Total serum N-glycans mark visceral leishmaniasis in human infections with Leishmania infantum

Summary Visceral leishmaniasis (VL) is a clinical form of leishmaniasis with high mortality rates when not treated. Diagnosis suffers from invasive techniques and sub-optimal sensitivities. The current (affordable) treatment with pentavalent antimony as advised by the WHO is possibly harmful to the patient. There is need for an improved diagnosis to prevent possibly unnecessary treatment. N-glycan analysis may aid in diagnosis. We evaluated the N-glycan profiles from active VL, asymptomatic infections (ASYMP) and controls from non-endemic (NC) and endemic (EC) areas. Active VL has a distinct N-glycome profile that associates with disease severity. Our study suggests that the observed glycan signatures could be a valuable additive to diagnosis and assist in identifying possible markers of disease and understanding the pathogenesis of VL. Further studies are warranted to assess a possible future role of blood glycome analysis in active VL diagnosis and should aim at disease specificity.


Visceral leishmaniasis (VL) has a distinct N-glycome profile
The N-glycome profile of VL reflects treatment response The N-glycome of asymptomatic infections doesn't differ from that of healthy controls N-glycome analysis has potential as an alternative to invasive diagnostic techniques

INTRODUCTION
Leishmaniasis is a major neglected tropical disease and a global public health problem since it is present in 98 countries of the world, with 700,000 to one million new cases every year and an estimated 350 million people at risk of infection ( 1,2 and https://www.who.int/news-room/fact-sheets/detail/leishmaniasis, accessed on Dec 2, 2022). Within the Americas the disease is present in 19 countries of which Brazil is the most affected country. The three main clinical forms are cutaneous (CL), mucosal (ML) and visceral leishmaniasis (VL). VL presents high mortality rates in patients being fatal in 95% of cases when not treated 3 and is the clinical form this study is focused on.
Leishmania infantum is the causative agent of VL in Brazil. 4 The disease was first recognized in the country in 1932 5 probably taken to the northeast region of Brazil by people or dogs from southern Europe or North Africa infected with the parasite. 6 The competent vector for L. infantum found in most countries in Latin America is the sandfly species Lutzomyia longipalpis and major reservoirs of L. infantum are dogs. 7,8 In addition, a role of asymptomatically infected humans as reservoirs for L. infantum has been suggested. 9,10 VL can be classified as an opportunistic infection because it is a co-infection in many HIV patients 11 and it is known that HIV infection massively amplifies the susceptibility as well as the severity of VL. 12 The susceptibility factors for VL include age -with children under one year old and adults above 50 years being the most affected 13 -as well as genetic background of the host, 14 nutritional status, 15 sex, 16 and immune suppression. 17 For asymptomatic infected individuals, it is difficult to predict whether and when the disease will become active, with environmental, parasitic and host-related factors playing a role. 18 The asymptomatic incubation period of VL has a variable duration and intermittent fever, malaise and shivering are included as early symptoms. Splenomegaly, accompanied or not by hepatomegaly, are symptoms manifested in overt disease. 19,20 The gold standard for diagnosis of VL is the parasitological exam with visualization of the amastigote, a motile, round and obligate form of the L. infantum, in biopsied material which should preferably be 1

The analytical pathway
A graphical representation of the analytical pathway followed is depicted in Figure 1. The serum and/or plasma N-glycomes from patients diagnosed with VL together with samples from control individuals were analyzed by mass spectrometry. Of the initial 756 collected samples, after spectra and analyte curation, a total of 661 serum samples characterized by 73 glycans were included in the study. Exclusions were based on parameters as described in the STAR Methods section. As described in this same section, a large subset of these samples (n = 544) was collected between 2012 and 2013 and used as the discovery set. The remaining samples were collected in 2018 and used as a validation set. Also, 76  iScience Article various timepoints after treatment were included in a longitudinal study. The characteristics of the participants cohort are summarized in Table 1.
For repeatability testing, 27 pooled serum samples and 67 VisuCon plasma samples spread over ten sample plates were analyzed regarding the 25 most abundant glycans, revealing an average overall CV of 17.1% ( Figure S1A).
To support the validity of the data and because the sex distribution of our samples was unbalanced between the groups, trends of glycans known to be influenced by sex 39,42 were evaluated and proven to be in accordance with the expectations ( Figure S1B).
The total serum N-glycome can distinguish VL from individuals without symptoms Principal Component Analysis (PCA) was performed to explore the Total Serum N-Glycome (TSNG) of asymptomatic, VL and control groups. The summary of fit is shown in Figure S2A, showing that the first five principal components already describe a large portion of the variance in the data. The PCA showed that TSNG can distinguish VL from the other three groups of interest. Asymptomatic patients largely overlapped with healthy individuals. The slightly disbalanced sex distribution was found to be no major confounder of the disease signature. Also age, another factor that is known to be of influence in the expression of certain glycans, 43 did not demonstrate any trends that would indicate confounding (Figure S2) as age was distributed evenly over the biological groups.
Sialylation of diantennary structures determines the first dimension of the PCA ( Figure S2B), but its influence on disease remains unclear, since most of the discrimination in the PCA score plot is visible in the second dimension. Tri-and tetra-antennary structures drive the separation in the second dimension with fucosylated glycans indicating VL patients, whereas afucosylated glycans marked the non-manifesting individuals (NC, EC, and ASYMP). To investigate the differences between the VL patients and the other groups in more detail, the glycans were individually tested with a Kruskal-Wallis test (Figures 2, S3, and  Table S3A) showing consistent signatures for males and females.

Few glycan differences were observed in onset of symptoms
To explore the differences in the four groups of interest in more detail by means of univariate analysis, the sample cohort was split into a discovery and a validation set (see STAR Methods).
To obtain more insights into the possible onset of symptoms we first focused on the two healthy control groups using a Wilcoxon rank-sum test. No consistent, replicated differences were found between the endemic and non-endemic healthy control groups.
Three glycans were different between non-endemic controls and asymptomatic for both discovery and validation set. They all belong to the diantennary, bisected glycans. Of interest, no replicated glycosylation differences were observed when comparing endemic controls with asymptomatic persons. The observed differences in glycans are reflected in the significantly different glycosylation traits of hybrid, mono-and diantennary glycans (Tables S3C and S3D).
Distinct N-glycome signature changes in active visceral leishmaniasis disease The majority of differences in this dataset are observed between VL-patients and the other groups (without clinical manifestations) (Figure 3; Tables S3E and S3F). Because the differences between non-endemic and endemic controls are not disease-related and the endemic controls are more closely related to the asymptomatic cases and to active VL compared to the non-endemic controls this latter group was disregarded in subsequent analyses. The Wilcoxon tests revealed 44 replicated glycans and glycosylation traits as After collection of serum (or plasma) from the participants in this study the glycans were released from the proteins, derivatized on the sialic acids to accommodate glycan isomer differentiation, purified using cotton HILIC and measured with a mass spectrometer. The data were processed using MassyTools software and data curation was performed to eliminate low quality data. A PCA analysis was performed on the full dataset, after which the data was split into a discovery and a validation set as described in the STAR Methods section. After analyte selection using a Wilcoxon rank-sum test, a model was built that was able to distinguish Visceral Leishmaniasis (VL) patients from Healthy endemic control individuals (EC). Data are represented both as box plots, which show the 0th and 100th percentiles, the sample median, and the first and third quartiles, and dots representing values for individuals. iScience Article significantly and sufficiently different between either EC or ASYMP and VL, in both the discovery and validation datasets (Table 2). Fucosylation was increased in samples from VL patients (represented by e.g., Figures 2A and S3 trait CF) compared to the other groups, mainly driven by core fucosylation, although antenna fucosylation also slightly increased on tri-and tetra-antennary glycans ( Figure S3 traits A3F, A3Fa, A4F and A4Fa). Examination of the results in more detail showed that bisection of diantennary structures require the use of additional therapy to treat the patient, which include antibiotics or blood products, with hemorrhage and increased risk of death, according to laboratory data. f NR means individuals whose sex or age has not been registered. g The asymptomatic individuals were identified by positive tests according to literature. 26,38 The same methods were applied to identify controls as. h Healthy individuals from an endemic area. i Healthy individuals from a non-endemic area through negative result.

N-glycome is not influenced by the severity of VL
To better understand underlying processes of active VL and progression of disease, we divided the VL group of the combined discovery and validation sets in two sub-groups: uncomplicated (U-VL) and complicated with clinical manifestations (C-VL). We did so, because the validation set only contained two complicated cases, which makes the set unsuitable for validation. The latter group included patients with additional therapy, hemorrhage or increased risk of death. The data as depicted in the Volcano Plot ( Figure S4C) indicate that none of the glycans or glycosylation traits are significantly different between these groups (Table S3G). Even in the case of experiment-wise multiple testing correction (a = 0.05/ 226 = 2.2124e-4) instead of study-wide multiple testing correction, only one glycan (H5N5F0E1L0) would  iScience Article be significantly different between these groups with an absolute log 2 (fold-change) of at least 0.3 (decrease in this case; À0.515) and possibly another one with an absolute log 2 (fold-change) of just below 0.3 (H6N5F1E1L0; À0.298). This result is corroborated by ROC analysis ( Figure S5C). The obtained single parameter model (H6N5F1E1L0) was poorly predictive, with an AUC of 0.60. Of interest, H5N5F0E1L0 was not selected as a discriminator in this ROC analysis.

N-glycome reverts to base levels after VL treatment
To evaluate the glycosylation changes on treatment and recovery from disease we assessed the difference in glycan levels in patients' plasma samples at day 0, 5, 90 and 180 after onset of treatment. The declining number of samples over time was caused by patients unfortunately not returning to the hospital at the indicated timepoint. Therefore, the number of samples was too low to perform statistical analysis. However, several trends were observed. On treatment, most of the glycosylation patterns returned to the situation as observed in healthy controls. Fucosylation in general ( Figures 5A, 5B, 5C, and 5F) and sialylation on the tetra-antennary glycans ( Figure 5G) decreased over time whereas a2,3and a2,6-linked sialylation on diantennary glycans increased ( Figures 5D and 5E). Bisection of fucosylated diantennary structures showed a decrease in VL patients, increasing over time on recovery.

DISCUSSION
TSNG analysis provided a strong signature that allowed to differentiate VL from both asymptomatic and endemic controls with very high sensitivity and specificity (AUCs of 0.88 and 0.92 for ASYMP and EC, respectively). Given the invasive nature of current diagnostic approaches for VL, blood glycome analysis may provide a promising approach for further evaluation and possible implementation in VL diagnosis. Blood glycome analysis can be performed from dried blood spots and provides a glycomic signature that is very similar to the one obtained by glycomic analysis of serum or plasma. 44 Hence, further research into dried bloodspot glycome analysis from clinical cohorts of active VL and conditions with similar clinical presentation is warranted to evaluate its potential for complementing or replacing current invasive diagnostic procedures. In addition, for clinical implementation, blood glycome analysis will have to be transferred from high-end mass spectrometry to e.g., a microtiter plate assay. 31,41,45 In our study we included both healthy controls from endemic as well as from non-endemic areas. The analysis indicated differences between these two groups that pointed toward the involvement of A B Figure 4. VL can be detected using a panel of four glycosylation traits ROC analysis showing models build on traits as determined by automated feature selection using the SES algorithm of the MXM R-package. A 10-fold crossvalidation procedure was used to assess the strength of the prediction models. The models were then validated using an independent dataset. The relevant analysis values are given in Table S4. iScience Article immunoglobulins. However, most of these differences could not be validated. As these differences are clearly not VL-induced, they might be caused by a difference in lifestyle and/or environment, as previously demonstrated. 46 People living in endemic areas might experience higher exposure to pathogens in general causing higher levels of proteins involved in protection against them. Because the endemic controls are more closely related to asymptomatic and active VL cases and to exclude bias introduced by demographic differences, we decided to exclude the non-endemic controls from further comparisons. iScience Article Following this reasoning we would like to stress that the discussion described hereafter should be considered as specific for the population under study as well as the parasite species.
Visceral leishmaniasis, when in its chronic course, presents clinical signs like those found in systemic erythematous lupus, 47 such as enlarged liver and spleen, and pancytopenia 48 and the clinical status of the disease is directly associated with a systemic inflammatory response. 49,50 It is known that the progression of liver diseases is associated with specific glycosylation alterations in serum proteins. For instance, in hepatocellular carcinoma, a disease with high risk for hepatic injury and inflammation, there is an increase in levels of core fucosylation. 51 The serum N-glycome from patients infected with Pseudomonas aeruginosa, an opportunistic gram-negative pathogen closely associated with cystic fibrosis affecting lung and liver, shows increased levels of core fucosylation (as well as decreased levels of sialylation, which will be discussed later). 52 Core fucosylation levels are also increased in advanced stages of pancreatic cancer, however the authors of that study suggested that the biological function seemed to be related to the disease itself and not to the inflammation. 53 Similar glycan profiles have been described for (auto)inflammatory diseases such as rheumatoid arthritis 54 and Crohn's disease, 30,55 in which fucosylation of a tri-antennary structure, indicated as an inflammatory marker, was elevated. To this extent, our results of increased fucosylation, both on the core and on the antennas, corroborate with all these observations indicating a large involvement of the inflammatory pathways. As has been shown previously, L. infantum leads to an elevation of circulating immunoglobulins in VL patients. 56,57 Of interest, patients with dengue virus 58 and COVID-19 59 had reduced levels of core fucosylation on disease-specific IgGs. This reduction was specifically correlated with enveloped viruses. The biological function of core fucose reduction is associated with an excessive activation of FcgRIIIa, which leads to an overreaction by the immune cells, increased antibody-dependent cellular cytotoxicity (ADCC) and cytokine storms. 60 As we observed the opposite response with regard to fucosylation in TSNG, and a similar increase in fucosylation was observed in total IgG of VL patients, 38 it would be interesting to investigate glycan profiles of Leishmania-specific IgG, especially of severe cases because these proinflammatory responses are also observed in patients presenting with severe VL 49 and they are possibly associated with a decreased fucosylation in specific IgGs, thus explaining this phenomenon. It is noteworthy that, compared to uninfected controls or asymptomatic individuals, in patients presenting with active VL the transcriptional profiles of genes expressed in peripheral blood leukocytes annotated into processes of leukocyte chemotaxis, neutrophil activation and B cell receptor activation were down-regulated whereas, on the other hand, up-regulated genes were mainly enriched into the network process of NK cell cytotoxicity, 61 suggesting the involvement of FcgRIIIa.
Besides increased fucosylation, other changes in inflammatory glycosylation patterns include changes in bisection, galactosylation and sialylation. To start with bisection, our results are in accordance with a study on thrombocytopenia where treatment of the disease resulted in an increase in bisection, suggesting reduced bisection on development of symptoms. 62 On the other hand, bisection was significantly higher in meningitis patients presenting with the most severe clinical outcomes. 63 Bisection also discriminates between pathogens that cause bacteremia, 52 possibly reflecting differences in their microbial-associated molecular patterns and respective strengths in activating innate immune signaling pathways.
With regard to galactosylation and sialylation, our results are in line with previous findings in other diseases with inflammatory profiles. A decrease of total IgG1 sialylation, galactosylation and bisection and increase of fucosylation was also observed in relapsing vasculitis patients with higher anti-neutrophil cytoplasmic antibody levels. 64 The previous work on the glycosylation patterns of IgG molecules of VL patients also showed a reduction of bisection, sialylation and galactosylation of, especially the Fc portion of IgG subclasses in VL patients and asymptomatic individuals, although the decrease in this last group was less pronounced, consistent with our findings for TSNG. Of interest, the decrease of galactosylation observed in our data was mainly caused by the effect on diantennary glycans and the decrease of sialylation as a result of decreased a2,6-linked sialic acids. These results are in accordance with the fact that Immunoglobulins comprise a large group of proteins in the human body and the Fc portion of IgGs carries diantennary glycans. 38 When present, sialylation on IgG is almost exclusively a2,6-linked. Both observations indicate a strong relation with IgG glycosylation. 65 Furthermore, a connection has been proposed between decreased galactosylation and both age and inflammatory diseases. 36 Because the mean age of the group with complicated cases is relatively low (23 years), and the proinflammatory effect of IgG is associated with an increase in agalactosylated structures, 66 we hypothesize that the observed change in galactosylation is a result of the inflammatory profile of VL.

ll
OPEN ACCESS iScience 26, 107021, July 21, 2023 iScience Article The above-described results all point toward a large contribution of immunoglobulins and their glycosylation to blood glycosylation signatures. With regard to hypergammaglobulinemia, one may speculate that specific antibody glycoforms may contribute to inflammation and regulate immunity (in much the same way that overall levels of serum cytokines determine immune profiles and clinical outcomes). Furthermore, there was indication in the results for some disease-specific changes as well. An increase in antennarity has been observed, especially the tetra-antennary structures. In contradiction with the general decrease in galactosylation and sialylation, these glycosylation traits showed an increase on tetra-antennary glycans as well, which for sialylation was not linkage specific. The question remains if a change in the number of antennae is indicative of a disease-specific phenomenon, as such processes also have been identified in patients exhibiting inflammation of different origin. For instance, after inflammation-mediated processing, the carbohydrate structures of various inflammatory glycoproteins shift from high-mannose type to di-, triand tetra-antennary glycans. 67 In addition, it has been reported that the glycans on alpha-1-acid glycoprotein (AGP), one of the major acute phase proteins in humans, are modified during an acute phase response, from bito tri-and tetra-antennary branching with increased fucosylation and sialylation. Similar modification of AGP with increased branching of the glycan structures has also been reported in some inflammatory diseases such as asthma and rheumatoid arthritis. 68 Of interest, in patients with liver cirrhosis AGP fucosylation was higher. An increase in the levels of both a2,3and a2,6-sialylation has been reported as general inflammatory markers shared by many diseases. 69 For example, in autoimmune diseases, inflammatory bowel diseases and acute inflammation the higher branching with increased sialylation has been described. 30 The presence of sialic acid has been reported both in the promastigote and in the amastigote form of Leishmania. However, the detection of these sugar molecules in the parasite is still questionable, since the biosynthetic machinery of the parasite to acquire sialic acid is not well understood, and most of the studies are performed in Leishmania donovani, without considering the fact that Leishmania species are heterogeneous with regard to molecules and virulence. Despite that, it has been shown that L. donovani can adsorb sialoglycans under various stimuli in the environment to possibly compensate for deficiency of sialic acid 70 and, as reviewed by Cavalcante et al. and Colli, to evade host immunity. 71,72 Such mechanisms may thus explain the decrease in sialic acid-containing glycans seen on proteins in patients presenting with VL. A similar hypothesis has been stated for individuals infected with P. aeruginosa. 73 Here, the bacterium was shown to express a sialidase that cleaves sialic acid residues from the host's glycoproteins thereby facilitating adherence to host cells, which was inhibited by sialylated glycans on the host cell surface. Khatua et al. also reported on P. aeruginosa capable of absorbing host a2,3and a2,6-sialoglycoproteins, thereby reducing neutrophil activity and increasing survival of P. aeruginosa. 74 Our results also may finally provide the basis of a mechanism to explain earlier findings, that sera from VL patients contain factors that enhance three-to 5-fold the lysis of amastigotes of L. donovani by the alternative pathway of complement when compared to normal human serum. 75 Another of the observations made at the onset of leishmaniasis involves ending the lytic route. 76 The parasite-specific IgG antibody induces lysis of Leishmania (and other trypanosomatids). 77 Sialic acids are crucial for the protection of the parasite against attack by the host complement system, 78 similar to what occurs, for instance, with Neisseria gonorrhoeae, which is protected from activated mannose-binding lectin (MBL) complement and death by sialylation both in epimastigote and trypomastigotes forms of the parasite, preventing the binding of lytic anti-galactose antibodies. 79 It is interesting to note that if the success of Leishmania infection in the cell depends on sialic acid, the greater the parasite's ability to adsorb it, the greater the parasite's proliferation and, probably, a greater damage to the host. Unfortunately, so far, the deeper analysis of glycosylation in relation to severity of the infection did not elucidate such close relationships and in this context, antigenspecific antibodies must be examined. Further possible mechanisms that might be the basis of the associations of TSNG profiles with the outcomes of infections with L. infantum seen in this study are the resulting interactions of L. infantum-specific antibodies with the several receptors on phagocytes that mediate this process (reviewed in 80 ), as well as the pathogen escape mechanisms that are possibly mediated by the receptors for IgG Fcs present on many pathogens, including trypanosomatid pathogens. 81 Although not significant, the two most promising glycans with a relation toward disease severity were already slightly reduced in asymptomatic cases and significantly reduced in VL patients overall. If we plot the two severity groups separately next to the asymptomatic and control groups (Figures S5A and  S5B)  iScience Article and the reduction of H6N5F1E1L0 can be almost entirely contributed to severe disease. The mean for this glycan is similar between the asymptomatic and the uncomplicated group, already slightly reduced compared to the healthy controls and lowest in the complicated VL group. Therefore, we dare to hypothesize that these glycans are influenced by the severity of the disease. Unfortunately, this change is not sufficiently pronounced to be used as a predictive marker as is shown with the ROC analysis. The biological meaning of this still remains unclear. Furthermore, it is notable that all the changes described above seem to revert on treatment.
Our research was driven by two main questions. One of these questions involved the predictability or early diagnosis of VL infection and severity. Early detection in the form of asymptomatic cases appears not to be possible using the methods described herein as (univariate) analysis did not yield enough consistent and replicated differences. Nonetheless, it would be insightful to undertake a longitudinal study in an endemic area that follows individuals bearing asymptomatic infections with L. infantum.
Our second question addressed the possibility whether the glycan analysis had potential for diagnosis as a substitute for the current invasive techniques. ROC analysis revealed that good models could be built for the discrimination of VL cases against both healthy controls and asymptomatic cases. In both cases, this resulted in a four-parameter model. Of interest, two parameters are shared between the models. Because the model comparing VL against healthy controls has the most logical medical meaning when used as a diagnostic tool, we feel that the four parameter EC-VL model (A2S0G + H5N4F0E0L0 + H5N4F0E0L2 + H7N6F2E1L2) is the final one, although the ASYMP-VL model performed slightly better in the validation set. This might be caused by a different sample distribution or the different diagnostic standard. Ongoing investigation using new samples will be required to enforce our model. Glycan analysis might be a good replacement of current methods to detect VL, because it is less invasive for the patient. However, the question remains if the model is disease-specific or, in view of the findings discussed above, a representation of general disease phenomena. Therefore, its diagnostic potential should be confirmed by cross validation with more samples from patients presenting with VL and relevant clinical controls (e.g., systemic inflammatory diseases that affect the liver, such as lupus, 47 and neoplasias involving bone marrow such as multiple myeloma, 82 etc.). It is worth mentioning at this point that VL can be confused with lupus and lymphoma and this can be fatal for the VL patient if not treated promptly because of this confusion. These appropriate controls should also include patients free from VL but suffering from diseases that can occur as concomitant conditions with VL, such as HIV, leprosy and Chagas Disease to exclude cross-reactivity; more research is needed in which multiple diseases involving similar phenotypes are compared to each other with respect to predictability.
It is also relevant to highlight that even after successful treatment and in case of immunosuppression, recurrence of VL may occur if there are viable parasites. 83 Besides that, there are VL cases with presence of antibodies beyond cure and there are some patients who are not able to produce sufficient antibodies, thus limiting the diagnosis of relapse or the prediction of cure by antibody-based tests such as rK39 rapid strip immunochromatographic test and ELISA. 24 In this sense, our results indicate that the glycomic signatures reflect treatment response and Total Plasma N-Glycome (TPNG) analysis (or TSNG analysis) could be used to monitor treatment success and eventually detect relapses. Finally, VL caused by L. infantum is a zoonosis with dogs being the reservoir. However, although there is no scientific evidence supporting culling of seropositive dogs to reduce the incidence of VL, asymptomatic dogs present low parasitism whereas symptomatic dogs are associated with high parasite load in various tissues such as skin, bone marrow and spleen. 84,85 The diagnosis in dogs faces the same issues as it does in humans, and it may also be worth examining TSNGs in this host to better define the de facto animal reservoirs for humans in this species.
In conclusion, TSNG could become a tool for detecting VL with high sensitivity and specificity, because the distinction of active VL from healthy controls or asymptomatic cases was clear, whereas between complicated and uncomplicated VL the TSNG signature appears to be largely the same. As for the individuals presenting with asymptomatic infections with L. infantum, although some structures seem to be slightly decreased in this group (ASYMP) compared to the controls (EC), their change has not sufficient statistical power to be used as a biological marker of infection nor were indications present of a subgroup at risk to evolve to VL. The changes in the N-glycome signature are associated with increased core-and antenna fucosylation and reduction of bisection, galactosylation and sialylation of especially diantennary glycans which has already been described for other diseases with an inflammatory profile, mainly the ones ll OPEN ACCESS iScience 26, 107021, July 21, 2023 iScience Article associated with liver injury. In addition, the increase in antennas as well as the increase in galactosylation and sialylation of these larger glycans may be associated with inflammation in active VL. Therewith, the overall glycan profile of VL has an inflammatory signature, which aids in understanding the pathogenesis of the disease, indicating a potential target for treatment and relapse detection if better explored. Until now, a disease specific signature could not be identified, which requires additional research.

Limitations of the study
It is noteworthy that infections were characterized with different strategies for the 2013 discovery and 2018 validation sets because of a change in regulatory standards: the Montenegro skin test reagent employed to elicit parasite-specific delayed hypersensitivity skin reactions, one of the approaches for diagnosis of infections with L. infantum, was discontinued after 2013. This required the development of a new assay based on the release of select cytokines by blood leukocytes. 26 The difference in assignment strategies may have caused a difference in assignment accuracy, which may in part explain why the validation outperformed the discovery. A second contribution to this outperformance may be because of differences in the sample distribution for EC versus VL in the discovery set (2:3) versus validation set (3:2). This reflects, however, the expected fluctuations seen in the incidence of VL by public health surveillance. 86 Furthermore, although the observed differences between VL and healthy or asymptomatic controls are striking, they do not elucidate disease specificity. To that purpose, further comparison with unrelated diseases with similar phenotypes or regularly observed co-infections is required.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:  The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We would like to thank all volunteers and study participants for providing body material. Furthermore, we thank dr. Oleg A. Mayboroda and dr. Rico Derks for their help and advice in statistical analysis.

Materials availability
This study did not generate new unique reagents.
Data and code availability d Raw mass spectrometry data in the form of .xy files have been deposited at GlycoPost 87 and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. This repository also contains the relevant metadata as well as the final curated data (Table S5) used in this manuscript.
d This paper does not report original code.
d Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS
This work was limited to human subjects. The total number of 661 participants was divided into a 338:264 ratio female:male (for 59 participants sex was not registered). Throughout the text, when sex is mentioned, the term sex assigned at birth is meant. Age ranged from 2 to 86 years. All participants originated from the Teresina or Aracajú region of Brazil. Other racial or ethnic information was not registered. More details are presented in Table 1. All available metadata per patient is also included in  38 or the production of cytokine profiles upon stimulation of peripheral blood leukocytes with an antigen extract prepared from promastigotes of L. infantum. 26 Due to a change in regulatory standards the Montenegro skin test was no longer permitted after 2013 by the Brazilian Health Ministry and was therefore only applied to the discovery set.
In this study we assessed the categories of clinical severity of 131 patients, which were classified in uncomplicated (U-VL: n=49) and complicated with required additional therapy or with hemorrhage and increased risk of death (C-VL: n=82). iScience Article and data quality assessment by MassyTools software version 1.0.2-alpha build 180703b 90 The extraction window ranged from 0.00719 to 0.08921 according to the formula '(0.00003 * "m/z") -0.02690', in order to accommodate for the increasing peak width with larger m/z values. The formula was obtained from the linear trendline derived from manual peak width measurements (full width at half maximum). To exclude poor quality spectra, the total intensity of a spectrum, the fraction of analyte area (minus background area) above signal to noise (S/N) 9, and the fraction of spectrum in analytes should be higher than the mean of these parameters minus three times the standard deviation. Each of these parameters was evaluated per biological group. The analyte curation was performed based on the following quality criteria: S/N should be greater than 9, isotopic pattern quality (IPQ) less than 0.2 and absolute PPM-error lower than 10 for at least 25% of all spectra per biological group. Areas of all curated glycans were normalized to the sum of these areas per spectrum and R-Studio software was used to calculate glycosylation traits (Table S2). 35

QUANTIFICATION AND STATISTICAL ANALYSIS
Analyses were performed in the R programming language version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria) and R-Studio software version 1.4.1106 (RStudio, Boston, MA). 91 To monitor the quality of the sample preparation, repeatability was tested using the Visucon standards and pooled serum samples, by analyzing the 25 most abundant glycans.
Principal Component Analysis (PCA) on the full dataset was performed to investigate if the four distinct groups of interest could be distinguished based on the respective N-glycome signatures. To do so, values were log 10 transformed and UV-scaled before PCA analysis was applied. To further evaluate the differences between the groups, the means of each individual glycan and glycosylation trait were compared using a Kruskal-Wallis test with, in case of significance, an additional post-hoc Dunn's test. The biological significance was then further evaluated by investigating the fold change in a group-by-group comparison using a Wilcoxon rank sum test on the discovery set and illustrated in a Volcano Plot. This same test was also applied to a severity group comparison (U-VL vs. C-VL). In significance testing a study-wide significance was set at a = 5.0e-5. This value was obtained by using a default a of 0.05, Bonferroni corrected by the study-wide number of tests performed (73 glycans + 153 glycosylation traits times one Kruskal-Wallis test plus four Wilcoxon tests per trait = 1130 tests: 0.05/1130 = 4.4248e-5 z 5.0e-5). The log 2 (fold-change) cut-off for the Volcano Plots was set at 0.3. Then these tests were applied to the validation set using an alpha of 0.05. Glycans and traits matching those in the discovery set with both significance and sufficient fold change were used as input for parameter selection in Receiver Operating Characteristics (ROC) analysis. Such analysis was performed in a group-by-group comparison for the comparisons EC-VL and ASYMP-VL on the discovery set and for the severity groups Uncomplicated -Complicated on the total set, since the validation set would only contain two complicated cases. Feature selection was achieved using the SES algorithm of the MXM R-package 92 applied to log 10 -normalized and scaled data on the remaining glycans and traits from the Wilcoxon tests. This algorithm could in theory output multiple sets of covariates for a single comparison. For each of these sets of covariates a general logistic model was calculated and the one with the lowest p-value for a Chi-square test was used in a ten-fold cross validation procedure. In each iteration of the cross-validation cycle the SES algorithm was applied again with the same selection procedure for covariates. From this cross-validation procedure, the Area Under the Curve (AUC) was calculated. As an independent validation the resulting models were applied to the validation set and a ROC curve was created.
The number of samples used for the follow-up after treatment was too low for statistical analysis. Therefore, only observed trends are discussed.