Association of Proximal Tubular Secretory Clearance with Long-Term Decline in Cognitive Function

Significance Statement

Patients with CKD are at high risk for cognitive impairment and progressive cognitive decline. The retention of organic solutes, which are predominantly cleared by tubular secretion, is hypothesized to contribute to cognitive impairment in such patients. In a multicenter prospective study of 2326 participants with CKD who were initially free of cognitive impairment, the authors found that lower 24-hour kidney clearance of a panel of secretory solutes was associated with cognitive decline, independent of eGFR and proteinuria. These findings highlight the potential contribution of kidney tubular clearance of secretory solutes to maintain normal cognitive function. Further work is needed to determine the mechanisms underpinning the associations between lower secretory clearance and cognitive decline.

Visual Abstract

Abstract

Background

People with chronic kidney disease (CKD) are at high risk for cognitive impairment and progressive cognitive decline. Retention of protein-bound organic solutes that are normally removed by tubular secretion is hypothesized to contribute to cognitive impairment in CKD.

Methods

We followed 2362 participants who were initially free of cognitive impairment and stroke in the prospective Chronic Renal Insufficiency Cohort (CRIC) Study. We estimated tubular secretory clearance by the 24-hour kidney clearances of eight endogenous solutes that are primarily eliminated by tubular secretion. CRIC study investigators assessed participants’ cognitive function annually using the Modified Mini-Mental State (3MS) Examination. Cognitive decline was defined as a sustained decrease of more than five points in the 3MS score from baseline. Using Cox regression models adjusted for potential confounders, we analyzed associations between secretory solute clearances, serum solute concentrations, and cognitive decline.

Results

The median number of follow-up 3MS examinations was six per participant. There were 247 incident cognitive decline events over a median of 9.1 years of follow-up. Lower kidney clearances of five of the eight secretory solutes (cinnamoylglycine, isovalerylglycine, kynurenic acid, pyridoxic acid, and tiglylglycine) were associated with cognitive decline after adjustment for baseline eGFR, proteinuria, and other confounding variables. Effect sizes ranged from a 17% to a 34% higher risk of cognitive decline per 50% lower clearance. In contrast, serum concentrations of the solutes were not associated with cognitive decline.

Conclusions

Lower kidney clearances of secreted solutes are associated with incident global cognitive decline in a prospective study of CKD, independent of eGFR. Further work is needed to determine the domains of cognition most affected by decreased secretory clearance and the mechanisms of these associations.

Introduction

Cognitive impairment is disproportionately prevalent in people with CKD.^1–5 Lower eGFR is associated with a greater risk of cognitive decline after controlling for established risk factors, suggesting that kidney disease or its metabolic complications may contribute to this condition. Specifically, the accumulation of potentially toxic solutes in CKD is one hypothesized mechanism of cognitive impairment.

Observational studies and experimental models have identified candidate uremic solutes that may contribute to cognitive impairment. Many of these solutes are protein bound and avidly secreted by the proximal tubules, but inefficiently cleared by glomerular filtration or traditional dialysis treatments.^6–8 CKD leads to the retention of tryptophan, kynurenic acid, and indole metabolites, which are plausibly linked with cognitive impairment.^9–11 For example, in a study of 199 patients with CKD and 84 controls with normal kidney function, higher serum concentrations of the protein-bound secretory solute indoxyl sulfate were associated with worse cognitive function, although this study did not adjust for eGFR.^12,13 Higher serum levels of p-cresol sulfate and hippurate, both protein-bound uremic solutes, have been associated with cognitive impairment in patients receiving maintenance hemodialysis.^14–16 The blood-brain barrier shares many of the same secretory transporters found in the proximal tubule for efflux of protein-bound solutes, such as hippurate and indoxyl sulfate, and increased levels of these solutes have been found in the cerebrospinal fluid of humans with kidney failure.^17–19

Previous studies of cognitive impairment in CKD have focused on serum concentrations of individual solutes. Given the general role of tubular secretory clearance in eliminating these substances, we tested the hypothesis that lower native secretory clearance is associated with incident cognitive decline in a national multicenter prospective cohort study of CKD.

Methods

Study Population

We used data from the Chronic Renal Insufficiency Cohort (CRIC) Study. Between June 2003 and August 2008, CRIC recruited 3939 participants with eGFRs in the range 20–70 ml/min per 1.73 m² from seven clinical centers across the United States (Ann Arbor/Detroit, MI; Baltimore, MD; Chicago, IL; Cleveland, OH; New Orleans, LA; Philadelphia, PA; and Oakland, CA).^20,21 People receiving maintenance dialysis and those with a prior kidney transplant were excluded. Additional exclusion criteria included polycystic kidney disease, multiple myeloma, HIV infection, cirrhosis, and severe (NYHA Class III or IV) heart failure. All participants provided written informed consent. The CRIC Study protocol was approved by Institutional Review Boards at each of the participating sites.

For the present analysis, we began with 2953 participants who had previously completed secretory solute clearance measurements at the CRIC baseline visit and who had at least one baseline and one follow-up Modified Mini-Mental State (3MS) score for analysis. We subsequently excluded 273 participants who had a history of stroke and 314 participants who had prevalent cognitive impairment at baseline, defined by a 3MS score <85 in participants aged <65 years, <80 in participants aged 65–80 years, and <75 in participants aged >80 years, as previously described.^22–28 As a surrogate for clinical diagnosis of dementia, we further excluded four participants who reported using antidementia medications (donepezil, galantamine, rivastigmine, and memantine) at baseline, giving a final analytic cohort of 2362 participants (Supplemental Figure 1).

Measurements of Cognitive Function

The 3MS is a test of global cognitive function that includes components of concentration, orientation, language, praxis, and memory.²² Scores range from 0 to 100, with higher scores indicating better cognitive function; scores <80 are highly sensitive for dementia.²² Trained CRIC Study personnel administered the 3MS annually for participants in the CRIC COG Cohort and biennially for others during phase 1 of CRIC (2003–2008), biennially in phase 2 (2008–2013), and, since 2013, annually in individuals >65 years of age and biennially in others. Consistent with previous studies, we defined cognitive decline by an absolute decrease in the 3MS score of more than five points from the baseline value.²⁹ To increase specificity, we further required the decline in 3MS score of more than five points to be sustained over at least two consecutive cognitive evaluations.

In secondary analyses, we explored the results of additional cognitive testing available in a subgroup of participants (n=1919). The median, interquartile range, and total range of the timing of these tests was 5 (4–6) and (2–13) years after the initial baseline visit; no baseline measures of these tests were obtained. Starting in 2006, the CRIC Cognitive Ancillary Study recruited participants aged ≥55 years from four of the seven CRIC Clinical Centers; these participants had additional cognitive tests administered at study visits. These included the Trail-Making Test, Forms A and B (Trails A and B), and the Buschke Selective Reminding Test (SRT). Trails A measures attention, visuospatial screening, and motor speed, whereas Trails B is primarily a test of executive function.³⁰ The Buschke SRT measures verbal memory with delayed components.³¹

Measurements of Secretory Solutes and Kidney Clearances

We previously selected candidate solutes suspected to be cleared primarily by proximal tubular secretion on the basis of one or more of the following characteristics: specificity for organic anion transporters type 1 and type 3(OAT1 and OAT3), an increase in circulating concentrations in OAT3-transporter knockout models, a high reported protein-binding percentage, and/or kidney clearances that substantially exceed GFR or creatinine clearance.^6,19,32 We then developed a targeted liquid chromatography-tandem mass spectrometry (LC-MS/MS) assay for these solutes in plasma and urine.³³

We measured secretory solutes in paired 24-hour urine and plasma samples from the baseline visit. We measured total plasma concentrations of secretory solutes using protein precipitation, solid phase extraction, and LC-MS/MS. We measured 24-hour urine concentrations using solid-phase extraction and LC-MS/MS. Calibration was achieved using a single-point calibration approach to account for potential drift that may be caused by changes in reagent lots, calibrator lots, or equipment performance. Additional details regarding standardization and coefficients of variation (which were generally low) have been previously published.^33,34 We calculated the kidney clearance of each solute as (U_X×V)/P_X, where U_X represents the urine concentration of solute X, V represents the 24-hour urine volume, and P_X represents the plasma concentration.

We created a summary measure of the clearances of secretory solutes to provide a singular metric to summarize associations with study outcomes. Because the kidney clearances of each solute inherently reside on different scales, we natural log transformed the individual clearances and then standardized them to a common 0–100 scale:

$$\text{Standardized clearance}=\langle \left\{\text{ln}\left[\text{clearance}\right]-\text{min}\left[\text{ln}\left(\text{clearance}\right)\right]\right\}/\text{range}\left[\text{ln}\left(\text{clearance}\right)\right]\rangle \times 100$$

where ln[clearance] represents the natural log-transformed clearance value, min[ln(clearance)] represents the minimum value of natural log-transformed clearance in the distribution, and range[ln(clearance)] represents the difference between the maximum and minimum values. We then computed the summary score of the average of the eight standardized clearances. The resultant computed summary score is therefore specific to the CRIC Study population, not the general population as may be ideal.³⁴

Measurements of Covariates

At baseline, participants provided information on their sociodemographic characteristics, medical history, medication use, and lifestyle behaviors. Race and ethnicity were self-reported. Educational attainment was categorized as less than a high school diploma, high school graduate or GED, some college without a degree, and college graduate or greater. Baseline cardiovascular disease status was determined by self-report and was defined as prior history of coronary artery disease, heart failure, or stroke. Baseline peripheral arterial disease was self-reported. BP and body mass index were assessed using standard protocols.³⁵ Diabetes mellitus was defined as a fasting glucose >126 mg/dl, a nonfasting glucose >200 mg/dl, or use of medications for diabetes mellitus, including insulin.

Serum creatinine concentrations were measured using an enzymatic method on an Ortho Vitros 950 (Raritan, NJ) at the CRIC Central Laboratory using a standardized enzymatic method.³⁶ eGFR was calculated from serum creatinine using the Chronic Kidney Disease Epidemiology Collaboration equation.³⁷

Statistical Analyses

We summarized study variables by tertiles of baseline 3MS scores. For analyses of decline in 3MS scores, participants were followed from the baseline visit until either they developed sustained cognitive decline (each participant’s second sequential evaluation with a 3MS score more than five points lower than their baseline 3MS score) or their data were censored due to death, lost to follow-up, withdrawal of consent, or the end of administrative follow-up, whichever came first. We calculated incidence rates of cognitive decline by dividing the number of events by the sum of follow-up time across the study population and multiplying by 100 and used a nonparametric bootstrap approach with 2000 replicates to obtain corresponding 95% confidence intervals (CI).³⁸

For each secretory solute, we calculated Pearson’s correlation coefficient between the log-transformed plasma concentration and urinary clearance at baseline. We constructed boxplots of solute clearances by tertiles of baseline 3MS score, comparing the difference in geometric mean solute clearance across tertile of baseline 3MS score via linear regression.

We fit discrete Cox regression models to estimate associations between clearance of each log-transformed secretory solute measured at baseline with study outcomes. We utilized a nested adjustment scheme to evaluate for potential confounding characteristics. Model 1 was adjusted for age, sex, self-reported race and ethnicity, educational attainment, and baseline 3MS score. Model 2 was adjusted for the variables in model 1 and diabetes, coronary artery disease, peripheral arterial disease, systolic BP, body mass index, high- and low-density lipoprotein, hemoglobin, high-sensitivity C-reactive protein, and use of statins, aspirin, β blockers, angiotensin converting enzyme inhibitors, and angiotensin receptor blockers. Model 3 was adjusted for the variables in model 2 and eGFR and 24-hour proteinuria. We further fit Cox regression models to estimate associations between plasma concentrations of each of our secretory solutes at baseline with study outcomes, using the same nested adjustment model as described above. We accounted for potential false positives due to multiple comparisons by utilizing a false discovery rate of 5%, implemented with the “qvalue” package in R.^39,40 Missing covariates were multiply imputed using chained equations.⁴¹ The multiple analyses over the imputations were combined using Rubin’s rules to account for the variability in the imputation procedure.⁴² To provide additional evidence that the association of interest was not confounded by lower eGFR, lesser educational attainment, and greater age, we tested for multiplicative interaction between our summary score and each of the following terms: continuous eGFR, categorical educational attainment, and age.

We performed several sensitivity analyses. In order to assess the validity of our definition of cognitive decline, we defined the outcome as a sustained decline in 3MS score from baseline of more than five points and a drop below an age-specific threshold for cognitive impairment (3MS score <85 in participants aged <65 years, <80 in participants aged 65–80 years, and <75 in participants aged >80 years), and the first instance of decline in 3MS score of more than five points from baseline regardless of whether it was sustained. In another sensitivity analysis, we utilized a Fine–Gray approach to account for the competing risk of death; in a final sensitivity analysis, we examined the associations of biomarker clearances with a composite outcome of sustained cognitive decline or death.^43–45 In each sensitivity analysis, we fit discrete Cox regression models to estimate associations between clearance of each log-transformed secretory solute at baseline with the outcomes, using a nested adjustment scheme as described above.

We also analyzed the associations between our summary score of renal solute clearance with performance on each supplemental cognitive test (Trails A and B, and Buschke SRT). For each participant with available scores for each of these tests, we selected the first instance of each test and the participant’s summary solute score. We fit linear regression models to estimate these associations, adjusted for the components of model 3 as described above. Additionally, for each participant with at least two instances of supplemental cognitive tests, we tested the associations between the summary score and changes in each supplemental cognitive test, adjusting for the components of model 3 as described above. The second instance of each test was administered a median of 2 years (interquartile range [IQR] 1.9–2.1 years) after the first instance.

All analyses were performed using R v4.0.2 (R Foundation for Computing, Vienna, Austria).⁴⁶

Results

Characteristics of the Study Population

Among 2362 participants, the mean age was 57.6±10.9 years; 1059 (45%) were women, and 1022 (43%) had diabetes. The mean eGFR was 45.9±14.9 ml/min per 1.73 m², and the median 24-hour urine protein excretion was 150 mg/day (IQR 70–730 mg/day). Compared with participants who had higher baseline 3MS scores, participants who had lower 3MS scores tended to be older, have more comorbidities, and have lower educational attainment (Table 1). Participants with lower baseline 3MS scores also had lower baseline eGFR, higher proteinuria, and higher systolic BP. Clearance of secretory solutes were only moderately correlated (Supplemental Table 1). Plasma concentrations of secretory solutes were weakly correlated at baseline (Supplemental Table 2). Baseline cognition scores were not significantly associated with over- or undercollection of 24-hour urine samples (Supplemental Figures 2 and 3).

Table 1.

Study population baseline characteristics (N=2362), overall and by tertiles of baseline 3MS score

Variable	Overall (N=2362)	3MS <93 (N=716)	3MS 93–97 (N=855)	3MS ≥98, (N=791)
Baseline 3MS score, median (IQR)	96 (91–98)	89 (87–91)	95 (94–96)	99 (98–100)
Age (yr)	57.6 (10.9)	60 (10.2)	57.6 (10.8)	55.4 (11.2)
Women, n (%)	1059 (45)	298 (42)	387 (45)	374 (47)
Self-reported race and ethnicity, n (%)
Hispanic	188 (8)	108 (15)	52 (6)	28 (4)
Non-Hispanic Black	867 (37)	372 (52)	328 (38)	167 (21)
Non-Hispanic White	1202 (51)	193 (27)	433 (51)	576 (73)
Othera	105 (4)	43 (6)	42 (5)	20 (3)
Educational level, n (%)
Less than high school	280 (12)	185 (26)	80 (9)	15 (2)
High school graduate	433 (18)	184 (26)	152 (18)	97 (12)
Some college	726 (31)	207 (29)	302 (35)	217 (27)
College graduate or higher	922 (39)	140 (20)	321 (38)	461 (58)
Current smoker	255 (11)	88 (12)	99 (12)	68 (9)
Alcohol use	1620 (69)	394 (55)	598 (70)	628 (79)
Medical history, n (%)
Diabetes mellitus	1022 (43)	380 (53)	370 (43)	272 (34)
Coronary artery disease	450 (19)	169 (24)	158 (18)	123 (16)
Peripheral vascular disease	117 (5)	43 (6)	46 (5)	28 (4)
Medication use, n (%)
Aspirin	987 (42)	316 (44%)	357 (42%)	314 (40%)
Statins	1259 (53)	407 (57%)	464 (54%)	388 (49%)
β blockers	1089 (46)	375 (52%)	407 (48%)	307 (39%)
ACE inhibitor or ARB	1589 (67)	501 (70%)	593 (69%)	495 (63%)
Systolic BP (mmHg))	125.4 (20.1)	129.4 (20.5)	126.1 (20.3)	121 (18.5)
Diastolic BP (mmHg))	71.3 (12.4)	71.6 (13.6)	71.1 (12)	71.2 (11.7)
BMI (kg/m2)	32 (7.6)	32.9 (7.5)	32 (7.6)	31 (7.7)
eGFR (ml/min per 1.73 m2)	45.9 (14.9)	42.6 (14)	45.6 (14.7)	49.4 (15.3)
24-hour urine protein (g/d), median (IQR)	0.15 (0.07–0.73)	0.2 (0.1–0.9)	0.2 (0.1–0.9)	0.1 (0.1–0.5)
Hemoglobin (g/dl)	12.8 (1.7)	12.4 (1.7)	12.8 (1.8)	13.1 (1.7)
LDL cholesterol (mg/dl)	103.2 (34.4)	102.4 (37.1)	103.1 (35.4)	103.9 (30.7)
HDL cholesterol (mg/dl)	48.2 (15.6)	46.4 (14.2)	48.4 (15.8)	49.5 (16.5)
CRP, median (IQR)	2.4 (1–5.8)	2.8 (1.2, 6.6)	2.5 (1–6)	1.9 (0.9, 4.6)

All values are mean (standard deviation) unless otherwise noted. Alcohol use was dichotomized as none versus any in the past 12 months; tobacco use was dichotomized as current tobacco use versus no tobacco use at time of cohort entry. ACE, angiotensin converting enzyme; ARB, angiotensin receptor blocker; BMI, body mass index; LDL, low-density lipoprotein; HDL, high-density lipoprotein; CRP, C reactive protein.

^aAmerican Indian or Alaska Native, Asian, or Pacific Islander.

Associations of Secretory Solute Clearances with Cognitive Impairment and Decline

At baseline, lower kidney clearances of all eight secretory solutes were associated with lower 3MS scores (Figure 1). The median number of follow-up 3MS examinations per participant was six (IQR 4–8). A total of 247 participants experienced cognitive decline over a median follow-up period of 9.1 years (IQR 4.9–11.8 years), yielding an incidence of cognitive decline of 1.3 events per 100 person-years (95% CI, 1.1 to 1.4 events per 100 person-years). After adjustment for eGFR, proteinuria, and other potential confounding variables, lower kidney clearances of five of the secretory solutes were significantly associated with cognitive decline (cinnamoylglycine, isovalerylglycine, kynurenic acid, pyridoxic acid, and tiglylglycine; Table 2). Each 50% lower clearance of these solutes was associated with a 17%–34% higher incidence of cognitive decline after adjustment.

Figure 1.

Box plots of biomarker clearances by tertiles of baseline 3MS score. All P values <0.001.

Table 2.

Associations of secretory biomarker clearance and sustained cognitive decline per 50% lower biomarker clearance

Secretory Solute	HR (95% CI)				Model 3 P Value
	Unadjusted	Model 1	Model 2	Model 3
Cinnamoylglycine	1.19 (1.09 to 1.3)	1.19 (1.08 to 1.31)	1.2 (1.09 to 1.32)	1.17 (1.06 to 1.3)a	0.002a
Indoxyl sulfate	1.23 (1.07 to 1.42)	1.25 (1.08 to 1.45)	1.27 (1.09 to 1.47)	1.2 (1.01 to 1.44)	0.04
Isovalerylglycine	1.35 (1.19 to 1.54)	1.27 (1.11 to 1.46)	1.29 (1.12 to 1.48)	1.24 (1.07 to 1.44)a	0.005a
Kynurenic acid	1.38 (1.2 to 1.6)	1.37 (1.17 to 1.6)	1.37 (1.17 to 1.61)	1.34 (1.11 to 1.6)a	0.002a
p-cresol sulfate	1.12 (1.01 to 1.25)	1.12 (0.99 to 1.27)	1.13 (1 to 1.28)	1.07 (0.92 to 1.23)	0.4
Pyridoxic acid	1.24 (1.11 to 1.39)	1.21 (1.07 to 1.37)	1.23 (1.08 to 1.4)	1.18 (1.02 to 1.37)a	0.03a
Tiglylglycine	1.27 (1.14 to 1.42)	1.21 (1.07 to 1.36)	1.22 (1.08 to 1.38)	1.17 (1.02 to 1.35)a	0.03a
Xanthosine	1.18 (1.06 to 1.31)	1.14 (1.02 to 1.27)	1.16 (1.04 to 1.29)	1.12 (0.99 to 1.26)	0.06
Summary score (per 10-point decrement)	1.44 (1.24 to 1.67)	1.41 (1.19 to 1.67)	1.45 (1.22 to 1.71)	1.44 (1.17, 1.77)	5.58E-04

Model 1: Adjusted for age, sex, race and ethnicity, education, and baseline 3MS. Model 2: Adjusted for the components of model 1 plus diabetes, coronary artery disease, peripheral arterial disease, systolic BP, BMI, HDL, LDL, hemoglobin, log-transformed CRP, and use of statins, aspirin, β blockers, ACE inhibitors, and ARBs. Model 3: Adjusted for the components of model 2 plus eGFR and log-transformed 24-hour proteinuria. HR, hazard ratio.

^aIndicates significance when controlling false discovery rate at 5%.

A 10-point lower summary secretion score, which comprises the average of all eight secretory clearances, was associated with an estimated 44% higher risk of cognitive decline after adjustment (95% CI, 17% to 77%).

We tested for effect modification by categories of age, eGFR, and educational attainment. The association of the summary secretion score with cognitive decline was statistically similar across these characteristics (P=0.22 for interaction by age, 0.27 for interaction by eGFR, and 0.46 for interaction by education).

Sensitivity Analyses

The associations between clearance of secretory solutes and cognitive decline were numerically similar when the cognitive decline outcome further required a final 3MS value below age-specified thresholds for cognitive impairment (<75, <80, or <85 for ages >80, 65–80, and <65 years, respectively; Supplemental Table 3). Associations were slightly attenuated by defining cognitive decline as an unrepeated drop of more than 5 points (Supplemental Table 4), and by accounting for the competing risk of death via a Fine–Gray approach (Supplemental Table 5). Associations were similar when considering sustained cognitive decline or death as a composite outcome (Supplemental Table 6).

Associations of Plasma Secretory Solute Levels with Cognitive Decline

After adjustment for potential confounding characteristics, plasma concentrations of the secretory solutes were not associated with risk of cognitive decline (Table 3).

Table 3.

Associations of plasma biomarker level and sustained cognitive decline per 50% plasma serum biomarker levels

Secretory Solute	HR (95% CI)				Model 3 P Value
	Unadjusted	Model 1	Model 2	Model 3
Cinnamoylglycine	0.96 (0.9 to 1.04)	0.97 (0.9 to 1.04)	0.98 (0.91 to 1.05)	1 (0.92 to 1.08)	0.96
Indoxyl sulfate	0.86 (0.75 to 0.99)	0.93 (0.8 to 1.06)	0.91 (0.78 to 1.04)	0.97 (0.84 to 1.13)	0.72
Isovalerylglycine	0.99 (0.83 to 1.17)	0.95 (0.8 to 1.13)	0.98 (0.82 to 1.17)	1.09 (0.9 to 1.32)	0.38
Kynurenic acid	0.94 (0.79 to 1.11)	0.92 (0.77 to 1.1)	0.93 (0.78 to 1.11)	1.13 (0.9 to 1.43)	0.29
p-cresol sulfate	0.8 (0.72 to 0.88)	0.88 (0.8 to 0.97)	0.88 (0.8 to 0.98)	0.91 (0.82 to 1.01)	0.09
Pyridoxic acid	1.08 (0.98 to 1.18)	1 (0.91 to 1.11)	1.00 (0.91 to 1.11)	1.05 (0.94 to 1.17)	0.38
Tiglylglycine	0.9 (0.79 to 1.02)	0.86 (0.75 to 0.99)	0.90 (0.78 to 1.04)	0.97 (0.83 to 1.14)	0.7
Xanthosine	0.98 (0.89 to 1.08)	1.05 (0.95 to 1.15)	1.03 (0.94 to 1.14)	1.07 (0.96 to 1.19)	0.2

Model 1: Adjusted for age, sex, race and ethnicity, education, and baseline 3MS. Model 2: Adjusted for the components of model 1 plus diabetes, coronary artery disease, peripheral arterial disease, systolic BP, BMI, HDL, LDL, hemoglobin, log-transformed CRP, and use of statins, aspirin, β blockers, ACE inhibitors, and ARBs. Model 3: Adjusted for the components of model 2 plus eGFR and log-transformed 24-hour proteinuria.

Associations Between Summary Score and Trails A, Trails B, and Buschke SRT Scores

Lower summary secretion scores were associated with slower completion of the Trails A and B tests. After adjustment for eGFR, proteinuria, and other potential confounding variables, each 10-point decrement in the summary score was associated with a 2.4-second (95% CI, 0.1 to 4.6 seconds) slower time to complete the Trails A test, and a 6.8-second (95% CI, 2.1 to 11.5 seconds) slower time to complete the Trails B test (Table 4). After adjusting for potential confounders, our summary score was not associated with higher or lower scores on the Buschke SRT. After adjustment for multiple confounding variables, the summary score was not associated significantly with changes in the Trails A, Trails B, or Buschke SRT (Supplemental Table 7).

Table 4.

Difference in Trails A, Trails B, and Buschke SRT scores per 10-point increase in summary score

Difference in test scores per 10-point increase in summary score	Unadjusted	Model 1	Model 2	Model 3	Model 3 P value
Trails A time (in seconds) (n=2362)	6.5 (4.7, 8.3)	3.3 (1.6, 5.0)	3.1 (1.4, 4.9)	2.4 (0.1, 4.6)	0.04a
Trails B time (in seconds) (n=2362)	17.2 (13.3, 21.2)	6.8 (3.3, 10.2)	6.3 (2.9, 9.8)	6.8 (2.1, 11.5)	0.004a
Buschke SRT score (n=1919)	−0.2 (-0.3, –0.1)	0.0 (–0.2, 0.1)	0.0 (–0.2, 0.1)	0.0 (–0.1, 0.1)	0.98

Positive values of Trails A and B indicate longer time to test completion (implying worse cognitive function); negative values of the Bushke SRT score are associated with worse cognitive function. Model 1: Adjusted for age, sex, race and ethnicity, education, and baseline 3MS. Model 2: Adjusted for the components of model 1 plus diabetes, coronary artery disease, peripheral arterial disease, systolic BP, BMI, HDL, LDL, hemoglobin, log-transformed CRP, and use of statins, aspirin, β blockers, ACE inhibitors, and ARBs. Model 3: Adjusted for the components of model 2 plus eGFR and log-transformed 24-hour proteinuria.

^aIndicates statistical significance.

Discussion

In a large population of patients with CKD without baseline cognitive impairment, we demonstrated associations between impaired tubular secretory clearance of a panel of uremic solutes and global incident cognitive decline as measured by performance on the 3MS. These associations were significant, even after controlling for conventional measures of glomerular filtration, including eGFR and 24-hour proteinuria. By contrast, plasma levels of these solutes were not associated with incident cognitive decline. This study highlights the potential contribution of kidney tubular clearance of secretory solutes to maintain normal cognitive function.

Previous studies have suggested that markers of impaired secretory clearance may be associated with cognitive impairment. For instance, a study of 199 patients with CKD and 84 controls with normal renal function found that elevated serum concentrations of indoxyl sulfate were associated with worse cognitive function.¹² Elevated serum concentrations of indoxyl sulfate has also been associated with cognitive impairment in a cross-sectional analysis of 260 patients with kidney failure on hemodialysis.¹⁴ The same study did not find a statistically significant association between serum levels of p-cresol sulfate and cognitive impairment, however. Kynurenic acid, a tryptophan metabolite, has been shown to accumulate in kidney failure and has been hypothesized to contribute to uremic symptoms.^47,48 Although these cross-sectional studies established associations, a noteworthy limitation is their inability to distinguish whether elevated serum levels of these solutes preceded cognitive impairment. Our study expands meaningfully on this literature by measuring the clearance of highly protein-bound solutes and their associations with future cognitive decline.

Our findings are biologically plausible. OAT1 and OAT3, expressed on the basolateral membrane of kidney tubules, mediate secretion of protein-bound secretory solutes from the plasma into the urine; these transporters are also expressed in the blood-brain barrier, where they are proposed to have a similar role in regulating the composition of cerebrospinal fluid.^49–52 Animal models have shown decreased efflux of uremic solutes from cerebrospinal fluid to blood in uremic conditions, possibly because of inhibition of OAT3.⁴⁹ Experimental models of CKD show decreased renal expression of OAT1 and OAT3.^53,54 It is possible that long-standing elevated serum levels of uremic solutes may cause elevated cerebrospinal levels of various uremic solutes, which are toxic to cells in the central nervous system.^55–57 For instance, indoxyl sulfate has been shown to induce cell death in neurons in a dose-dependent fashion.⁵⁸ Decreased cinnamoylglycine clearance has been associated with increased peroxisome proliferator-activated receptor α in mice; peroxisome proliferator-activated receptor α may have neuroprotective properties via metabolic and anti-inflammatory pathways.^59,60 Isovalerylglycine is a derivative of isovaleryl-CoA, which has been suggested to exert neurotoxic effects via oxidative stress on the basis of in vitro research in the rat brain cortex.⁶¹ Kynurenic acid is a tryptophan metabolite that is suspected of causing cognitive impairment via antagonizing the action of glutamate at N-methyl-D-aspartate (NMDA) receptors throughout the central nervous system.⁶² Pyridoxic acid is a metabolite of vitamin B6, which is elevated in patients with Alzheimer’s and mild cognitive impairment versus healthy controls; underlying mechanisms have not been thoroughly investigated but may be related to inhibition of neurite outgrowth.^63,64 Tiglylglycine is a product of isoleucine catabolism that has been suggested as a marker of mitochondrial dysfunction, which is postulated to play a role in several neurodegenerative diseases.^65,66 We found that lower clearance of cinnamoylglycine, indoxyl sulfate, isovalerylglycine, kynurenic acid, and tiglylglycine were associated with cognitive decline; however, higher baseline plasma levels of these same solutes were not associated with cognitive decline. In this setting, despite prior results, our data do not clearly support the roles of these solutes as causative molecules of cognitive decline. Possibly, decreased secretory clearance of these solutes may serve as a marker of cerebrospinal concentrations of uremic solutes; however, we find it more likely that decreased secretory clearance may lead to chronic elevation of other yet unidentified secretory solutes that may cause cognitive impairment.

Regarding specific domains of cognitive function in our study, we were able to analyze data from a subset of patients who had Trails A, Trails B, and Buschke SRT performed several years after their initial study visit. We found significant associations between lower summary scores and higher prevalent times (indicating worse performance) on the Trails A and B tests, which are predominantly tests of executive function.^30,67 We did not observe significant associations between our summary score and performance on the Buschke SRT, which is a test of verbal memory.³¹ Further, our summary score was not significantly associated with changes in these supplemental tests, although our follow-up was limited. This may suggest that worse tubular secretory clearance of uremic solutes has a greater effect on executive function than verbal memory, as has been previously suggested, although our results could be affected by multiple time-varying factors since baseline, or concomitant prevalent factors at time of testing.^68,69

A novel approach in this study was the development of the summary renal clearance score—a singular metric containing data on the tubular clearance of all of the secretory solutes of interest. Although several previous studies have evaluated decreased glomerular filtration with cognitive impairment, we have demonstrated the associations of decreased secretory function with cognitive decline, independent of eGFR, rather than relying on potentially variable serum levels of specific solutes.^2–5 A further innovation in our study was the longitudinal evaluation of 3MS scores over time; this approach addresses the limitations of previous cross-sectional studies, which could identify associations without temporality. We have demonstrated that impaired secretory clearance preceded the outcome of cognitive decline.

This study has several notable strengths. First, it was performed in a cohort of patients with a wide range of CKD severity. We analyzed patients without baseline cognitive impairment and controlled for a variety of possible covariates. We analyzed clearance of multiple secretory solutes as a method to determine tubular secretory clearance, which expands on prior studies associating increased serum levels of solutes with cognitive impairment. However, our study does have several limitations. First, a large number of medications can inhibit the organic anion transporters and decrease the secretory clearance observed in this study; we were not able to adjust for all such medications. Considering the hundreds of medications that can exert such effects, controlling for them is likely not feasible with this study design. Second, the 3MS is known to have ceiling effects, which may limit its ability to detect cognitive decline in high-functioning individuals. It was not feasible with the study design to adjust fully for the multiple factors that could contribute to cognitive decline, including clinically silent strokes, toxin exposures, brain injury, and other neurologic conditions; as with all cohort studies, unmeasured confounding remains a limitation. Third, we lacked baseline measures of the Trails A, Trails B, and Buschke SRT, which limited our analyses to global cognitive decline rather than domain-specific decline. Our analyses of these additional tests may have been additionally affected by confounders that changed since baseline, which we were unable to adjust for. Finally, the cohort was composed of research volunteers who were closely followed in nephrology subspecialty clinic, which may limit generalizability to other CKD populations.

In conclusion, we demonstrated that decreased tubular secretory clearance of cinnamoylglycine, indoxyl sulfate, isovalerylglycine, kynurenic acid, and tiglylglycine were significantly associated with global cognitive decline in a large cohort of patients with CKD, independent of baseline eGFR and proteinuria. These findings highlight the importance of tubular secretory function in clearing potentially deleterious solutes and may suggest novel mechanisms for the development of cognitive impairment in patients with CKD, or possible therapeutic targets in these patients. Further work is needed to determine the domains of cognition most affected by decreased secretory clearance and to determine the mechanisms underpinning the associations between lower secretory clearance and cognitive decline.

Disclosures

N. Bansal reports an advisory or leadership role for Kidney360 (Associate Editor). M.A. Dobre reports honoraria from Relypsa, Inc., and Tricida, Inc., and an advisory or leadership role for Relypsa (Resistant Hypertension Working Group) and Tricida, Inc. (Metabolic Acidosis Working Group). A.N. Hoofnagle reports consultancy for Kilpatrick Townsend & Stockton LLP; ownership interest in Seattle Genetics; research funding from Waters; patents or royalties from SISCAPA Assay Technologies; an advisory or leadership role for Clinical Chemistry (Associate Editor); and other interests or relationships as an expert witness for Kilpatrick, Townsend, and Stockton, LLC. B. Kestenbaum reports consultancy for and honoraria from Reatta Pharmaceuticals. M. Kurella Tamura reports honoraria from the American Federation for Aging Research, and an advisory or leadership role for Beeson External Advisory Committee, the CJASN Editorial Board, and the Clin-Star Advisory Board. S.E. Rosas reports consultancy for Astellas (consultant for event adjudication for a study), AstraZeneca, Fibrogen, and Reata; research funding from AstraZeneca, Bayer, Ironwood, and the National Institutes of Health National Institute of Diabetes and Digestive and Kidney Diseases; honoraria in 2019 from Bayer and Reata; and an advisory or leadership role for ACKD (editorial board), CJASN, National Kidney Federation Scientific Advisory Board, and National Kidney Federation New England Medical Advisory Board. M.J. Sarnak reports consultancy for Akebia, being on the steering committee of a trial (funds come to the Division of Nephrology), and consultancy for Cardurian; ownership interest in Lilly (spouse is an employee); and research funding from the National Institute of Health. S. Seliger reports consultancy for Tricida, Inc. (Endpoint Adjudication Committee); research funding from Kadmon Pharmaceuticals, Palladio Biosciences, Reata Pharmaceuticals, Roche Diagnostics, Inc., and Sanofi US; patents or royalties from the University of Maryland, Baltimore, and University of Texas, Southwestern (Methods for Assessing Differential Risk for Developing Heart Failure); and an advisory or leadership role for CJASN (Associate Editor), a member of Medical Review Board ESRD Network 5, a member of Endpoint Adjudication Committee VALOR-CKD trial (Tricida, Inc.), and a member of the Editorial Board for Circulation. J. Sondheimer reports consultancy for Davita Kresge Hemodialysis Unit (Medical Director); ownership interest in Doximity (250 shares); and an advisory or leadership role for DaVita (Physician Advisory Board for Information Technology). K. Yaffe reports an advisory or leadership role for Eli Lilly (DSMBs) and studies sponsored by the National Institute of Health and is on the Board of Directors for Alector. L.R. Zelnick reports consultancy for the Veterans Medical Research Foundation and is statistical editor for CJASN. All remaining authors have nothing to disclose.

Funding

This study was supported by R01 DK107931 (to B. Kestenbaum) and T32 DK007467 (B. Lidgard). Additional support was provided by an unrestricted fund from the Northwest Kidney Centers. Funding for the CRIC Study was obtained under a cooperative agreement from the National Institute of Diabetes and Digestive and Kidney Diseases (U01DK060990, U01DK060984, U01DK061022, U01DK061021, U01DK061028, U01DK060980, U01DK060963, U01DK060902, and U24DK060990). In addition, this work was supported in part by: the Perelman School of Medicine at the University of Pennsylvania Clinical and Translational Science Award NIH/NCATS UL1TR000003, Johns Hopkins University UL1TR-000424, University of Maryland GCRC M01 RR-16500, Clinical and Translational Science Collaborative of Cleveland, UL1TR000439 from the National Center for Advancing Translational Sciences (NCATS) component of the National Institutes of Health and NIH roadmap for Medical Research, Michigan Institute for Clinical and Health Research (MICHR) UL1TR000433, University of Illinois at Chicago CTSA UL1RR029879, Tulane COBRE for Clinical and Translational Research in Cardiometabolic Diseases P20 GM109036, Kaiser Permanente NIH/NCRR UCSF-CTSI UL1 RR-024131, Department of Internal Medicine, University of New Mexico School of Medicine Albuquerque, NM, R01DK119199. Additionally, some authors were supported by R01 DK069406 and R01 HL141846 (MD).

Data Sharing Statement

Original data created for the study are available in the CRIC repository under the following accession number: 10.1097/01.asn.0000070149.78399.ce.

Author Contributions

N. Bansal and B. Kestenbaum were responsible for supervision; N. Bansal, B. Kestenbaum, and B. Lidgard were responsible for conceptualization; A.N. Hoofnagle, B. Kestenbaum, M. Kurella Tamura, B. Lidgard, S. Seliger, and L.R. Zelnick were responsible for the methodology; A. Hoofnagle, B. Kestenbaum, and B. Lidgard were responsible for the investigation; A.N. Hoofnagle and B. Kestenbaum were responsible for resources; B. Kestenbaum was responsible for funding acquisition and project administration; B. Kestenbaum, B. Lidgard, and L.R. Zelnick were responsible for the formal analysis and visualization; B. Lidgard wrote the original draft of the manuscript; B. Lidgard and L.R. Zelnick were responsible for validation; L.R. Zelnick was responsible for data curation and software; and all authors reviewed and edited the manuscript.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2021111435/-/DCSupplemental.

Supplemental Table 1. Pearson’s correlation coefficients between log-transformed clearance of each uremic solute at baseline.

Supplemental Table 2. Pearson’s correlation coefficients between log-transformed baseline plasma levels of each uremic solute.

Supplemental Table 3. Associations of secretory biomarker clearance and sustained cognitive decline of more than five points and a drop below the threshold for cognitive impairment.

Supplemental Table 4. Associations of secretory biomarker clearance and unrepeated cognitive decline of more than five points.

Supplemental Table 5. Associations of secretory biomarker clearance and sustained cognitive decline, per 50% lower biomarker clearance, accounting for competing risk of death with Fine–Gray approach.

Supplemental Table 6. Association of secretory biomarker clearance and sustained cognitive decline or death.

Supplemental Table 7. Average decrease in Trails A, Trails B, and Buschke SRT scores per 10-point increase in summary score over median of 2 years (IQR 1.9–2.1 years).

Supplemental Figure 1. CONSORT diagram.

Supplemental Figure 2. Twenty-four-hour urine volume by baseline 3MS score deciles.

Supplemental Figure 3. Twenty-four-hour creatinine excretion by baseline 3MS score deciles.