Albumin-dependent and independent mechanisms in the syndrome of kwashiorkor

The syndrome of kwashiorkor is a striking phenotype of childhood severe malnutrition (SM) comprising oedema, fatty liver, and skin and hair changes. Despite high fatality, the aetiology and pathophysiology of kwashiorkor remain enigmatic, including the role of serum albumin on oedema development. Here, we demonstrate that serum albumin is associated with the presence and severity of oedema among severely malnourished children. Further, in two independent cohorts of children in Malawi and Kenya, we show albumin-independent mechanisms are associated with oedema in SM, including oxidative stress and extracellular matrix (ECM) remodelling. Plasma concentrations of ECM-related proteins: lumican, podoplanin, lymphatic vessel endothelial hyaluronan receptor 1 (LYVE1) and matrix metalloproteinase (MMP)2 were associated with kwashiorkor. We therefore conclude that the pathophysiology of kwashiorkor has both albumin-dependent and independent mechanisms. We discuss the ways in which albumin-independent mechanisms may explain the clinical features observed in kwashiorkor.


Introduction
The syndrome of kwashiorkor (a.k.a oedematous severe malnutrition) is a striking phenotype of childhood malnutrition comprising oedema, fatty liver, hair depigmentation, a desquamating skin rash and behavioural changes 1 . It is distinct from the syndrome of marasmus (a.k.a severe wasting), characterised by low weight-for-height (<-3 z scored from 2006 WHO standards), low (<115 mm) midupper arm circumference (MUAC), visible atrophy and loose skin. Kwashiorkor and marasmus may cooccur (i.e. marasmic-kwashiorkor). Whilst both phenotypes are assumed to result from inadequate nutritional intake, the aetiology of the oedematous phenotype remains elusive 1 . Several hypotheses on the pathophysiology of kwashiorkor have been proposed but none are supported by robust epidemiological or mechanistic evidence 1 .
The earliest explanation of kwashiorkor was a low protein diet leading to hypoalbuminemia causing pathognomonic oedema 2 . However, in 2007, prospective assessment of food and nutrient intake in a population at risk for kwashiorkor in India and Malawi found no association between measured protein intake and kwashiorkor 3,4 . Furthermore, oedema in SM is observed to resolve independently of protein intake 5 and without an increase in serum albumin 6 . However, the latter has been questioned upon reanalysis of the original data suggesting serum albumin could have increased in these earlier studies 7 .
Circulating essential amino acid (EAA) concentrations, particularly sulphated AAs, are reported to be lower in kwashiorkor compared to marasmus 8,9 . Some of the clinical characteristics of kwashiorkor are similar to those occurring in methionine deficiency, notably the skin lesions and reduced plasma glutathione levels. Furthermore, supplementation study of cysteine has been associated with faster resolution of oedema compared to alanine supplementation 10 . However, supplementation of methionine itself did not yield the same effect 11 , nor did supplementation of an antioxidant mixture containing Nacetylcysteine prevent the development of kwashiorkor 12 . Recently, DNA hypomethylation has been reported in kwashiorkor relative to marasmus during active disease, but not in subsequently recovered cases in adulthood, which the authors attribute to reduced concentrations and methyl-flux of methionine and so did not include faecal samples from children with marasmus 28 . Children with SM in general have an altered gut metagenome composition compared to healthy children 28,29 , indicating that dysregulation of the microbiome may be a cause or consequence of SM, but this does not explain the pathophysiology of kwashiorkor.
Our aim was to determine albumin-independent differences in pathophysiology between the two SM phenotypes to inform targeted prevention and treatment strategies 30,31 .

Results
Low serum albumin is necessary but not sufficient to develop kwashiorkor. We initially determined the association between serum albumin concentration and kwashiorkor using data from a clinical trial of reformulated therapeutic milk among hospitalized children with severe acute malnutrition conducted in Malawi and Kenya (the discovery cohort) 32 . Within this trial, 79% (662/843) of the children had admission serum albumin data with a median (interquartile range; IQR) of 34g/L (IQR 24 -40). HIV, older age, and enrolment at the Malawi site were associated with lower albumin concentrations, while breastfeeding, pre-existing heart disease, and presenting with severe pneumonia or diarrhoea were associated with higher albumin (all p<0.05) in multivariable analysis. Serum albumin concentration was negatively associated with the presence of kwashiorkor (aOR = 0.75 [95% CI: 0.71, 0.78] per g/L, p < 0.001). We observed a significant decline in serum albumin concentration with increasing oedema severity graded according to the WHO classification ( Figure 1a). Almost all children with kwashiorkor had serum albumin levels below 35g/L, but many children with similarly low levels of serum albumin did not have oedema or other features of kwashiorkor (Figure 1a Oedema grading data were not collected for this study.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Oedema resolved within 3 days of hospitalization in almost half (48%) of children admitted with kwashiorkor in the discovery cohort and oedema had improved in 31%. A small proportion (3%) of those admitted with marasmus developed oedema during treatment (Figure 1d). Among children whose nutritional oedema resolved or improved, there was a small increase in serum albumin concentration during 3 days of hospitalisation (0.68 g/L mean increase, p=0.02), whereas serum albumin remained unchanged among those whose oedema did not improve (Figure 1e). However, despite the small increase serum albumin among children whose oedema resolved or improved, concentrations at 3 days (median 20 g/L, IQR 16.5 -26 g/L) were still far below clinically recognized norms in children (34 -54 g/L).
Adjusting for regression to the mean indicated no differences in changes in serum albumin between those without oedema and those with oedema which either improved (p = 0.93) or worsened (p = 0.38).
These findings strongly suggest that although low serum albumin is associated with kwashiorkor, other factors play an essential role in the pathophysiology of the kwashiorkor phenotype. Figure 1. (a) Association between serum albumin concentration and degree of oedema severity in the F75 reformation clinical trial 32 : "None" means no oedema (marasmus) whereas "+" means oedema on both feet; "++" -oedema in both feet and legs; "+++"oedema in both feet, legs, arms, hands and face. Oedema severity was assessed by trained clinicians following World Health Organization guidelines; (b) probability of presenting with oedema based on serum albumin concentration at hospital admission . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint in the discovery cohort: green and red dots indicate those that presented with or without oedema respectively; (c) difference in serum albumin concentration between kwashiorkor and marasmus in the validation cohort; (d) distribution of oedema status after 3 days of hospitalization in the F75 reformulation clinical trial; (e) changes in serum albumin concentration among kwashiorkor during admission and after 3 days of hospitalization Selection of a sub-population matched on serum albumin levels. To determine factors associated with kwashiorkor in conjunction with low albumin, we further selected children from the F75 reformulation trial discovery cohort ( Figure 2). This sub-cohort comprised children with kwashiorkor and marasmus who had been matched on exact serum albumin levels. In the discovery cohort, age and sex distributions were similar in oedematous and non-oedematous groups. Mid-upper arm circumference (MUAC) was higher among children with kwashiorkor (p < 0.001), whereas HIV was more prevalent among children with marasmus (p < 0.001). A greater proportion of oedematous children were recruited in Malawi than in Kenya.
To validate the results obtained from the discovery sub-cohort, a validation sub-cohort was selected from the validation set from the trial in Kenya, with kwashiorkor and marasmus matched on age, sex, site of recruitment and sex, but not matched for serum albumin concentration (more details in the Methods section) (Figure 2). In the validation cohort, 47 children with marasmus and 51 with kwashiorkor were selected based on 25 strata. As with the discovery cohort, serum albumin was higher in marasmus than in kwashiorkor thus, for succeeding analyses, variables were normalized by serum albumin concentration in order to assess differences occurring independently of serum albumin concentrations to harmonize approaches with the discovery cohort. The baseline description of both cohorts is presented in Table 1 and flowcharts of the selection for both cohorts are presented in Figure   2.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Figure 2. Recruitment flow diagram for the discovery and validation cohorts . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 1, 2021.  Volcano plot showing the log odds ratio (x-axis) and -log p value after false-discovery rate adjustment of plasma proteins (y-axis). False discovery adjustment was performed separately for proteins and metabolites. The horizontal line signify the FDR p = 0.05 and p = 0.01 cut-offs, whereas vertical broken line signify log odds ratio = 0. Features on the upper right quadrant represent those with FDR-corrected p values < 0.05, and are associated with kwashiorkor compared to marasmus phenotype, whereas those in the upper left quadrant are associated with marasmus. Estimates were obtained using conditional logistic regression adjusting for age, sex, HIV status and site of recruitment stratified for admission serum albumin concentration. (B) Boxplot showing the association between the plasma concentrations of significantly associated features and the degree of oedema severity. "None" means no oedema (marasmus) whereas "+" means oedema on both feet; "++" -oedema in both feet and legs; "+++"oedema in both feet, legs, arms, hands and face. Oedema severity was assessed by trained clinicians following World Health Organization guidelines; p values were estimated using ordinal logistic regression adjusted for age, sex, HIV status, site of recruitment and serum albumin concentration. (C) Radar plot comparing plasma amino acid content in kwashiorkor and marasmus both with matched serum albumin and serum albumin > 35g/L (high albumin). * denotes difference at p <0.05 between kwashiorkor and marasmus matched with serum albumin. # denoted different at p < 0.05 between kwashiorkor and marasmus with serum albumin >35g/L (high albumin) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Plasma lipids and extracellular matrix proteins are albumin-independent factors associated with kwashiorkor. To discover albumin-independent protein and metabolite factors associated with kwashiorkor, plasma samples from the discovery sub-cohort were subjected to untargeted liquid chromatography tandem mass tag (TMT)-based proteomics and targeted metabolomics (Biocrates™ p 180). After data cleaning and pre-processing, 187 out of 456 proteins and 155 out of 205 metabolites were retained for further analysis. A complete list of annotated proteins and targeted metabolites and their association with nutritional oedema is provided in Supplementary table S1. Nine metabolites including 1 phosphatidylcholine (PC), 6 lysoPC, and 1 sphingomyelin species, apolipoprotein C-I (Apo [95% confidence interval: 0.42, 1.08]) nor with oedema severity. In fact, an almost complete overlap between each free amino acid concentration was observed between kwashiorkor and marasmus ( Figure   3c). Although 5-HIAA was negatively associated with kwashiorkor, plasma levels of the other members of the tryptophan pathway, i.e. tryptophan, serotonin and kynurenine, were not (Supplementary figure   S1). Comparing kwashiorkor with children with marasmus but with serum albumin > 35g/L (high albumin), almost all measured amino acid concentrations were higher (p < 0.05) in the high albuminmarasmus group compared to kwashiorkor, indicating an association between serum albumin concentration on measured plasma amino acid content. Asparagine, aspartic acid, citrulline, ornithine . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint and proline concentrations were not different between kwashiorkor and marasmus, whereas glycine and histidine were significantly lower in the high albumin-marasmus group compared to kwashiorkor. Each module is characterised by an eigenvector (E (q) , which is the 1 st principal component of module q that represents the overall behaviour of the module 35 . Regressing E (q) to a binary outcome (kwashiorkor or marasmus) adjusting for age, sex, HIV and recruitment site, stratified by each value of admission serum albumin, we found seven modules (ME4, 6, 8, 10, 11 and 15) to be significantly associated with kwashiorkor ( Figure 4a,b).

Multi
Most of the associated modules were composed of lipids (ME4, 6, 10-12). However, there was a preferential increase in plasma concentration of unsaturated ester-bound PCs (ME4, 10) compared to ether-bound PCs (ME1) and PCs with lower degree of unsaturation (ME9) in kwashiorkor. Modules . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint containing both ester-and ether-bound PCs were also not associated with kwashiorkor, further emphasizing the preference for esterified PCs. Plasma levels of unsaturated acylcarnitines (ME6) were also preferentially increased in kwashiorkor compared to saturated acylcarnitines (clustered in ME7).
LysoPCs (ME11) and sphingomyelins (ME12) were found to be positively associated with kwashiorkor compared to marasmus. These positive associations of lipid modules to kwashiorkor are corroborated by the results of lipid specific associations ( Figure 3a) and the positive association of proteins involved in lipid transport and metabolism (ME15). Proteins associated with oxidative stress (ME8) was also positively associated with kwashiorkor.
We further observed that proteins linked with extracellular matrix remodelling (ME15) were positively associated with kwashiorkor. This also corroborates with the results of protein specific associations ( Figure 3a), showing that ECM-associated proteins lumican, ITIH2 and HRG were positively associated with kwashiorkor.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Plasma markers of endothelial glycocalyx integrity are associated with oedema severity but not specifically with kwashiorkor. We quantified lumican using ELISA to validate the untargeted proteomics results in both discovery and validation cohorts. Lumican was found to be positively associated with kwashiorkor compared to marasmus (aOR = 1.49 [95% CI: 1.23, 1.79]) per µg/mL, adjusting for age, sex, HIV status and site of recruitment in agreement with the untargeted proteomics results (Figure 5a). Also consistent with untargeted proteomics findings, plasma lumican concentration was positively associated with increasing degree of oedema severity (p < 0.001) (Figure 5b). The positive association of plasma lumican and kwashiorkor was replicated in the validation cohort for lumican levels normalized on serum albumin concentration (p = 0.002) (Figure 5c). Comparing circulating lumican levels between discharge (still SM but free of underlying infections) and 60 days post-discharge (fully recovered from SM and infections) in the validation cohort, lumican was found to significantly increase (p < 0.001) among children with marasmus, whereas it decreased among children with kwashiorkor (p < 0.03).
As lumican has been reported to adsorb on the endothelial glycocalyx 36 , we reasoned that the increase in plasma lumican likely reflected a disruption of the endothelial glycocalyx. Hence, we determined the association between kwashiorkor and plasma endothelial glycocalyx (EG) markers. Of the several EG integrity markers reported in the literature, and given limited sample volumes, we prioritised analysis of two abundant markers in plasma previously reported to be increased in diseases with known EG dysfunction 37 : syndecan-1 (Syn1), a proteoglycan bearing sulphated glycosaminoglycans (GAG), and hyaluronan (HA), a non-sulphated GAG.
Based on literature suggesting increased serum levels of EG markers in diseases associated with leakage of intra-vascular fluid and proteins leading to oedema, such as dengue 37 , we initially hypothesized that increased plasma levels of Syn1 and HA would be associated with kwashiorkor. Surprisingly, neither Syn1 nor HA were associated with kwashiorkor compared to marasmus in the discovery cohort ( Figure   5d-i). Their plasma levels were however negatively associated with increasing degree of oedema severity. These results were replicated in the validation cohort, except for a modest but significant . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. increase in HA among children with kwashiorkor ( Figure 5i). Plasma levels of Syn1 and HA significantly decreased after 60 days post discharge among children with either initial phenotype who fully recovered from malnutrition without further acute illness following discharge. Figure 5. Association between SM phenotype and lumican and endothelial glycocalyx markers.

ECM remodelling is an albumin-independent mechanism in kwashiorkor
Plasma levels of matrix metalloproteinase (MMP)2 in the discovery cohort had a positive association with kwashiorkor (aOR = 1.89 [95% confidence interval: 1.38, 2.58]) per µg/mL, and was associated with severity of oedema (p < 0.0001) (Figure 6a). Because of this, we measured plasma levels of ECM . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint remodelling regulators, i.e. MMP2, tissue inhibitors of MMP (TIMP)1, and TIMP2, and other ECM proteins (podoplanin and LYVE1) in the validation cohort, both at hospital discharge and at day 60 postdischarge among children who achieved full nutritional recovery.
ECM remodelling regulators MMP2 and TIMP1 were positively associated with kwashiorkor.
Furthermore, plasma levels of these proteins significantly reduced during nutritional rehabilitation ( Figure 6b). Apart from lumican, other ECM proteins were altered in plasma of children with kwashiorkor. Podoplanin, a mucin-like glycoprotein found in the alveoli, heart and lymphatic vascular system, was negatively associated, whereas lymphatic vessel endothelial HA receptor 1 (LYVE1) was positively associated with kwashiorkor. Plasma levels of LYVE1 also reduced during nutritional rehabilitation in both the kwashiorkor and marasmus phenotypes, but podoplanin remained unchanged.
As ECM remodelling is activated by inflammation, we further measured key plasma markers of systemic inflammation in both SM phenotypes. None of the inflammatory cytokines measured were differentially abundant between kwashiorkor and marasmus cases matched for serum albumin, suggesting inflammatory response-independent mechanisms were driving ECM remodelling in kwashiorkor. It can be observed however that plasma levels of IL8, IL10, VEGFR3, GRO alpha KC and sCD14 decreased significantly in plasma of validation cohort children with kwashiorkor during the 60 days post-hospital discharge but remained unchanged in marasmus. Plasma IL6 significantly decreased in both kwashiorkor and marasmus upon full nutritional and clinical recovery ( Figure 6B).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Figure 6. Association between SM phenotype and markers of ECM remodelling markers and systemic inflammatory.

Discussion
There is a universal understanding within the medical literature of the role of serum albumin concentration on oedema formation in general, as explained by Starling forces. However, in . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint hypoalbuminemic states such as the nephrotic syndrome and inflammation, evidence suggests that other factors beyond a decline in colloid osmotic pressure are also responsible for the altered fluid distribution 38 . The role that albumin plays in kwashiorkor has been a topic of debate among scientists, with some postulating a causal role of hypoalbuminaemia in its aetiology 7,39 while others rejecting their association 6,40 . In this study, we showed that kwashiorkor is associated with both albumin-dependent and independent mechanisms. We demonstrated for the first time a role of ECM degradation as an albuminindependent mechanism in the pathophysiology of kwashiorkor.
Our data from two independent cohorts show that serum albumin is lower among children with kwashiorkor compared to marasmus and negatively associated with the degree of oedema severity. In fact, there were almost no cases of kwashiorkor where serum albumin concentrations were more than 35 g/L in the discovery cohort. In the validation cohort who were enrolled towards the end of their hospital admission having initiated therapeutic feeding and no longer suffering acute infection, 35% of kwashiorkor cases had serum albumin levels above 35 g/L compared to 67% for the marasmus cases.
These data provide the first physiological validation of the current WHO nutritional oedema grading system 41 . However, in the absence of active acute infection (validation cohort), serum albumin concentrations overlapped more between groups compared to at admission to the hospital (F75 reformulation trial). Furthermore, we observed that resolution or improvement of oedema was accompanied by a small but not likely clinically relevant increase in serum albumin concentrations, agreeing with previous observations 5, 6 . These led us to postulate that other factors beyond albumin play important roles in oedema formation in SM.
Using a targeted metabolomics approach, we previously showed that serum concentrations of 141 metabolites (including AAs) tended to be lower in kwashiorkor compared to marasmus in general, and not only AA 42 . In this current study, we found that plasma AA profile was very similar between kwashiorkor and marasmus when matched for serum albumin concentrations, contrary to previous reports in which serum albumin concentrations were not considered 8,9,42 . These data indicate an association between serum albumin and serum free AA concentrations. One explanation for this is water . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint displacement by albumin. In conditions of low albumin content, water replaces the space that would have been occupied by albumin per volume of serum, thereby causing a dilution of polar metabolites.
Hence, an apparently lower content per unit volume of polar metabolites is observed in kwashiorkor compared to marasmus with high serum albumin concentration. Another potential explanation is that the lower serum albumin concentration is a reflection of low concentrations of AA precursors needed for albumin synthesis. We therefore addressed these issues by matching for serum albumin when comparing kwashiorkor and marasmus in the discovery cohort and normalising for serum albumin in the validation cohort.
Our results also demonstrate that kwashiorkor is associated with increased levels of oxidative stress (increased hydrogen peroxide catabolism, oxygen transport, response to reactive oxygen species and cellular oxidant detoxification). However, whether increased oxidative stress causes or is a consequence of kwashiorkor requires further investigation. Nonetheless, we showed in this study that oxidative stress was increased in kwashiorkor independently of the plasma AA profile, which agrees with one of the postulates of the free radical hypothesis 16 . However, we did not find evidence for an association between plasma levels of endothelial glycocalyx components, Syn1 and HA, and kwashiorkor. Instead we found other ECM proteins that are associated with kwashiorkor, such as lumican, LYVE1, podoplanin and MMP2. It is important to note however that endothelial glycocalyx components are also structurally linked to ECM proteins.
Lumican is a leucine-rich proteoglycan with keratan sulphate, a sulphated glycosaminoglycan (sGAG), side chains and is a major component of corneal, dermal and muscle connective tissue. Downregulation of lumican results in skin fragility and laxity, and corneal opacity 43 . Hence, degradation of lumican could explain the skin changes 44,45 and corneal opacity 46,47 reported to occur in some children with kwashiorkor. Although, no study has linked lumican specifically with the type of skin changes in kwashiorkor, i.e. "flaky paint" or "peeling paint" dermatosis, reduction of lumican levels have been reported in skin diseases such as actinic keratosis and Bowen disease 48 .
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint The lymphatic system could also be involved in the clinical manifestations of kwashiorkor, especially oedema. LYVE1 and podoplanin are both markers of lymphatic endothelial integrity predominantly expressed in lymphatic vessels [49][50][51] . LYVE1 and podoplanin are essential for lymphatic system development and ablation of podoplanin and LYVE1 in transgenic mice resulted in diminished lymphatic transport and lymphedema 52,53 . Hence, differential plasma levels of these markers indicate that lymphatic system could be compromised in kwashiorkor, leading to poor fluid homeostasis.
Degradation of ECM in the lymph may also explain the disturbances in lipid metabolism observed in kwashiorkor, being the main route for chylomicrons from enterocytes to reach the bloodstream 54 .
Interestingly, reduction of podoplanin by knock-out of T-synthase in mice caused fatty liver by chylomicrons being diverted directly to the portal vein instead of being transported to the bloodstream via the lymphatic system 55 . It is however worth noting that liver steatosis is not a common observation among children with congenital intestinal lymphangiectasia, a disease associated with lymphatic damage 56 . Hepatic lipid accumulation in animal models of SM has so far been proposed to be related to mitochondrial dysfunction, peroxisomal damage 57,58 or choline deficiency 59 . However, in these animal studies, rodents were fed low-protein diets which did not induce oedema. ECM degradation, especially in the lymph, could be a contributing mechanism to liver fat accumulation in kwashiorkor, which is testable. Analysis of post-mortem portal vein triglyceride content using minimally invasive tissue sampling strategies could test this hypothesis. Alternatively, high content of triglycerides, apolipoprotein B48 and lymphocytes in ascitic fluids could further demonstrate the role of lymphatic degradation in kwashiorkor. However, ascites in kwashiorkor may already indicate a more advanced condition or an entirely different clinical picture, which could influence this observation.
Both LYVE1 and podoplanin (via colocalization with CD44) are major HA receptors 50,60 . Although we did not find significant association between kwashiorkor and HA in the discovery cohort, plasma albumin-normalized HA was higher in kwashiorkor in the validation cohort, indicating higher HA shedding. Observed differences between our cohorts could be because children in the discovery cohort were both severely undernourished and clinically ill at the time of sampling, whereas children in the validation cohort were still severely undernourished but severe infection had resolved, micronutrients, . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint protein and energy had been initiated, and they were tolerating therapeutic feeds, which is regarded as a sign of nutritional stabilisation.
Increased levels of plasma ECM were accompanied by increased plasma concentrations of MMP2, which is an active regulator of ECM remodelling. MMP2 is a 72 kDa type IV collagenase that is distributed in many tissues and associated with several serious diseases. Along with MMP9, it is also expressed in lymphatic endothelial cells 61 and plays a key role in lymphatic vessel formation 62 .
Oxidative stress is a strong activator of MMP2 63 , and reactive oxygen-nitrogen species-induced MMP2 activation has been shown to play major roles in various diseases such as cardiac injury 64 , hepatic fibrosis 65 and atherosclerosis 66 . Our findings therefore provide a plausible mechanism for previous data suggesting generalized loss of sGAG in kwashiorkor 24,67-70 . Our results suggest an MMP-induced degradation of the core ECM proteins that these sGAG are structurally linked to. We also observed that this disruption is not exclusive to sGAG-binding ECM but also to ECM proteins linked to non-sulphated GAGs, such as LYVE1 and podoplanin.
We did not find evidence of greater systemic inflammation in kwashiorkor compared to marasmus, when albumin was matched. This is surprising considering the well-described interaction among oxidative stress, inflammation and ECM remodelling 71 . It is thus plausible that children with kwashiorkor have a predisposition to a heightened ECM remodelling given the same inflammatory insult experienced by children with marasmus. Hence, examining variability in ECM remodelling-associated genes, such as MMPs, would be a promising next step to understand the aetiology of kwashiorkor. These results also highlight the role of non-nutritional factors such as (epi)genetic factors in the aetiology of kwashiorkor.
Although our results do not give a definite answer to what causes kwashiorkor, we reveal important findings on the pathophysiology of oedema, especially the role that albumin plays. This finding is not only relevant to SM but also to other oedematous diseases such as nephrotic syndrome and sepsis. We recognize that matching patients by admission serum albumin to specifically examine albuminindependent mechanisms associated with kwashiorkor precludes comparison with the full spectrum of . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint marasmus cases. Nonetheless, our results remained consistent in the validation cohort where we normalized by albumin rather than individually matched. A limitation of the study is the lack of data on the renal function, which has also been reported to be associated with LYVE1 72 , podoplanin 73 and lumican 74 . Although it is plausible that any renal dysfunction in kwashiorkor could be linked with the degradation of renal ECM proteins.
Given these findings, we propose that the mechanism for oedema in kwashiorkor involves both reduced serum albumin concentration and increased oxidative stress leading to heightened ECM degradation resulting in (1) reduction in interstitial integrity and (2) impaired lymphatic integrity causing poor interstitial fluid drainage. The ECM is a highly dynamic structure in which components are continuously synthesized, degraded and regenerated 75 . This potentially explains the resolution of oedema despite minimal increase in serum albumin concentrations. Hence, targeting restoration of ECM during treatment could help rapidly restore interstitial rigidity and lymphatic integrity allowing the drainage of interstitially displaced fluids back to circulation.

Overall study design
This study comprised of two separate nested case-control studies formed from a sub-selection of children with either kwashiorkor or marasmus from two clinical trials in Malawi and Kenya, as further described below. A hypothesis generating discovery cohort was used to explore albumin-independent mechanisms, which were then validated using the second cohort.

Study population and setting
Discovery cohort: The discovery cohort was nested within a randomised controlled trial (NCT02246296) that aimed at determining the effect of a lactose-free, low-carbohydrate F75 milk formulated to limit carbohydrate malabsorption, diarrhoea and refeeding syndrome among children hospitalized for complicated SM in Queen Elizabeth Central Hospital in Malawi, and Kilifi County . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Hospital and Coast General Teaching and Referral Hospital in Kenya 32 . The trial enrolled children aged 6 months to 13 years at admission to hospital if they had complicated SM, defined as: mid-upper arm circumference (MUAC) < 11.5 cm or weight-for-height Z score < −3 if younger than 5 years of age, BMI Z score < −3 if older than 5 years, or oedematous malnutrition at any age. The children were admitted to hospital because of medical complications or failed an appetite test (8/843, 0.9%) as defined by WHO guidelines 76 . Children were excluded if they had a known allergy to milk products or did not provide consent. The primary outcome of the trial was the time to initial stabilization, defined as having reached the 'transition' phase of treatment and switched to a standard higher caloric feed based on WHO guidelines. Biological samples for research including serum and plasma samples were collected upon admission but before randomization and stored at -80°C until analysis. For the trial, biochemical tests were performed to determine serum albumin concentration. Clinical findings were also recorded such as presence of shock, pneumonia, malaria, heart disease, cerebral palsy and diarrhoea, as well as breastfeeding. The trial recruited a total of 843 children of which 8.9% died prior to stabilization while another 6.2% died after the first stabilization 32 .
Validation cohort: The validation cohort was nested within a randomised controlled trial (NCT00934492) that tested the efficacy of daily co-trimoxazole prophylaxis in reducing post-discharge mortality among HIV-uninfected children aged 60 days to 59 months admitted to hospital and diagnosed with SM in four hospitals in Kenya (two rural hospitals in Kilifi and Malindi, and two urban hospitals in Mombasa and Nairobi)) 33 . Children were eligible for inclusion in the trial on the basis of the midupper-arm circumference (MUAC) measurements (<11·5 cm for children aged ≥6 months and <11·0 cm for infants aged 2-5 months) or presence of kwashiorkor; had a negative HIV rapid-antibody test; and had completed the stabilisation phase of treatment. Children were recruited into the trial for a median of 6 days from admission to the hospitals. Children were actively followed up for a total of 1 year, monthly in the first 6 months and every 2 months until month 12 for growth, readmission or death and traced at home if they defaulted. Samples were stored at -80°C until analysis.
Variables and data source/measurement . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint The presence of oedema, which was evaluated by trained research clinical staff. Kwashiorkor was diagnosed based on the presence of oedema regardless of concurrent wasting. Children without nutritional oedema and with either mid-upper arm circumference <11.5cm (or <11cm if age <6 months) or weight-for-length/height (WFL/H) < -3 were considered as marasmus.

Data sources and management
Clinical data was obtained from two independent clinical trials 32,33 , as above.

Study size
Of the 843 study children recruited in the F75 reformulation clinical trial, 46% (385) had grade one (+, n=67), grade two (++, n=113), and grade three (+++, n=29) nutritional oedema at admission. For this study, children with nutritional oedema (n=72; 19%) and with known serum albumin concentrations was selected as the discovery cohort. A total of 181 (21%) had missing albumin concentration at admission. Hence, for the discovery cohort, children with kwashiorkor were matched on serum albumin concentrations to those with marasmus (n=72; 16%). Sample size was limited by finding exact matches of serum albumin concentrations between SM phenotypes and hence all matched case-control pairs were included.
The sample size calculation for the validation cohort was based on the results of the discovery cohort.
Based on results for lumican (proteomics) and PC.ae.C34:3 (metabolomics), we calculated that a sample size of 40 per group would be sufficient to achieve >80% power at α = 0.05, accounting for stratification based on oedema severity and multiple testing. Subjects for the validation cohort were selected if they had achieved total nutritional and clinical rehabilitation (defined as having a MUAC >12.5 cm, absence of oedema and/or disease needing hospitalization) at day 60 post-hospital discharge. Selection was also limited to children who had sufficient plasma samples at enrolment and at month 2 of follow-up.
Kwashiorkor (n=40) were matched to marasmus (n=40) based on age, sex, and site of recruitment and randomisation arm in the trial.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Laboratory analyses
Untargeted plasma proteomics analysis Liquid chromatography tandem mass spectrometry plasma proteomics analysis was performed for the discovery cohort using the plasma samples collected at enrolment during admission. Briefly, plasma proteomics was performed using Tandem Mass Tag (TMT; Thermo Scientific) as described in our previous study 77 . Plasma samples were depleted of abundant proteins using spin columns (Thermo Scientific) then reduced and alkylated respectively with 40mM tris(2-carboxyethyl)phosphine and 80mM iodoacetamide. The proteins were subsequently precipitated using pre-chilled (-20°C) acetone followed by centrifugation at 8,000g for 10 min at 4°C. Precipitated proteins were then subjected to trypsin digestion (1:15 trypsin:sample ratio) and labelled with TMT 10plex kit (Thermo Scientific) according to manufacturer's instructions. The TMT-tagged peptides generated were then separated on the Dionex Ultimate 3000 nano-flow ultra-high-pressure liquid chromatography system (Thermo Scientific) with a 75 µm × 25 cm C18 reverse-phase analytical column (Thermo Scientific) at 40 °C.
Elution was carried out with mobile phase B (80% acetonitrile with 0.1% formic acid) gradient (2 to 35% B) over 310 min at a flow rate of 0.3 μl/min. The chromatographic outflow was connected to a Q Exactive Orbitrap Mass Spectrometer (Thermo Scientific) via a nano-electrospray ion source. Further details of the chromatographic and mass spectrometry parameters were as previously described by Njunge, et al. 77 .
Mass spectrometer raw files were analysed by MaxQuant software version 1.6.0.1 78 and peptide spectra were searched against the human Uniprot FASTA database using the Andromeda search engine 79 .
Further details on the proteomics dataset extraction and pre-processing were previously in Njunge, et al. 77 . A complete list of all annotated proteins is provided in Supplemental Table S1.
Targeted plasma metabolomics analysis . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  Table S1.

Targeted plasma analysis of endothelial glycocalyx components
Plasma levels of Syn1 and HA were performed using quantitative ELISA (Thermo Scientific, MD, USA) following manufacturer's instructions at admission for the discovery cohort, and at admission and 60 days post-discharge for the validation cohort.  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Data Preprocessing
For proteomics, columns containing the protein identifiers (IDs), protein names, gene names, and corrected reporter ion intensity in the protein group matrix file from MaxQuant were obtained. Data preprocessing included removing features with more the 20% missing values. The dataset was then log transformed and batch corrected using using ComBat function as implemented in sva R package 80 .
Missing values were then imputed using an imputation algorithm based on k-nearest neighbour.
For metabolomics, metabolites with more than 20% missing values in both SM phenotype groups were removed before analysis. Then, metabolites with concentrations below the limit of detection (LOD) for each metabolite were set as half the LOD. Lastly, only metabolites that have <30% coefficient of variation among the quality control samples were retained for further data analysis. Both proteomics and metabolomics data were log transformed prior to analysis.

Data analysis
Data analyses were all performed using R v 3.6 81 . Baseline patient characteristics are provided as either mean ± standard deviation (SD), median (25 and 75 th percentile) or proportions, as applicable. For the discovery cohort, difference in exposures (untargeted proteome and targeted metabolite levels, Syn1, HA, lumican, MMP2, TIMP1 and TIMP2, and a panel of 29 cytokines and chemokines) between marasmus and kwashiorkor were analysed using conditional logistic regression adjusted for age, sex, HIV status and recruitment site, to account for the sparseness introduced by exactly matching by serum albumin. To assess the association between increasing oedema severity and plasma/serum levels of glycocalyx components, individual protein and metabolite levels, we used an ordinal regression with age, sex, HIV status, recruitment site and serum albumin as additional covariates. High MUAC signifies the absence of wasting but also increases with oedema in kwashiorkor. Hence, no adjustment for MUAC was made in the models as this would obscure the interpretation of the results. Longitudinal analyses were performed using linear mixed models with the individual subjects set as random effect. Correction for multiple testing was performed using Benjamini-Hochberg false-discovery rate method 82 .
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint Multi-omics weighted co-expression network analysis (WCNA) was then performed using the WGCNA R package 83 . WCNA, is a data-driven network approach that is used to find clusters, known as modules, of tightly correlated features that can then be used to assess associations towards a specific outcome 83 .
Understanding which modules relate to clinical outcomes and determining pathways involving features that comprise the modules enables us to uncover molecular pathways that could be associated to the outcome of interest. After pre-processing and combining both metabolomics and proteomics datasets, a soft threshold for network construction needed to achieve a scale-free topography typical of biological networks (r² ≳ 0.8) 35 was determined. Scale-free topography for our data was achieved using β = 18 for an signed network. A Pearson correlation (sij) matrix was then generated between each multi-omics pairs

Data availability and access statement
The processed data and codes (STATA and R) will be deposited in the KEMRI-Wellcome data repository on the Harvard Dataverse under the Biosciences Dataverse subtheme (https://dataverse.harvard.edu/dataverse/kwtrp) and issued with Digital Object Identifiers (DOI) at the . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint time of deposit. The data will be titled "Albumin-dependent and independent mechanisms in the syndrome of kwashiorkor". The anonymised mass spectrometry raw files generated and analysed in the current study will be deposited to The ProteomeXchange Consortium: http://www.proteomexchange.org/ and assigned a unique identifier.

Acknowledgement
We thank the parents and guardians of the study participants who patiently participated in in both clinical trials in Malawi and Kenya. We acknowledge the enthusiastic work of our nurses, field workers, clinical and non-clinical staff who tirelessly collected the data and samples and provided administrative support to the project. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 1, 2021. ; https://doi.org/10.1101/2021.05.31.21257914 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.