Novel urinary glycan profiling by lectin array serves as the biomarkers for predicting renal prognosis in patients with IgA nephropathy

In IgA nephropathy (IgAN), IgA1 molecules are characterized by galactose deficiency in O-glycans. Here, we investigated the association between urinary glycosylation profile measured by 45 lectins at baseline and renal prognosis in 142 patients with IgAN. The primary outcome was estimated glomerular filtration rate (eGFR) decline (> 4 mL/min/1.73 m2/year), or eGFR ≥ 30% decline from baseline, or initiation of renal replacement therapies within 3 years. During follow-up (3.4 years, median), 26 patients reached the renal outcome (Group P), while 116 patients were with good renal outcome (Group G). Multivariate logistic regression analyses revealed that lectin binding signals of Erythrina cristagalli lectin (ECA) (odds ratio [OR] 2.84, 95% confidence interval [CI] 1.11–7.28) and Narcissus pseudonarcissus lectin (NPA) (OR 2.32, 95% CI 1.11–4.85) adjusted by age, sex, eGFR, and urinary protein were significantly associated with the outcome, and they recognize Gal(β1-4)GlcNAc and high-mannose including Man(α1-6)Man, respectively. The addition of two lectin-binding glycan signals to the interstitial fibrosis/tubular atrophy score further improved the model fitness (Akaike’s information criterion) and incremental predictive abilities (c-index, net reclassification improvement, and integrated discrimination improvement). Urinary N-glycan profiling by lectin array is useful in the prediction of IgAN prognosis, since ECA and NPA recognize the intermediate glycans during N-glycosylation of various glycoproteins.


Results
Patient characteristics. Among 157 patients diagnosed with isolated IgAN who received a renal biopsy from December 2010 to August 2017 at Okayama University Hospital, 142 patients were enrolled in the current study ( Supplementary Fig. 1). The baseline characteristics of the patients at the time of the renal biopsy are shown in Table 1. The patients were 42.7 ± 16.3 years old, and 48% men. The mean baseline eGFR was 70.6 ± 25.9 mL/ min/1.73 m 2 , and the median 24-h urinary protein (UP) was 0.73 g/day (interquartile range [IQR] 0.27-1.53). The primary outcome was defined as an estimated glomerular filtration rate (eGFR) decline (> 4 mL/min/1.73 m 2 /year), or eGFR ≥ 30% decline from baseline, or initiation of renal replacement therapies within 3 years. During a median follow-up period of 3.4 years, 26 patients reached the renal outcome (Group P), while 116 patients were with good renal outcome (Group G). The systolic blood pressure (SBP), serum IgA levels, and UP were significantly higher in Group P (n = 26) than Group G (n = 116). The percentage of treatments, such as tonsillectomy and/or steroid therapy, was not significantly different between Group P and G. In addition, there was no significant difference in the use of antihypertensive agents, such as angiotensin converting enzyme inhibitor (ACE-I) or angiotensin receptor blocker (ARB) and calcium channel blocker, between two groups. At the final follow-up, UP was significantly reduced, and the patients treated with ACE-I or ARB was increased compared with baseline in both Group P and G (Supplementary Table 1). Furthermore, SBP, diastolic blood pressure (DBP) and mean arterial pressure (MAP) demonstrated no significant differences between 2 groups at the end of observation (Supplementary Table 2).
Relationship between the renal outcome and lectin binding signals. The median follow-up period was 3.4 years (IQR 2.2-5.2 years). The data of net glycan intensity (Net-I) in Group P and G at baseline are shown in Supplementary Fig. 2. The lectin signals were generally higher in Group P versus Group G and the urinary protein excretions were higher in Group P versus Group G (Table 1). Cy3 fluorescent is labelled to amine-containing proteins and the background signals of albumin lacking glycosylation are the critical concerns regards the specificity of lectin array. Actually, Net-I in each lectin demonstrated significant correlation with urinary protein concentrations ( Supplementary Fig. 3). However, the correlation matrix among 45 lectin binding signals demonstrated that r values between lectins with similar glycan recognition are very high, while r values between lectins with distinct glycan specificity are very low or even minus ( Supplementary Fig. 4). These data supported the elimination of artifacts and specificity of urinary lectin array.
The odds ratios (ORs) for a poor renal outcome by 45 lectin binding signals from urine samples are shown in Fig. 1, and the reported glycan structures specific to each lectin are shown in Supplementary Table 3 (Tables 2, 3). The inclusion of variables, i.e. IgA, SBP or T score into the models, ECA and NPA remained statistically significant. In the stepwise models, age, eGFR, and ECA/NPA were selected as statistically significant independent variables (Tables 2, 3). In another multivariate logistic regression models, we employed statistically significant parameters in univariate analyses, such as ECA/NPA, age, UP, IgA, SBP and T score, Influence of pathological scoring, lectin binding signals and steroid use on the renal outcome. The Oxford classification and pathological grading are shown in Table 4. There were no statistically significant differences in most of the pathological findings between Group P and G, whereas T score of Oxford classification was significantly higher in Group P than in Group G (P = 0.04). The classification of pathological features and their renal outcomes are shown in Table 5. Although only 11 patients were in T2 category, T2 score of Oxford classification and percentage of interstitial fibrosis/tubular atrophy (IFTA) were significantly related to the renal outcome in both univariate and multivariate models. Next, we investigated the correlation of ECA and NPA with pathological parameters. ECA signals demonstrated mild correlation with T score (r = 0.25, P < 0.01), cellular crescent (r = 0.22, P = 0.01), global sclerosis (r = 0.22, P = 0.01), and IFTA (r = 0.21, P = 0.01), while NPA signals also revealed mild correlation with the T score (r = 0.32, P < 0.01), cellular crescent (r = 0.36, P < 0.01), and IFTA (r = 0.27, P < 0.01) (Supplementary Table 6). The comparisons of ORs in the groups stratified according to ECA/NPA signals and T score are shown in Fig. 2. In ECA high and NPA high 2-quantile groups, the elevation of risks for poor renal outcome were prominent in patients with severe interstitial disease (T1/T2). We further analyzed the association between steroid therapy and the renal outcome in subgroups stratified by MEST-C scores, including mesangial hypercellularity (M), segmental sclerosis (S), interstitial fibrosis/tubular atrophy (T) lesions, and crescents (C) (Supplementary Table 7). As a result, the patients with cellular crescent and adhesion were more likely to receive steroid therapy than those without cellular crescent (p = 0.01) or adhesion (p = 0.03).
In contrast, the patients with higher T scores were more likely to avoid steroid therapy (p < 0.01).
Incremental predictive power of urinary glycan levels of ECA and NPA, plus T score. The Akaike information criterion (AIC) for evaluating the model fitting, concordance index (C-index), category- www.nature.com/scientificreports/ free net reclassification improvement (NRI), and integrated discrimination improvement (IDI) for predicting the primary renal outcome at the median follow-up time (3 years) obtained by adding the ECA, NPA, T score, and their combinations are summarized in Table 6. Adding the ECA, NPA, or T score to the multivariate model displayed improved the models, as shown by a decreased AIC and increased NRI. However, the addition of single parameters did not improve other model fitting indexes such as the C-index and IDI. Next, we investigated the effects of various combination of T score, ECA, and NPA signals. The combination of 2 parameters improved

Discussion
Glycans play pivotal roles in various physiological and pathological processes such as development, inflammation, autoimmune, hormone action, cell adhesion, and cancer [20][21][22] . In the current investigation, we firstly demonstrated that urinary excretion of glycans originated from the N-glycosylation process was tightly associated with the renal prognosis of IgAN. We found that the urinary excreted levels of glycans binding to ECA and NPA were significantly higher in IgAN patients with a poor renal outcome (Group P). Gal(β1-4)GlcNAc bound  24 . Although all 45 lectins including ECA and NPA recognize specific sugar structures, any protein carriers with specific sugar structures could be detected by lectin array systems and lectin signals are not confined to specific protein carrier, such as IgA. In organelles, N-glycosylation process begins in the endoplasmic reticulum (ER), and the complex-and hybrid-type glycans are synthesized from the high mannose-type glycans by its trimming and subsequent glycan elongation through the Golgi 25 . More specifically, high-mannose including Man(α1-6)Man is synthesized in the ER, Cis-Golgi, and part of the Medial-Golgi, subsequently complex-and hybrid-type glycans are generated in another part of the Medial-Golgi and Trans-Golgi 25,26 . In the final step in the Golgi, sialyltransferase, which enables sialic acid to bind to Gal(β1-4)GlcNAc, also functions in the Trans-Golgi (Supplementary Fig. 7a) 26 . The knockout of genes encoding α1,2-mannosidase-I and N-acetylglucosaminyl-transferase-I in HEK293 cells resulted in removed hybrid-and complex-type N-glycans and only high mannose-type N-glycans among recombinant proteins 27 .
In addition, we previously raised the possibilities that urinary glycan excretion could reflect kidney-specific Table 5. Logistic regression analysis of the renal outcome. Oxford classification; M1, Mesangial hypercellularity score > 0.5; E1, any endocapillary hypercellularity; S1, any segmental sclerosis; T, tubular atrophy and interstitial fibrosis (T0 ≤ 25%, 25% < T1 ≤ 50%, T2 > 50% of cortical area). CI, confidence interval; IFTA, interstitial fibrosis / tubular atrophy. a The absence of each pathological parameter is defined as a reference.  Figure 2. The comparisons of the odds ratios in the groups stratified according to ECA/NPA signals and T score. All patients were divided into four groups by lower or higher median lectin binding signals for ECA, NPA and T score (T0, T1, and T2). The odds ratio for renal outcome was calculated by a logistic regression analysis. The box and neighboring number indicate the odds ratio, and the bar shows the standard error. *P < 0.05 (vs reference group www.nature.com/scientificreports/ alterations of glycosylation rather than circulating serum glycosylation changes 18 . Taken together, the increased urinary high-mannose including Man(α1-6)Man and Gal(β1-4)GlcNAc could reflect the glycosylation abnormality in the Trans-Golgi and Medial-Golgi of renal tissues, and those glycosylation abnormalities might be involved in the progression of IgA nephropathy (Supplementary Fig. 7b).
Since the Oxford classification was published, a number of studies have proved that the classification is useful for predicting the renal prognosis of IgAN. The relationship of the renal prognosis with mesangial and endocapillary hypercellularity is still controversial, while IFTA has been reported to be a strong predictor of the renal outcome in several studies [28][29][30][31] . In our study, IFTA was significantly associated with the renal prognosis independent of baseline proteinuria and eGFR, which was compatible with previous reports [28][29][30][31] . Intriguingly, IFTA demonstrated mild correlation with the ECA and NPA signals, suggesting that the glycans detected by ECA and NPA might be involved in the mechanism of IFTA progression (Table 6). Furthermore, the urinary ECA and NPA signals had incremental predictive abilities when they were added to the model containing IFTA (Table 6). The addition of T scores in multivariate logistic regression model 1 in Tables 2 and 3; Supplementary Tables 4  and 5 did not alter the ORs of ECA and NPA signals, respectively. Therefore, we speculate that ECA/NPA binding glycans and interstitial renal injuries shown by IFTA are independently associated with progression of IgAN.
As well as IFTA, glomerular crescents tightly associated with a poor renal outcome of IgAN, resulting in the new inclusion of cellular and/or fibrocellular crescent scoring in the updated Oxford classification 32,33 . In the current investigation, crescents and segmental sclerosis were not associated with the renal prognosis. We found that patients with cellular crescents were more likely to receive steroid therapy than those without cellular crescents, and steroid therapy was associated with a good renal prognosis in the cellular crescent ( +) group. Likewise, in the patients with segmental sclerosis, those with steroid therapy tended to have a better renal prognosis, although the difference did not reach statistical significance (Supplementary Table 7). Given these associations, the treatment strategies and their therapeutic effects might affect the prognosis, resulting in different consequences from previous studies.
The localization of glycans in kidney tissues has been investigated only in the limited studies. ECA has been reported to bind to the proximal, distal tubules in the cortex, and the loops of Henle in the inner medulla on human kidney tissues 34 . The reported localization of ECA-recognizing Gal(β1-4)GlcNAc may support the link between elevation of urinary ECA signals and IFTA. NPA is a member of a large family of monocot mannosebinding proteins. The preferred glycan structures differ among lectins belonging to the same family. For example, NPA binds to α1-6-linked mannosyl residues, while GNA has a higher affinity to α1-3-linked mannosyl residues 35 . Since the relationship between NPA and kidney diseases has not been investigated, further experimental research is needed.
One of the limitations was the observational study with relatively smaller number of the enrolled patients. For example, only 26 renal events were observed, and 11 patients were classified as T2 category. We could not completely negate the possibility that the treatments for IgAN might not be standardized and the potential confounders could not be fully adjusted in the analyses. However, we observed no statistical differences in major treatment factors, such as ACE-I or ARB use, steroid therapy, and blood pressure control between Group P and G (Supplementary Table 2). Moreover, our sensitivity analyses revealed that ECA and NPA signals were still significant in the multivariate logistic regression analysis even after adjustment for various potential confounders Table 6. AIC, category-free NRI, and IDI for predicting the 3-year outcome with glycan index data, and difference of C-index between estimation models with or without glycan index and T score. Covariates (crude) were age, sex, estimated glomerular filtration rate, and log-transformed urinary protein excretion at the time of renal biopsy. AIC, Akaike's information criterion; NRI, net reclassification improvement; IDI, integrated discrimination improvement; 95% CI, 95% confidence interval; C-index, concordance index; T score, tubular atrophy and interstitial fibrosis (T0 ≤ 25%, 25% < T1 ≤ 50%, T2 > 50% of cortical area); ECA, Erythrina cristagalli lectin; NPA, Narcissus pseudonarcissus lectin; NA, Not applied. a In combination with ECA and NPA model, the AIC was higher than the single models of ECA, NPA, and T score, and other statistical analyses were not performed. www.nature.com/scientificreports/ (Tables 2, 3; Supplementary Tables 4 and 5). Another limitation is that the lectin microarray system may not determine the complete glycan structure and unknown preferred glycans to lectin potentially result in some bias. However, less time-consuming and more cost effective than conventional methods such as mass spectrometry (MS) were the benefits of lectin microarray 20 .
In conclusion, we showed that urinary excretion of glycans binding to two lectins, ECA recognizing Gal(β1-4) GlcNAc, and NPA binding to high-mannose including Man(α1-6)Man, were significantly associated with a poor renal prognosis in patients with IgAN. Furthermore, the addition of one of the two lectin binding signals and the Oxford classification T score to known renal prognostic factors can significantly improve the prediction of renal outcome. We need the further research to prove the underling mechanisms why the ECA and NPA signals could increase in urine of IgAN progressors and the abnormalities of glycosylation, especially N-glycosylation which was commonly recognized by ECA and NPA, might be involved with the progression of IgAN.

Methods
Study design. The current study was conducted as a retrospective cohort study. Among 157 patients diagnosed as "isolated IgAN" by performing biopsies from December 2010 to August 2017 at Okayama University Hospital, 142 were eligible for the enrollment. The patients with ≤ 3 glomeruli on biopsy specimens, < 1 year of follow-up, < 3 repeated measurements of eGFR, and a baseline eGFR < 10 mL/min/1.73 m 2 were excluded (Supplementary Fig. 1).
Ethics statement. This study was conducted in accordance with the principles of the Declaration of Helsinki, and the protocol was approved by the ethics committee of Okayama University Hospital (authorization number: 1709-039). Written informed consents were obtained from all participants. For patients < 18 years old, informed consent was obtained from parents or legal guardian. The study is registered with the University Hospital Medical Information Network Clinical Trials Registry (UMIN000029336).
Lectin microarray analysis. Urine samples collected and stored at renal biopsy were used to measure urinary glycan levels. All specimens were stored at -80 °C, and thawed once to perform this study. We previously described a novel technique of glycan profiling by the evanescent-field fluorescence-assisted lectin microarray. In brief, 20 μL of urine samples were labeled with 100 μg of Cy3 (GE Healthcare) and free Cy3 was removed by Zeba Desalt Spin Column (Pearce). We applied urinary Cy3-labeled glycoproteins on the wells of LecChip  36 . For patients < 18 years old, the eGFR was calculated using the equation reported by the Japanese Society for Pediatric Nephrology 37 . Occult blood in urine was defined as > 5 urinary red blood cells /high-powered field in multiple urinalysis before a renal biopsy.
Diagnosis and Oxford classification. The diagnosis of IgAN performed by 3 nephrologists and a renal pathologist by confirming mesangial proliferative glomerulonephritis in light microscopy, mesangial IgA deposition by immunofluorescence, and electron-dense deposits in the mesangial area by electron microscopy 38 . The following pathological scoring systems were employed, including MEST scores by Oxford classification 39 , presence of crescent formation (cellular/fibrocellular/fibrous) & tuft adhesion, glomeruli with global sclerosis (%), and IFTA (10% increments). There was very good agreement among 3 nephrologists' scores in the percentage of IFTA (weighted κ value 40 : 0.92). The Oxford classification excludes cases with fewer than 8 glomeruli from analysis 39 . In this investigation, 9 patients (7 ≥ glomeruli ≥ 4) were included and the Oxford classification was performed by the consensus of 3 nephrologists.
Outcomes. The primary outcome was defined as meeting at least one of the following criteria: eGFR decline of > 4 ml/min/1.73 m 2 /year during follow-up, 30% decline in eGFR from the baseline within three years, and commencement of renal replacement therapy for end-stage renal disease within the same period. The participants who reached the renal outcome were defined Group P (poor renal outcome; n = 26) and the others as Group G (good renal outcome; n = 116). We selected the composite outcome because the absolute eGFR decline, but the percent eGFR decline, is not affected by the baseline eGFR, and the composite outcome including absolute and percent eGFR decline is employed in a recent biomarker study 41 Non-normally distributed variables were subjected to log-transformation to improve normality before analysis. To evaluate inter-observer concordance, we calculated weighted κ statistics 40 . The logistic regression model was used to calculate the OR and 95% CI. To avoid the type I errors in null hypothesis testing when conducting multiple comparisons, FDR is calculated by Benjamini and Hochberg. In the multivariate model, ORs were adjusted for age, sex, eGFR, and log-transformed UP (g/day) at the time of renal biopsy. These covariates were selected according to biological plausibility and the findings of previous reports 44 . In addition, as sensitivity analyses, other potential covariates were incorporated into the multivariate models one by one. We tested for a formal interaction of each glycan index with eGFR or log-transformed UP in the multivariate regression models. Any glycan index × eGFR/log-transformed UP was not statistically significant. We also divided all patients into four groups by median of the glycan index (ECA or NPA) and T score (T0 / T1 or T2), and calculated the ORs to the renal outcome by a logistic regression analysis. Furthermore, several analyses were employed to evaluate the incremental predictive value of preferable glycan biomarkers and pathological scores. We first used AIC to compare the model fitting. Next, C-index was compared between multivariate logistic regression models with or without biomarkers. Finally, improvement in discriminating the three-year risk of the outcome was assessed by analysis of category-free NRI and IDI, as reported elsewhere [45][46][47] . The 95% CIs for the differences in the C-index, category-free NRI, and IDI were computed based on 500 bootstrap samples. ROC of estimation models with and without glycan index and T score were used to evaluate the characteristics of biomarkers. The cutoff points were calculated by Youden's method.

Data availability
The main clinical data and lectin binding signal data generated during the current study are available in the Supplementary datasheet.