Identification of Glomerular and Plasma Apolipoprotein M as Novel Biomarkers in Glomerular Disease

Abstract

Introduction

Dysregulation of sphingolipid and cholesterol metabolism contributes to the pathogenesis of glomerular diseases (GDs). Apolipoprotein M (ApoM) promotes cholesterol efflux and modulates the bioactive sphingolipid sphingosine-1-phosphate (S1P). Glomerular ApoM expression is decreased in patients with focal segmental glomerulosclerosis (FSGS). We hypothesized that glomerular ApoM deficiency occurs in GD and that ApoM expression and plasma ApoM correlate with outcomes.

Methods

Patients with GD from the Nephrotic Syndrome Study Network (NEPTUNE) were studied. We compared glomerular mRNA expression of ApoM (gApoM), sphingosine kinase 1 (SPHK1), and S1P receptors 1 to 5 (S1PR1–5) in patients (n = 84) and controls (n = 6). We used correlation analyses to determine associations between gApoM, baseline plasma ApoM (pApoM), and urine ApoM (uApoM/Cr). We used linear regression to determine whether gApoM, pApoM, and uApoM/Cr were associated with baseline estimated glomerular filtration rate (eGFR) and proteinuria. Using Cox models, we determined whether gApoM, pApoM, and uApoM/Cr were associated with complete remission (CR) and the composite of end-stage kidney disease (ESKD) or ≥40% eGFR decline.

Results

gApoM was reduced (P < 0.01) and SPHK1 and S1PR1 to 5 expression was increased (P < 0.05) in patients versus controls, consistent with ApoM/S1P pathway modulation. gApoM positively correlated with pApoM in the overall cohort (r = 0.34, P < 0.01) and in the FSGS (r = 0.48, P < 0.05) and minimal change disease (MCD) (r = 0.75, P < 0.05) subgroups. Every unit decrease in gApoM and pApoM (log₂) was associated with a 9.77 ml/min per 1.73 m² (95% confidence interval [CI]: 3.96−15.57) and 13.26 ml/min per 1.73 m² (95% CI: 3.57−22.96) lower baseline eGFR, respectively (P < 0.01). From Cox models adjusted for age, sex, or race, pApoM was a significant predictor of CR (hazard ratio [HR]: 1.85; 95% CI: 1.06–3.23).

Conclusions

pApoM is a potential noninvasive biomarker of gApoM deficiency and strongly associates with clinical outcomes in GD.

Patients with GD who do not achieve remission are at risk for significant morbidity and progressive kidney failure. The most common lesions underlying primary GD (MCD, FSGS, membranous nephropathy [MN], and IgA nephropathy [IgAN]) account for more than half of all cases of GD in the United States and are the third most common cause of ESKD after diabetes mellitus and hypertension.¹ In current clinical practice, GDs are classified according to the histopathologic pattern of injury observed on kidney biopsy, despite evidence that functionally distinct diseases can present with similar histologic patterns.² Therapeutic decisions and outcome prediction rely on this histopathologic classification, combined with the use of surrogate clinical parameters, such as serum creatinine and proteinuria, and response to empiric therapy.³ In view of the heterogeneity of the underlying biological mechanisms, the ability to identify specific biomarkers of molecular pathway activation is a critical step that can enable the development of targeted therapies and a precision medicine approach to the treatment of patients with GD.

Lipid-induced kidney injury is a highly relevant molecular pathway in GD. Dysregulation of intrarenal metabolic pathways involved in reverse cholesterol transport and sphingolipid metabolism plays a key role in mediating lipid-induced podocyte injury in GDs, including FSGS.4, 5, 6, 7 Among several genes involved in cellular lipid homeostasis and cholesterol efflux, decreased expression of ApoM has been identified in the glomerular transcripts of patients with FSGS.⁷ ApoM is mainly located in high-density lipoprotein (HDL) particles, where it exerts a crucial role in HDL metabolism and the mobilization of cellular cholesterol.8, 9, 10, 11, 12, 13 ApoM is highly expressed in the liver and kidneys⁸ and acts as the physiologic carrier and modulator of the bioactive sphingolipid, S1P.¹⁴ The vasoprotective and antiapoptotic actions of ApoM are mediated by the HDL-associated ApoM-S1P complex.14, 15, 16 The role of this pathway in GD is further supported by the finding that recessive mutations in the gene encoding S1P lyase, the main regulator of intracellular S1P levels, lead to the intracellular accumulation of various bioactive sphingolipid intermediates in animal models and the development of steroid-resistant nephrotic syndrome in children.17, 18, 19

These observations led us to hypothesize that GD represents a state of glomerular ApoM deficiency. Further, we sought to determine whether gApoM expression correlates with the glomerular expression of genes involved in S1P metabolism and signaling and with noninvasive measures of ApoM in plasma and urine. Finally, we evaluated whether these markers can predict clinical end points, including proteinuria, eGFR, and probability of disease remission. To do so, we investigated these markers in a cohort of patients with GD from NEPTUNE.

Methods

Study Cohort and Participant Inclusion

The study included 84 participants with GD from NEPTUNE, a multicenter prospective cohort study of children and adults with primary GD enrolled at the time of a clinically indicated kidney biopsy.²⁰ Participants were assigned into a specific study cohort based on detailed histopathological analysis (FSGS, MCD, MN, IgAN, and “Other” histopathology not consistent with the other major patterns). Key exclusion criteria include nephropathy due to systemic diseases such as systemic lupus erythematosus, diabetes, or multiple myeloma. Biopsy tissue, clinical data, and biospecimens were obtained after written informed consent (adults) and assent with written parental consent (children) with approval by the University of Michigan Instituional Review Board and local institutional review boards at each participating center.²⁰ Participants with MicroArray glomerular compartment gene expression data, proteinuria ≥1 g/g, and a baseline visit within 30 days of biopsy were included. The latter was chosen to reduce the possibility of including patients experiencing spontaneous remission, in view of the more liberal inclusion criterion of proteinuria ≥0.5 g/g in the initial phase of NEPTUNE.

Study Measures

Clinical variables were centrally measured at the NEPTUNE laboratory. Biopsy tissue was manually microdissected to separate the glomerular and tubulointerstitial compartments, and compartment-specific measurement of mRNA expression was performed on an Affymetrix 2.1 ST platform as previously described.²¹ Gene expression was normalized, quantitated, and annotated at the EntrezGene level. Kidney tissue from 6 living kidney donors was used to obtain glomerular gene expression in controls. Plasma and urine specimens obtained at the baseline visit were used to perform ApoM measurements. pApoM was measured using a sandwich enzyme-linked immunosorbent assay as previously described.²² Due to a retained signal peptide in the circulating protein, pApoM remains anchored to lipoproteins and does not circulate in a free form.^9,22,23 Calculated from a midrange control sample (ApoM = 0.98 μmol/l), intra-assay and interassay coefficients of variation were 3.2% and 7.9%, respectively. Urinary ApoM was measured using a competitive enzyme-linked immunosorbent assay as previously described.²⁴ Intra-assay and interassay coefficients of variations were below 17%, and the detection limit was 0.59 nM/l. All plasma and urine ApoM measurements were analyzed in duplicate.

Outcomes

eGFR was calculated using the Chronic Kidney Disease in Children U25 formula for patients <25 years and the Chronic Kidney Disease Epidemiology Collaboration formula without the race coefficient for patients 25 years of age and older.^25,26 The urine protein-to-creatinine ratio (UPCR, in g/g) was determined using a 24-hour urine (in 75 subjects) or a spot urine specimen (in 9 subjects). UPCR was not normally distributed and was log₂-transformed. UPCR (log₂) was used for all analyses. CR was defined as a reduction in UPCR to ≤0.3 g/g. Partial remission was defined as a >50% reduction in UPCR and a final UPCR ≤3.5 g/g but >0.3 g/d. The kidney outcome was defined as a composite of ESKD or ≥40% decline in eGFR.²⁰

Statistical Analysis

Demographic, clinical, and histopathologic characteristics were described for the overall study cohort and for each histological subgroup (FSGS, MCD, MN, IgAN, and Other). Data are presented as mean ± SD for normally distributed continuous variables, median and interquartile range (IQR) for non-normally distributed continuous variables, and counts (percentages) for categorial variables. Levels of uApoM were normalized to urinary creatinine excretion (uApoM/Cr). pApoM and uApoM/Cr levels were not normally distributed and were log₂-transformed. pApoM (log₂) and uApoM/Cr (log₂) were used for all analyses. Wilcoxon rank-sum test was used to compare glomerular gene expression levels for genes of interest (ApoM, sphingosine kinases 1 and 2 [SPHK1 and SPHK2, respectively] and S1PR1−5) in the overall GD cohort and within each histological subgroup to controls. Kruskal-Wallis test was used to compare gene expression levels across the different histological subgroups (FSGS, MCD, MN, IgAN, and Other) and in the histological subgroups versus controls. Spearman’s rank correlation was used to determine associations between gApoM and the expression levels for SPHK1, SPHK2, and S1PR1 to 5 in the entire sample of 84 patients and 6 controls. To determine whether gApoM was correlated with noninvasive measures of ApoM in the plasma and urine, Pearson correlation analyses were used to determine associations between gApoM with pApoM and uApoM/Cr in the overall study cohort and within each histological subgroup. Subsequent analyses were performed to determine the relationship between gApoM, pApoM, and uApoM/Cr with the outcomes of interest. Linear regression analysis was used to determine whether gApoM, pApoM, and uApoM/Cr were associated with baseline eGFR and UPCR. Linear regression analysis was performed unadjusted and adjusted for age (model 1); age, sex, and race/ethnicity (model 2); age, sex, race/ethnicity, HDL cholesterol, low-density lipoprotein (LDL) cholesterol, and use of lipid-lowering medication (model 3); and age, sex, race/ethnicity, HDL, LDL, use of lipid-lowering medication, and serum albumin (model 4). Cox proportional hazards models were used to determine whether gApoM, pApoM, and uApoM/Cr were associated with time to CR, CR/partial remission, and the composite of ESKD or ≥40% eGFR decline. Cox models were performed unadjusted and adjusted for potential confounders, including age, sex, race/ethnicity, HDL, LDL, use of lipid-lowering medication, and serum albumin, baseline eGFR, and baseline UPCR. Statistical significance was defined as a 2-tailed value of P < 0.05. Analyses were performed using RStudio version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Baseline characteristics for the overall study cohort and each histological subgroup are presented in Table 1. The study consisted of 84 patients, including 23 with FSGS, 7 with MCD, 31 with MN, 12 with IgAN, and 11 with Other histopathology. The mean age was 40.3 years. In the overall cohort, 17% were children (age <18 years) and 83% were adults; 69% were men and 31% were women; 46% were non-Hispanic White, 22.6% were non-Hispanic Black, 16.7% were Hispanic, 10.7% were Asian, and 3.6% were multiracial. Median time from disease onset to biopsy was 3 (IQR 0−13) months. Mean eGFR at baseline was 80.5 ± 30.1 ml/min per 1.73 m² and median UPCR was 2.91 (IQR 2.07−13.66) g/g. Of the patients, 21% were receiving immunosuppressive therapy prior to biopsy, 61% were receiving renin-angiotensin system blockade at baseline, and 45% were receiving lipid-lowering medication at baseline.

Table 1.

Participant characteristics

Characteristics	Overall (N = 84)	FSGS (n = 23)	MCD (n = 7)	MN (n = 31)	IgAN (n = 12)	Other (n = 11)
Age, mean (SD)	40.3 (19.1)	30.5 (18.6)	23.3 (14.6)	51.0 (14.7)	43.3 (19.0)	38.3 (17.1)
<18 yr, n (%)	14 (16.7%)	7 (30.4%)	3 (42.9%)	1 (3.2%)	1 (8.3%)	2 (18.2%)
Sex, n (%)
Female	26 (31.0%)	6 (26.1%)	2 (28.6%)	11 (35.5%)	3 (25.0%)	4 (36.4%)
Male	58 (69.0%)	17 (73.9%)	5 (71.4%)	20 (64.5%)	9 (75.0%)	7 (63.6%)
Race/Ethnicity, n (%)
Non-Hispanic White	39 (46.4%)	11 (47.8%)	1 (14.3%)	19 (61.3%)	5 (41.7%)	3 (27.3%)
Non-Hispanic Black	19 (22.6%)	7 (30.4%)	5 (71.4%)	5 (16.1%)	1 (8.3%)	1 (9.1%)
Asian	9 (10.7%)	0 (0%)	1 (14.3%)	1 (3.2%)	3 (25.0%)	4 (36.4%)
Multi-Racial	3 (3.6%)	1 (4.3%)	0 (0%)	2 (6.5%)	0 (0%)	0 (0%)
Hispanic, n (%)	14 (16.7%)	4 (17.4%)	0 (0%)	4 (12.9%)	3 (25.0%)	3 (27.3%)
Months from disease onset to biopsy, median (IQR)	3 (0–13)	3 (0–34)	0 (0–2)	7 (1–10)	4.5 (1–41.8)	2.5 (0–8.5)
Days from biopsy to baseline, median (IQR)	16 (7–28)	17 (7.5–27)	8 (3–12.5)	13 (7–22)	28 (19.75–31)	23 (0.5–31.5)
IST use before biopsy, n (%)	18 (21.4%)	4 (17.4%)	6 (85.7%)	4 (12.9%)	4 (33.3%)	0 (0%)
On RAAS blockade, n (%)	51 (60.7%)	12 (52.2%)	2 (28.6%)	22 (71.0%)	9 (75.0%)	6 (54.5%)
On lipid-lowering medication, n (%)	38 (45.2%)	8 (34.8%)	2 (28.6%)	19 (61.3%)	5 (41.7%)	4 (36.4%)
eGFR, mean (SD)a	81.7 (29.9)	76.1 (31.5)	108.5 (36.9)	86.0 (24.2)	74.6 (25.2)	71.4 (34.7)
UPCR, median (IQR)b	2.91 (2.07–13.7)	2.39 (1.34–3.77)	3.34 (2.48–6.23)	3.80 (2.62–7.63)	2.13 (1.74–2.79)	2.97 (2.18–5.80)
Serum albumin, mean (SD)	3.31 (0.77)	3.63 (0.876)	2.68 (0.81)	3.21 (0.57)	N/A	N/A
Total cholesterol, mean (SD)c	271 (94.3)	262 (105)	388 (144)	274 (62.9)	241 (62.4)	239 (93.5)
HDL cholesterol, mean (SD)c	62.3 (23.1)	62.4 (28.7)	74.0 (32.5)	66.2 (19.1)	52.8 (15.5)	54.2 (17.2)
LDL cholesterol, mean (SD)c	174 (80.7)	163 (85.6)	278 (129)	174 (54.1)	160 (56.7)	145 (80.9)
Triglycerides, median (IQR)c	164 (118–245)	153 (105–267)	212 (156–230)	165 (135–233)	124 (115–176)	204 (136–353)

eGFR, estimated glomerular filtration rate; FSGS, focal segmental glomerulosclerosis; HDL, high-density lipoprotein; IgAN, IgA nephropathy; IQR, interquartile range; IST, immunosuppressive therapy; LDL, low-density lipoprotein; MCD, minimal change disease; MN, membranous nephropathy; RAAS, renin-angiotensin-aldosterone system; UPCR, urine protein-to-creatinine ratio.

^aeGFR was calculated from serum creatinine using the Chronic Kidney Disease in Children (CKiD) U25 formula for patients <25 years and the CKD Epidemiology Collaboration (CKD-EPI) formula without the race coefficient for patients 25 years of age and older.

^bThe urine protein-to-creatinine ratio was obtained from a 24-hour urine specimen.

^cLevels of total cholesterol, HDL, and triglycerides were determined at the central laboratory from a fasting blood draw obtained at the baseline visit. The LDL cholesterol was calculated using the Martin-Hopkins methods based on the other 3 measures.

Glomerular ApoM Deficiency and pApoM/S1P Pathway Modulation in GD

gApoM was lower in patients with GD compared with controls (overall cohort vs. controls, P < 0.001 in Figure 1a and in each subgroup vs. controls in Figure 1b, P < 0.05 for all comparisons except for MCD vs. controls), with no significant differences between subgroups (P = 0.13). The glomerular expression of SPHK1 (P < 0.05), but not SHPK2 (P = 0.76), was significantly higher in GD compared with controls (Figure 1c and d). The glomerular expression of S1PR1−5 was higher in GD compared with controls, P < 0.005 for all comparisons except S1PR5 versus controls (P = 0.064, Figure 1e–i). There were no significant differences in gene expression levels of SPHK1, SPHK2, or S1PR1−5 across the GD subgroups (not shown). gApoM negatively correlated with the glomerular expression of SPHK1 (r = −0.21, P < 0.05) and S1P receptors (r = −0.42, P < 0.01 for S1PR1; r = −0.3, P < 0.01 for S1PR2; r = −0.22, P < 0.05 for S1PR3; r = −0.2, P = 0.06 for S1PR4; and r = −0.21, P = 0.05 for S1PR5), but not SPHK2 (not shown).

Figure 1.

Glomerular ApoM expression and ApoM/S1P pathway modulation in GD. (a) gApoM expression in GD patients vs controls, P <0.001. (b) gApoM expression in subgroups vs controls, ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001; ns, nonsignificant. (c) Glomerular SPHK1 expression in GD patients vs controls, P < 0.05. (d) Glomerular SPHK2 expression in GD patients vs controls. (e-h) Glomerular expression of S1PR1-5 in GD patients vs controls, P < 0.05 for all comparisons except S1PR5 vs controls. ApoM, apolipoprotein M; FSGS, focal segmental glomerulosclerosis; GD, glomerular disease; MCD, minimal change disease; MN, membranous nephropathy; S1P, sphingosine-1-phosphate; SPHK1, sphingosine kinase 1; SPHK2, sphingosine kinase 2; S1PR1-5, S1P receptors 1 to 5.

Correlations Between gApoM With pApoM and uApoM as Biomarkers of ApoM Expression

pApoM levels are shown in Supplementary Figure S1. The median (IQR) for pApoM was 1.17 (0.88) μmol/l. gApoM positively correlated with pApoM in the overall cohort (r = 0.34, P < 0.01) and in the FSGS (r = 0.48, P < 0.05) and MCD (r = 0.75, P < 0.05, Figure 2a–c) subgroups. uApoM levels are shown in Supplementary Figure S1. The median (IQR) for uApoM was 0.44 (0.92) nmol/l. gApoM did not correlate with uApoM/Cr levels in the GD cohort or within subgroups, before or after exclusion of the 33 patients who had undetectable levels of ApoM in the urine (Figure 2d).

Figure 2.

Correlations between gApoM, pApoM, and uApoM/Cr in GD and among subgroups. (a) Correlation between gApoM expression and pApoM in the GD cohort. (b) Correlation between gApoM expression and pApoM in the FSGS subgroup. (c) Correlation between gApoM expression and pApoM in the MCD subgroup. (d) Correlation between gApoM expression and uApoM/Cr in the GD cohort. FSGS, focal segmental glomerulosclerosis; GD, glomerular disease; MCD, minimal change disease.

Associations Between gApoM, pApoM, and uApoM with Baseline eGFR

The associations of gApoM, pApoM, and uApoM/Cr with baseline eGFR are shown in Figure 3. In the unadjusted linear regression model, gApoM showed a significant positive association with eGFR, such that each unit decrement of gApoM was associated with a 9.77 ml/min per 1.73 m² (95% CI: 3.96–15.57) lower baseline eGFR (R² = 0.12, P < 0.01, Figure 3a). The association was also significant in the FSGS (β = 13.90, 95% CI: 2.02–25.78; R² = 0.22, P < 0.05) and IgAN (β = 14.13; 95% CI: 4.17−24.08, R² = 0.5, P = 0.01) subgroups. A significant association persisted after adjusting for age in model 1 (R² = 0.18, P < 0.01); for age, sex, and race in model 2 (R² = 0.38, P < 0.01); for age, sex, race, HDL, LDL, and use of lipid-lowering medications in model 3 (R² = 0.42, P < 0.05); and for age, sex, race, HDL, LDL, use of lipid-lowering medications, and serum albumin in model 4 (R² = 0.46, P < 0.01). In the unadjusted linear regression model, pApoM showed a significant positive association with eGFR (β = 13.26; 95% CI: 3.57–22.96, R² = 0.08, P < 0.01, Figure 3b). A significant association persisted in model 1 (R² = 0.14, P < 0.05) and model 2 (R² = 0.33, P < 0.05), although not in model 3 (R² = 0.37, P = 0.91) or model 4 (R² = 0.39, P = 0.92). uApoM/Cr was not significantly associated with eGFR at baseline (not shown).

Figure 3.

Associations between gApoM and pApoM with baseline eGFR. (a) Linear regression showing the association between gApoM expression and baseline eGFR in the GD cohort. (b) Linear regression showing the association between pApoM and baseline eGFR in the GD cohort. ApoM, apolipoprotein M; eGFR, estimated glomerular filtration rate.

Associations Between gApoM, pApoM, and uApoM With Baseline UPCR

The associations of gApoM and pApoM with baseline UPCR are shown in Figure 4. There was no significant association between gApoM and UPCR in the GD cohort (P = 0.26). In the univariate model, there was a positive association between pApoM and UPCR in the GD cohort (β = 0.81, 95% CI: 0.51−1.11, R² = 0.27, P < 0.01, Figure 4a) and in the FSGS (β = 1.36, 95% CI: 0.79−1.93, R² = 0.54, P < 0.01) and IgAN (β = 0.45, 95% CI: 0.024−0.88, R² = 0.36, P < 0.05) subgroups. To further explore this association, we repeated the analyses after stratification by potential covariates. The association between pApoM and UPCR in the GD cohort remained significant in patients with baseline eGFR > 60 ml/min per 1.73 m² (R² = 0.32, P < 0.01) but not among those with baseline eGFR < 60 ml/min per 1.73 m² (R² = 0.044, P = 0.35), in patients with UPCR > 3 g/g (R² = 0.13, P < 0.05) but not among those with UPCR < 3 g/g (R² = 0.065, P = 0.09), and in patients with plasma S1P levels above the median (R² = 0.47, P < 0.01) but not among those with plasma S1P levels below the median (R² = 0.049, P = 0.16, Figure 4b–g). The association between pApoM and UPCR in the GD cohort remained significant in model 1 (R² = 0.27, P < 0.01), model 2 (R² = 0.30, P < 0.01), model 3 (R² = 0.36, P < 0.05), and model 4 (R² = 0.46, P < 0.05). For uApoM/Cr, there was a positive association between uApoM/Cr and baseline UPCR in the GD cohort (R² = 0.27, P < 0.01, Figure 4h) that remained significant in model 1 (R² = 0.32, P < 0.01), model 2 (R² = 0.34, P < 0.01), model 3 (R² = 0.60, P < 0.01), and model 4 (R² = 0.83, P < 0.05).

Figure 4.

Associations between gApoM, pApoM, and uApoM with baseline UPCR. (a) Association between pApoM and baseline UPCR. (b) Association between pApoM and UPCR, stratified by baseline eGFR >60 mL/min/1.73 m². (c) Association between pApoM and UPCR, stratified by baseline eGFR <60 mL/min/1.73 m². (d) Association between pApoM and UPCR, stratified by baseline UPCR >3 g/g (UPCR log₂ >1.585). (e) Association between pApoM and UPCR, stratified by baseline UPCR <3 g/g (UPCR log₂ >1.585). (f) Association between pApoM and UPCR, stratified by plasma S1P above the median. (g) Association between pApoM and UPCR, stratified by plasma S1P below the median. (h) Association between uApoM/Cr and baseline UPCR in the GD cohort. eGFR, estimated glomerular filtration rate; pS1P, plasma sphingosine-1-phosphate; UPCR, urine protein-to-creatinine ratio.

Associations Between gApoM, pApoM, and uApoM With Longitudinal Outcomes

Median follow-up time was 48.5 months (IQR 30–55 months). The associations of gApoM, pApoM, and uApoM/Cr levels with the longitudinal outcomes of CR, CR/partial remission, and the combined kidney outcome are shown in Table 2 and Figure 5. In the univariable Cox proportional-hazards model, gApoM (HR: 0.92; 95% CI: 0.68–1.26, P = 0.61), pApoM (HR: 1.44; 95% CI: 0.90–2.33, P = 0.13), and uApoM/Cr (HR: 1.07; 95% CI: 0.82–1.40, P = 0.61) were not significantly associated with CR in the overall cohort. However, in the FSGS subgroup, pApoM was borderline significant for predicting CR (HR: 2.19; 95% CI: 0.96–5.01, P = 0.063). In the multivariable Cox model in the overall cohort, pApoM was a significant predictor of CR after adjustment for age, sex, and race (HR: 1.85; 95% CI: 1.06–3.23, P < 0.05). pApoM remained significantly associated with CR after further adjusting for either baseline eGFR (HR: 1.82; 95% CI: 1.03–3.21, P < 0.05) or baseline UPCR (HR: 1.99; 95% CI: 1.01–3.92, P < 0.05), and borderline significant after further adjusting for both baseline eGFR and UPCR (HR: 1.96; 95% CI: 0.99–3.89, P = 0.055) but not after adjustment for age, sex, race, HDL, LDL, use of lipid-lowering medication, and serum albumin (HR: 1.17, 95% CI: 0.43–3.18, P = 0.76). gApoM, pApoM, and uApoM/Cr were not associated with CR/partial remission or the combined outcome of ESKD or ≥40% eGFR decline.

Table 2.

Associations between gApoM, pApoM, or uApoM/Cr with longitudinal outcomes

Variables	Predictor	Unadjusted risk	P-value	Adjusted for age/sex/race	P-value
Outcome		HR (95% CI)		HR (95% CI)
Complete remission
	Glomerular ApoM	0.92 (0.68–1.26)	0.612	0.84 (0.60–1.17)	0.296
	Plasma ApoM (log2)	1.44 (0.90–2.33)	0.132	1.85 (1.06–3.23)	0.0304a
	Urine ApoM/Cr (log2)	1.07 (0.82–1.39)	0.613	1.20 (0.86–1.69)	0.28
Partial or complete remission
	Glomerular ApoM	0.98 (0.78–1.24)	0.866	1.15 (0.87–1.54)	0.331
	Plasma ApoM (log2)	1.29 (0.90–1.83)	0.162	1.41 (0.94–2.13)	0.101
	Urine ApoM/Cr (log2)	0.99 (0.82–1.20)	0.925	1.09 (0.79–1.49)	0.60
ESKD or 40% decline in eGFR and eGFR <90
	Glomerular ApoM	0.94 (0.61–1.45)	0.783	1.02 (0.61–1.71)	0.927
	Plasma ApoM (log2)	0.90 (0.47–1.72)	0.753	1.08 (0.51–2.26)	0.85
	Urine ApoM/Cr (log2)	1.29 (0.93–1.80)	0.125	1.17 (0.76–1.81)	0.49

ApoM, apolipoprotein M; CI, confidence interval; eGFR, estimated glomerular filtration rate; ESKD, end-stage kidney disease; HR, hazard ratio.

^aIndicates estimate is significant (P < 0.05).

Figure 5.

Associations between pApoM and complete remission. CI, confidence interval; eGFR, estimated glomerular filtration rate; HR, hazard ratio; UPCR, urine protein-to-creatinine ratio.

Discussion

We found that gApoM was reduced in patients with GD compared with controls with no significant differences between histological subgroups, suggesting the presence of gApoM deficiency as a common molecular signature in GD, irrespective of the underlying histologic diagnosis. The glomerular expression of SPHK1 and S1PR1−5 was increased in patients with GD, supporting the hypothesis that activation of S1P-mediated signaling and modulation of S1P metabolism occur in the glomeruli of patients with GD and are associated with gApoM deficiency. Furthermore, we found that gApoM deficiency was associated with lower levels of pApoM, but not of uApoM, suggesting a potential role for pApoM as a noninvasive biomarker of gApoM deficiency. Lower gApoM and pApoM were significantly associated with a lower baseline eGFR. Although gApoM was not associated with baseline proteinuria in the GD cohort, there was a positive association between pApoM and baseline UPCR, particularly among patients with baseline eGFR > 60 and UPCR > 3 g/g, and among patients with higher plasma S1P levels. Importantly, pApoM was a significant predictor of CR, such that lower pApoM was associated with a lower likelihood of CR after adjusting for potential confounders, including baseline eGFR and proteinuria.

ApoM is a 25 kDa apolipoprotein that is a member of the lipocalin protein superfamily owing to its ability to bind small lipophilic ligands in a hydrophobic binding pocket.²⁷ ApoM is primarily expressed in the liver and kidneys, with high levels of expression in proximal tubular epithelial cells and lower levels in glomerular cells, including podocytes.^28,29 ApoM exerts anti-inflammatory and antiatherogenic effects by enhancing the cholesterol efflux capacity and antioxidative effects of HDL.9, 10, 11, 12, 13 In particular, ApoM plays a crucial role in HDL metabolism by stimulating the formation of pre-β-HDL, a key acceptor of cellular cholesterol upon efflux from peripheral tissues.^10,13 Pre-β-HDL mediates ATP-binding cassette transporter A1-dependent efflux of cholesterol from macrophage-derived foam cells and podocytes through the process of reverse cholesterol transport.^5,7 ApoM deficiency is associated with the formation of defective pre-β-HDL leading to reduced ATP-binding cassette transporter A1-mediated reverse cholesterol transport and cholesterol accumulation in macrophages, thus promoting atherosclerotic lesions in mice.^9,10 In the kidney, intracellular cholesterol accumulation due to impaired cholesterol efflux in association with decreased glomerular ATP-binding cassette transporter A1 expression impairs podocyte function and contributes to the development and progression of diabetic kidney disease, FSGS, and Alport syndrome.^4,5,7 Decreased expression of ApoM has been described in the glomerular transcripts obtained from kidney biopsy tissue in patients with FSGS enrolled in the NEPTUNE study.⁷ Our data further support that gApoM expression is significantly reduced in patients with GD compared with controls.

Furthermore, the vasoprotective and antiapoptotic effects of ApoM are mediated by the HDL-associated ApoM-S1P complex, and ApoM may promote renal protection in part by modulating the effects of S1P, which exerts numerous biologic effects through autocrine and paracrine routes via binding to 5 G protein-coupled receptors (S1PR1−S1PR5).14, 15, 16 In the hyper-IgA mouse model for IgAN, ApoM expression in the kidney is suppressed, and knockdown of ApoM results in mesangial proliferation, worsened proteinuria, and increased expression of fibrosis-related genes. Conversely, adenoviral ApoM overexpression ameliorates mesangial proliferation, proteinuria, and fibrosis.³⁰ Similarly, in a model of lipopolysaccharide-induced organ injury, ApoM overexpression ameliorates the adverse effects of lipopolysaccharide administration on mortality and organ injury, whereas knockout or knockdown of ApoM worsens survival and results in significantly higher plasma creatinine levels compared with lipopolysaccharide-treated wild-type mice.³¹ In our study, the increased glomerular expression of SPHK1 and S1PR1 to 5 in patients with GD in association with decreased gApoM compared with controls is consistent with S1P pathway activation in GD. As others have reported that inhibition of S1PR1 and S1PR3 signaling could partially reverse the protective effect of ApoM overexpression,^30,31 the effect of ApoM and/or ApoM-S1P complex on individual S1P receptor pathway activation remains to be established and is the subject of ongoing experimental work.

ApoM expression in the liver and kidneys and pApoM levels are also altered in a variety of systemic conditions, including type 2 diabetes mellitus, systemic lupus erythematosus, sepsis, and heart failure.31, 32, 33, 34, 35, 36 In patients with systemic lupus erythematosus, pApoM levels are reduced compared with controls and correlate inversely with the presence of active nephritis, inflammation, and overall disease severity.³⁴ pApoM levels are modestly reduced in patients with type 2 diabetes mellitus and negatively correlate with body mass index and insulin resistance.37, 38, 39 Several single nucleotide polymorphisms in the ApoM promoter region have been shown to confer an increased risk of developing type 1 diabetes mellitus and type 2 diabetes mellitus in Chinese and Swedish populations.^40,41 Furthermore, low pApoM is independently associated with adverse outcomes including all-cause mortality in patients with heart failure, and lower pApoM levels are associated with higher levels of multiple inflammatory markers, including TNF-alpha and TNF-alpha receptors.³⁶ Finally, in 409 patients with chronic kidney disease (CKD), including 32% with “chronic glomerulonephritis/vasculitis,” pApoM levels were significantly lower in patients with CKD stages 3–5 than among patients with CKD stages 1 and 2 and healthy controls. Moreover, pApoM levels were significantly lower among CKD patients with known cardiovascular disease compared with those without known cardiovascular disease, among CKD patients with type 2 diabetes mellitus compared with nondiabetic CKD, and among patients with ESKD compared with controls.^24,42 The present findings of a positive association between gApoM and baseline eGFR are consistent with a renoprotective role for ApoM. In our study, each unit decrement in gApoM was associated with a ∼10 ml/min per 1.73 m² lower baseline eGFR in the GD cohort and in the FSGS and IgAN subgroups, and each unit decrement in log₂-transformed pApoM levels was associated with a ∼13 ml/min per 1.73 m² lower baseline eGFR in the GD cohort. Likewise, and consistent with renoprotection, per every log₂ higher pApoM level, there was an 85% higher likelihood of CR in the overall cohort in the minimally adjusted analysis and a 96% higher likelihood of CR after additional adjustment for baseline kidney function and proteinuria, although the association was attenuated in fully adjusted models.

Due to a retained hydrophobic signal peptide, pApoM is not present in its free form in the circulation but rather is bound to lipoproteins, in particular HDL, such that 95% of pApoM is associated with HDL and the remainder with LDL, very low density lipoprotein, and chylomicron particles.^9,23 However, the concentration of pApoM is relatively low (∼0.9 μmol/l or ∼23 mg/l), and only about 5% of HDL particles carry ApoM. Though the liver is the primary source of pApoM, there is also a component of pApoM derived from the kidney and other cell types.⁴³ In our study, 12%, 23%, and 56% of the variance in pApoM levels in the overall GD cohort and in the FSGS and MCD subgroups, respectively, was explained by lower gApoM, supported by coefficients of determination in the correlation analyses between gApoM and pApoM. pApoM does not seem to undergo glomerular filtration due to its binding to large lipoproteins, and while it undergoes secretion and megalin-mediated reuptake in the proximal tubule, there is no detectable ApoM in the urine under normal conditions.⁴⁴ Data on associations between pApoM and uApoM levels and proteinuria are scarce, and whether uApoM correlates with the degree of proteinuria is not known. Whereas some studies have found pApoM levels to be higher in patients with diabetic nephropathy compared with those with diabetes without albuminuria, others have found no correlation between pApoM levels and albuminuria.^45,46 Here, we found that pApoM correlated positively with proteinuria, particularly in patients with baseline eGFR > 60, UPCR > 3 g/g, and higher plasma S1P levels. In addition, we found that uApoM/Cr levels correlated positively with proteinuria, but not with eGFR. In a recent study, while uApoM/Cr levels were increased among some patients with acute kidney injury, CKD stages 3 to 4, and proteinuria > 3 g/d compared with low or undetectable levels among healthy controls, there was no statistically significant difference between these groups.²⁴ These findings suggest that HDL-associated ApoM does not undergo glomerular filtration, even when the glomerular filtration barrier is impaired. However, it remains unknown whether the increase in uApoM detected in our cohort is due to glomerular filtration in the presence of marked proteinuria. Our findings support the conclusion that uApoM is not a clinically useful biomarker in GD.

Our study has several limitations. First, we did not evaluate the differential effects of circulating HDL-ApoM/S1P complex and albumin-S1P complex (unbound to HDL-cholesterol) on outcomes. Recent studies have demonstrated that S1P in the ApoM/HDL-bound form (∼65% of total plasma S1P) exerts antiapoptotic, anti-inflammatory, and vasoprotective effects, whereas albumin-bound S1P (∼30% of total plasma S1P) is associated with mesangial cell proliferation, vasoconstriction, increased vascular permeability, and organ fibrosis, suggesting that ApoM/HDL-bound S1P might be associated with a protective effect, whereas albumin-S1P in the absence of ApoM will be associated with worse clinical end points.^14,15 This differential association will be investigated in future studies. Second, we did not explore the interaction or modulating effects between S1P pathway activation, ApoM deficiency, and outcomes. Third, our study has limited power to detect differences between histological subgroups of patients with GD and between subgroups and controls. The present findings are consistent with the hypothesis that ApoM deficiency is common to GDs and not limited to one disease subgroup or histologic pattern of injury. Whether the lack of differential expression identified in the MCD subgroup compared with controls is due to reduced statistical power or whether this represents a distinct molecular pathway in MCD needs to be confirmed in larger cohorts with increased power to detect differences between disease subgroups. Finally, our study may have limited power to detect significant associations with clinical outcomes in fully adjusted models due to the low number of events (47 events for CR, 21 events for the composite of ESKD or 40% decline in eGFR). The addition of potential confounders including HDL cholesterol, LDL cholesterol, and use of lipid-lowering medications attenuated the primary association between pApoM with outcomes (namely, eGFR and CR). However, adjustment for multiple variables may result in an overfitted model due to the low number of events, and adjustment for age, sex, and race may represent a more parsimonious model that is able to detect significant association between pApoM and clinical outcomes. Results will be confirmed in larger cohorts with increased power to detect significant associations in fully adjusted models. Our study has several strengths. We have leveraged the multicenter, prospective NEPTUNE cohort that is collecting data across the genotype-phenotype continuum, enabling the integration of clinical and histological data with comprehensive genetic, transcriptomic, proteomic, and metabolomic datasets.⁴⁷ In addition, NEPTUNE is a diverse cohort that allows for the generalization of findings.

Conclusion

ApoM expression was reduced in the glomeruli of patients with GD compared with healthy controls. Decreased gApoM was associated with increased expression of SPHK1 and S1PR1 to 5, indicating that the ApoM/S1P pathway is modulated in GD. Whether S1P pathway activation contributes to disease pathogenesis and/or progression in GD warrants further study. The observation that gApoM positively correlates with pApoM suggests a potential role for pApoM as a noninvasive biomarker of gApoM deficiency. Reduced gApoM and lower pApoM were significantly associated with a lower baseline eGFR. Finally, pApoM was a significant predictor of CR among patients with GD. Taken together, our study identifies gApoM deficiency and pApoM as potential biomarkers of adverse clinical outcomes in patients with GD. Whether ApoM has a causal role in regulating lipid metabolism and cholesterol accumulation in the kidney remains to be investigated.

Disclosure

YD has nothing to disclose. AF and SM are inventors on pending or issued patents (US10,183,038, US10,052,345) aimed to diagnose or treat proteinuric renal diseases. They stand to gain royalties from their future commercialization. AF is Chief Scientific Officer of L&F Health LLC, holds equity interests in L&F Research, and is the inventor of assets developed by ZyVersa Therapeutics. ZyVersa has licensed worldwide rights to develop and commercialize hydroxypropyl-beta-cyclodextrin for treatment of kidney disease from L&F Research. AF holds equity in River 3 Renal Corporation. SM holds equity interest in L&F Research.

Supplementary Material

Figure S1. Plasma and urine ApoM levels in GD patients.