestine chip offers a unique advantage over other advanced in vitro models, such as intestinal organoids, because direct access to the two parallel flow channels of the device enable quantitation of intestinal barrier function as well as transepithelial absorption and transport that are not possible in static three-dimensional enteroid cultures. Using this approach, we explored the effect of nutritional deficiencies on the manifestation of the disease at both the phenotypic and functional levels. Importantly, our results showed that human intestine chips lined by intestinal epithelial cells isolated from organoids derived from EED patients mimic key features of the transcriptome signature of EED patients, including upregulation of antimicrobial genes and downregulation of metallothioneins and genes involved in digestion and metabolism, but only when exposed to nutritional deficiency, which we modelled by removing niacin and tryptophan from the medium. In contrast, nutritional deficiency induced similar intestinal villus atrophy, disruption of barrier function, and changes in amino acid and fatty acid absorption in chips lined by cells from both healthy and EED patients. Thus, we were able to attribute these various responses specifically to nutritional deficiency, genetic or epigenetic changes in the intestinal epithelium, or a combination of both—distinctions that have not been possible to make in clinical studies, at least for the clinically interesting subpopulation of EED patients studied here who are referred to biopsy after failing nutritional intervention.

Importantly, although the number of EED donors used in this study was limited, the results obtained with EED intestine chips were consistent between donors and compared favourably with genetic data from clinical studies with much larger numbers of EED patients carried out in Pakistan (SEEM study), as well as in similar clinical studies in Zambia and Bangladesh. The Zambian study compared stunted paediatric patients who were unresponsive to nutritional intervention, to paediatric patients or adults with severe acute malnutrition, and EED relevant changes were seen in NADPH oxidases (such as DUOXA2), chemokines (such as CXCL5), metallothioneins (for example, MT1X), membrane ion transport (such as SLC4A7 and SLC23A3), antimicrobial defence (in particular, SAA1, SAAA2 and DUOXA2) and mucosal protection (such as TFF1, TFF2 and MUC6)55. Interestingly, a duodenal proteomic profile that correlated with dysbiosis in a nutritionally refractory stunted Bangladeshi patient cohort also showed overlap with genes related to brush border structure, mucosal function and inflammation (downregulation of CDHR5, CLIC5 and TFF4; upregulation of LCN2)56. Additional organ-on-a-chip studies will need to be carried out with cells from a larger number of EED patients to rule out any potential biases due to small sample size and to determine their generalizability to other EED patient subpopulations.

Owing to the modular design of the intestine chip, future studies using the EED intestine chip could provide a more in-depth analysis of the immune compartment in EED by incorporating immune cells into the model. Because the immune response plays an important role in EED, addition of this complexity to the model could provide greater insight into associated inflammatory processes, their interplay with nutritional deficiencies as well as bacterial infections, and most importantly, provide an improved testbed for EED treatments. As these intestine chips can support co-culture with complex human gut microbiome26, this model could also enable future studies with and without healthy or EED patient-derived duodenal bacterial isolates to further define the separate and combinatorial causes of dysbiosis and intestinal dysfunction in EED pathophysiology.

Malnutrition in EED can be regarded as both a cause and an outcome of the disease. In this study, we examined the effect of dual deficiency of an essential amino acid tryptophan and the vitamin niacin (in the form of niacinamide) because of their reported effect on intestinal development and function, and the suggested correlation between their deficiencies and EED development18,53. Remarkably, we observed a 6-fold increase in the number of genes differentially expressed in EED intestine chips exposed to nutritional deficiency compared with healthy chips grown in complete medium. Moreover, many of the affected pathways were associated with nutritional uptake and cellular energy processing. These results imply that nutritional deficiency itself leads to derangements in nutritional processing that create a positive feedback loop, which further worsens the nutritional deficiency in EED patients.

Villus blunting and barrier dysfunction are hallmarks not only of EED, but also of other intestinal pathological conditions, such as inflammatory bowel diseases, coeliac, diarrhoea and small-intestine bacterial overgrowth57,58. Indeed, we were able to show in this study that these phenotypes can manifest because of nutritional deficiencies regardless of whether the intestinal epithelium was derived from healthy or EED patients. This implies that villus blunting and barrier dysfunction observed in EED may be a response to environmental conditions, rather than an inherent genetic or developmental feature of this disease. This finding has high clinical relevance because the intestinal epithelium undergoes continuous shedding and renewal every 3–5 d; hence, the chronic negative effect of nutritional deficiencies might explain a fundamental aspect of the malabsorption and poor response to oral vaccination seen in EED patients. The ability to dissect and assess the factors leading to villus blunting in EED (as well as in other intestinal conditions) in vitro is a unique capability enabled by intestine chips that support the formation of three-dimensional villus-like structures as well as mucus production and quantification of transepithelial transport, absorption and secretion.

Fat is a macronutrient responsible for 30–40% of total caloric intake in children and 20–35% in adults, and fatty acid composition has a direct effect on health and development, including inflammatory status and cognitive development50,51. Our transcriptomic analysis was sensitive enough to detect several genes, including ApoB, that were downregulated in nutritionally deficient epithelium. We confirmed that the expression of this molecule is decreased, using immunofluorescence microscopy, and that fatty acid absorption is impaired in both EED −N/−T and healthy −N/−T chips (compared with their respective controls), using a fatty acid uptake assay.

Given that dietary protein is an important macronutrient directly linked to linear growth and the results from our transcriptome-wide analysis revealed downregulation of amino acid transporters in EED chips, we also conducted an untargeted metabolomic analysis that identified several amino acids that are significantly reduced in the EED control chips compared with healthy control chips. More interestingly, we identified metabolites that were secreted by the intestinal epithelium into the lower channel. Thus, the intestine chip may be used as a nutritional and metabolic screening tool where uptake, utilization and secretion of specific metabolites by and through the intestinal epithelium can be followed in high resolution and quantified over time. Moreover, future analysis of molecules released into the lower flow channel could lead to identification of biomarkers of disease severity and/or progression that might be detectable in blood.

While reduction of intestinal absorptive surface area due to villus blunting caused by nutritional deficiency could impair nutrient uptake by the intestine, exposure to nutritional deficiency also directly suppressed expression of multiple genes related to nutrient absorption, specifically in chips lined by cells from EED patients. Thus, these results suggest that nutritional deficiency has a two-fold effect in these patients, which would probably manifest in a greater degree of intestinal dysfunction and a more severe EED phenotype. In our study, we also found that EED intestine chips exposed to nutritional deficiencies produced greater amounts of inflammatory cytokines compared with healthy chips grown under the same −N/−T conditions. This is a critical detail as inflamed intestine has higher caloric demands for basic maintenance and renewal, which could result in a negative caloric balance unable to support catch-up growth. Furthermore, chronic inflammation may negatively affect the efficacy of oral vaccines in the EED intestine. As such, it is possible that the lack of catch-up growth in EED children receiving nutritional intervention is due to current interventions only supplying nutrients aimed at replenishing the deficiency in tissues responsible for linear growth (bone, muscles)59. A more effective approach might be to first administer a diet composition that preferentially promotes intestinal recovery with formation of new villi and restores increased intestinal absorptive area before moving to supplementation required for catch-up growth.

The ability of the intestine chips to allow the study of the effects of environmental factors (such as cell source and nutrient levels) individually or in combination in a controlled manner and to explore multiple clinically relevant outcomes, such as cell and tissue morphology, barrier function, nutrient metabolism and absorption, inflammatory status and transcriptome modifications, enabled us to distinguish between manifestations more common to multiple intestinal diseases versus responses that are unique to EED. For example, we identified the CD36 gene, which has several functions (in particular, as a fatty acid translocase, as a regulator of inflammation, and in oxidative stress and angiogenesis) relevant to EED as well as to intestinal cancer and other intestinal diseases, to be significantly downregulated under nutritional deficiencies in both healthy and EED intestine chips. Remarkably, our findings relating to EED are directly in line with recently published clinical data that explored unique signatures of EED-affected children compared with healthy controls and children with coeliac disease38. Thus, this in vitro EED model may be useful for gaining further insight into the pathophysiology of this disease as well as for development of potent therapeutics. The intestine chip could also find uses in personalized medicine and nutrition by leveraging clinical biopsies, potentially allowing for personalized (patient-specific) digestion, absorption and allergic reactions to be assessed for different nutrients.


Organoid cultures and intestine chips

Organoids from healthy donors or EED patients were generated from biopsy samples collected during exploratory gastroscopy following a procedure previously described60. A total of 3 healthy and 2 EED donors were used to generate the data in this study (Supplementary Table 4). For the healthy chips, de-identified endoscopic tissue biopsies were collected at Boston Children’s Hospital from grossly unaffected (macroscopically normal) areas of the duodenum in 2-, 10- and 11-year-old patients undergoing endoscopy for gastrointestinal complaints. Informed consent was obtained at Boston Children’s Hospital from the donors’ guardians. All methods were carried out in accordance with the approval of the Institutional Review Board of Boston Children’s Hospital (Protocol number IRB-P00000529). For the EED chips, samples were collected in Pakistan and newborns in this study were registered and followed every month for anthropometry. Children with weight for height (WHZ) scores of less than −2 at their 3–6 month follow up visit were selected as cases. The de-identified endoscopic tissue biopsies were collected from affected areas of the duodenum in 1.5-year-old and 1.9-year-old patients undergoing endoscopy following unsuccessful educational and nutritional intervention for wasting. Informed consent was obtained at the household level in a rural district of Matiari, Sind, Pakistan from the donors’ guardians. All methods were carried out in accordance with the AKU Ethical Review Committee’s approval (ERC number 3836-Ped-ERC-15). Organoids were kept in complete growth medium28,60, and passaged every 7 d in a 1:4 ratio. Before cell seeding, S-1 chips (Emulate) were activated using ER1/2 (Emulate) and UV exposure for 20 min. Chips were then coated with 200 μg ml−1 collagen I (BD Corning) and 100 μg ml−1 Matrigel (BD Corning) in serum-free DMEM-F12 (Gibco) for 2 h at 37 °C. After washing, organoids were broken into smaller fragments using enzymatic activity (TryplE, Gibco) and seeded in the luminal upper channel of the chips. They were then allowed to adhere for 24 h before introduction of flow and mechanical deformation as previously described28. For the −N/−T treatment, niacinamide and tryptophan were removed from the basal medium (DMEM-F12, Gibco) used to prepare the expansion culture medium (used for the luminal and lower channels) and no additional niacinamide was added28. After 16–18 d in culture with continuous flow (60 µl h−1) and mechanical deformation (10%, 0.15 Hz), the medium was changed to differentiation medium (serum and Wnt-3A free28, and −N/−T free for the respective group) in the luminal top channel and expansion culture medium in the lower channel for 4 additional days.

Microarray sequencing and bioinformatics analysis

Initial microarray experiments were carried out using one healthy donor (n = 3 biological replicates) and one EED donor (n = 3 biological replicates) and were reflective of a recent clinical EED transcriptomic signature derived from a larger population (SEEM study, n = 25 healthy donors, 52 EED donors). Subsequent validation studies were carried out using 1–3 healthy donors (n = 3–9 biological replicates) and 1–2 EED donors (n = 3–8 biological replicates) per experiment. RNA samples were processed using the Genechip WT PLUS reagent kit and hybridized to Affymetrix Human Clariom D arrays. Robust multichip average (RMA) was used to generate normalized expression intensity data for subsequent analysis (R package, oligo; function, rma). Differential expression analysis was performed with limma61 R package for each comparison pair; gene expression values were averaged for each condition. Template matching was used to extract genes that are differentially expressed between these conditions. Differential gene expression heat map analysis was performed using Euclidean distance and McQuitty’s linkage within the R package heatmaply39. To normalize values across experiments, columns were scaled to generate a Z-score. Pathway analysis was performed using the natural language-processing algorithm COmprehensive Multi-omics Platform for Biological InterpretatiOn (COMPBIO) to generate a holistic contextual map of the core biological themes associated with gene expression changes. Enriched concepts associated with differentially expressed genes were compiled from PubMed abstracts using contextual language-processing. Themes were scored using a complex function that incorporates an empirical P value and a standard score similar to a Z-score to give a final value representing fold enrichment over a random clustering as follows: 3–9, weak relationship; 10–99, modest relationship; 100–999, strong relationship; 1,000+, very strong relationship. A theme map was generated where themes are represented as nodes and interconnections between themes are represented as edges, with the thickness of an edge relating to the degree of interconnection.

We also compared intestine chip results with those obtained from samples from the SEEM study in which Pakistani patients were enrolled as newborns and their growth trajectories were followed to 24 months. Children who were malnourished (weight for height < −2.0) at 3–6 months of age were selected and given nutritional intervention if they continued to be malnourished at 9 months of age. Those EED patients who were unresponsive to nutritional intervention underwent esophagogastroduodenoscopy and histological evaluation as part of the clinical workup. SEEM samples were sequenced using RNA-seq technology as previously detailed38. Comparisons between the two studies were made using Z-score-scaled fold-change data.

Immunofluorescence microscopy

Immunofluorescence microscopic imaging was carried out using the following steps: the apical and basal channels of the chips were gently washed with PBS and fixed with 4% paraformaldehyde (Electron Microscopy Sciences, 157-4) in PBS for 30 min, then washed twice with PBS and kept at 4 °C. The fixed samples were sectioned to 150–250 µm sections using a vibratome (Leica), and then permeabilized and blocked with 0.1% Triton X-100 solution and 10% donkey serum in PBS for 30 min at room temperature. Then primary antibody (Apo-B; Abcam, ab20737) was added (1:100 in 1.5% BSA/PBS solution) and the samples were incubated overnight at 4 °C, followed by multiple PBS washes. Cells were then incubated with secondary fluorescent antibody (Invitrogen, SA5-10038) and phalloidin (Invitrogen, A12380) at room temperature for 60 min and washed with PBS; nuclei were co-stained with Hoechst 33342 (Sigma, 14533). Microscopy was performed with a laser scanning confocal microscope (Leica SP5 X MP DMI-6000 or Zeiss TIRF/LSM 710).

Paracellular permeability measurements

To assess paracellular permeability, 50 μg ml−1 of Cascade blue (596 Da; ThermoFisher, C687) was introduced to the luminal channel (at 60 ml h−1). After flowing overnight to saturate the microfluidic channels, outflows were discarded and collection for measurements were conducted for ~24 h. The fluorescence intensities (390 nm/420 nm) of the top and bottom channel effluents were measured using a multimode plate reader (BioTek NEO). The apical-to-basolateral flux of the paracellular marker was calculated using the following equation: Papp = (dQ/dt)/AdC, where dQ/dt (g s−1) is molecular flux, A (cm2) is the total area of diffusion and dC (mg ml−1) is the average gradient.

Mucus assessment

Mucus was visualized using a wheat germ agglutinin (WGA)–Alexa Fluor 488 conjugate (ThermoFisher, W11261) for live cell imaging, as described previously26 with some modifications. Briefly, WGA solution (25 μg ml−1 in Hanks’ Balanced Salt Solution, HBSS) was allowed to flow through the epithelium channel for 30 min and then washed with continuous flow of HBSS for 30 min. Intestine chips were then cut sideways parallel to the length of the channel and imaged with an epifluorescence microscope (Zeiss Axio Observer Z1) with ×5 objective.

Fatty acid uptake

Chips were starved for 1 h by replacing the luminal and lower channel media with HBSS. Then, fluorescently labelled dodecanoic acid combined with a quencher (to eliminate any unspecific signal) were added according to the manufacturer instructions (BioVision, K408) to the luminal upper channel of the intestine chips. The entire length of the channels was then imaged with an epifluorescence microscope (Excitation/Emission = 488/523 nm, Zeiss Axio Observer Z1) with ×5 objective at 5, 10, 30 and 60 min.


The medium in the lower channel of the chips was changed to HBSS and allowed to flow for 30 min to clear any residues. Then collection of outflows was conducted over 5 h. Samples were frozen immediately after collection (−80 °C) and submitted for LC–MS/MS analysis. The Metabolon global metabolomics platform was used to measure biochemicals in cell and media samples. Samples were prepared using the automated MicroLab STAR system (Hamilton). Several recovery standards were added before the first step in the extraction process for quality control purposes. To remove protein, dissociate small molecules bound to protein or trapped in the precipitated protein matrix, and to recover chemically diverse metabolites, proteins were precipitated with methanol under vigorous shaking for 2 min (Glen Mills GenoGrinder 2000), followed by centrifugation. The resulting extract was divided into five fractions: two for analysis by two separate reverse phase (RP)/Ultra Performance Liquid Chromatography (UPLC)–MS/MS methods with positive ion mode electrospray ionization (ESI), one for analysis by RP/UPLC–MS/MS with negative ion mode ESI, one for analysis by Hydrophilic interaction (HILIC)/UPLC–MS/MS with negative ion mode ESI, and one sample was reserved for backup. Samples were placed briefly on a TurboVap (Zymark) to remove the organic solvent. Raw data were extracted, peak-identified and QC processed using Metabolon’s hardware and software.

Beginning with the OrigScale values from Metabolon, which are normalized in terms of raw area counts, we calculated total ion count in the outflow to visualize sample-wise variance. Next, we normalized each sample by its total ion count and rescaled each metabolite’s value by dividing each metabolite by its root mean square. We visualized metabolite abundance using a heat map. Statistical analysis and heat map generation was performed using R (R Foundation for Statistical Computing). We performed differential expression analysis by using the limma61 R package (v3.32.10) to fit a linear model to the data. log2 fold change, P value and adjusted P value were calculated for each comparison using an unmoderated Student’s t-test and the false discovery rate (FDR) method for multiple testing correction62. Adjusted P values are shown in volcano plots. To assess transporter mediated uptake and transfer from the luminal upper channel to the lower channel, glycyl-sarcosine (Gly-Sar, 1 mM, Sigma) alone or in combination with the specific PEPT1 transporter inhibitor Gly-Gly dipeptide (1 mM, Sigma) was added to the luminal medium and their abundance in the lower channel outflow was assessed.

Statistical analysis

Each intestine chip was used as a biological repeat for one terminal assay. Either a Student’s t-test or 2-way analysis of variance (ANOVA) was performed to determine statistical significance, as indicated in the figure legends (error bars indicate s.e.m.; P < 0.05 was considered significant).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.