Glomerular structural–functional relationship models of diabetic nephropathy are robust in type 1 diabetic patients

Abstract

Studies of structural–functional relationships have improved understanding of the natural history of diabetic nephropathy (DN). However, in order to consider structural end points for clinical trials, the robustness of the resultant models needs to be verified. This study examined whether structural–functional relationship models derived from a large cohort of type 1 diabetic (T1D) patients with a wide range of renal function are robust. The predictability of models derived from multiple regression analysis and piecewise linear regression analysis was also compared. T1D patients (n = 161) with research renal biopsies were divided into two equal groups matched for albumin excretion rate (AER). Models to explain AER and glomerular filtration rate (GFR) by classical DN lesions in one group (T1D-model, or T1D-M) were applied to the other group (T1D-test, or T1D-T) and regression analyses were performed. T1D-M-derived models explained 70 and 63% of AER variance and 32 and 21% of GFR variance in T1D-M and T1D-T, respectively, supporting the substantial robustness of the models. Piecewise linear regression analyses substantially improved predictability of the models with 83% of AER variance and 66% of GFR variance explained by classical DN glomerular lesions alone. These studies demonstrate that DN structural–functional relationship models are robust, and if appropriate models are used, glomerular lesions alone explain a major proportion of AER and GFR variance in T1D patients.

INTRODUCTION

The annual incidence of diabetic nephropathy (DN), by far the most common cause of end-stage renal disease (ESRD) in the USA, continues to increase [1] despite overall improvements in glycemia and blood pressure control and the availability of renin–angiotensin system-blocking agents (RASB) [2]. Although most persons with diabetes reaching ESRD have type 2 diabetes [1], type 1 diabetes (T1D) results in >4500 new ESRD cases per year in the USA, an increase of >35% since 1990 [3]. Studies of relationships between DN lesions and renal function help to better understand the natural history of DN. Lesions that are closely associated with increased urinary albumin excretion rate (AER) or glomerular filtration rate (GFR) loss are more likely to play a role in the pathogenesis of these conditions as well as to predict DN progression. Our previous studies described important relationships between renal structure and function in T1D [4–6]. However, given the small number of patients, these analyses were vulnerable to overfitting and require validation in a larger cohort. The goals of the present studies in a large cohort of T1D patients across a wide range of renal function are 2-fold: to study the robustness of the DN structural–functional relationship models and to compare multiple versus piecewise linear regression analyses in these structural–functional relationship models.

MATERIALS AND METHODS

Patients

A total of 160 Caucasian T1D research volunteers who had renal function studies and research native kidney biopsies performed between 1982 and 2001 in the General Clinical Research Center (GCRC) at the University of Minnesota (UMN) were included in this study. All research procedures were in accordance with the ethical standards of the responsible Institutional Review Board committee at the UMN. Informed consents were obtained from all subjects prior to enrolment. Inclusion criteria were T1D duration of at least 8 years and adequate tissue from these research kidney biopsies for the measurements described below. Exclusion criteria were serum creatinine >180 μmol/L (GFR <40 mL/min/1.73 m²), presence of nondiabetic renal disease, single kidney, anticoagulant therapy and known secondary causes of hypertension, including renal artery stenosis and endocrinopathies. Ninety-one of these patients were also included in one of our previous studies of structural–functional relationships in T1D [5].

Glycated hemoglobin was measured by a nonautomated ion-exchange chromatography micro-column from 1983 to 1986 (HbA₁ reference range: 5.5–8.5%), by an automated high-performance ion-exchange liquid-chromatography (HPLC; Bio-Rad) from 1986 to 1992 (HbA₁ reference range: 5.1–7.3%; HbA_1c reference range: 4.3–6.1%) and by the same method standardized to diabetic control and complications trial values since then (HbA₁ reference range: 5.4–7.4%; HbA_1c reference range: 4.3–6.0%). Results prior to 1992 were converted to current HbA_1c standards using published equations [7]. Serum and urinary creatinine were measured by automated methods using the Jaffe's reaction and, more recently, by enzymatic methods. AER was measured in at least three 24 h or timed overnight sterile urine collections by a sensitive fluorimetric assay [8]. Patients were classified as normoalbuminuric (AER <20 µg/min; n = 62), microalbuminuric (AER 20–200 µg/min; n = 63) or proteinuric (>200 µg/min; n = 35) based on at least two out of three sequential AER measurements being in the same range. Median AER values were used for analyses. The GFR at the time of renal biopsy was measured by iothalamate (n = 27) or iohexol (n = 17) clearance or as the mean of multiple 24-h creatinine clearances (n = 118) from the samples collected at GCRC, the latter previously shown to highly correlate with inulin clearances [4]. Blood pressure was measured at least 10 times with the patients at rest in the GCRC and the values provided are the mean of these repeated measurements.

Renal structural studies

Research kidney biopsies performed under ultrasound guidance were immediately processed for electron microscopy (JEOL CX100, Japan) [4, 5]. Random glomerular profiles were prepared as previously described [5, 9]. At least three glomeruli per biopsy with complete profiles and intact Bowman's capsules were examined. Overlapping electron micrographs at ×3900 were obtained to make montages of the entire glomerular profile to estimate fractional volume of mesangium per glomerulus [Vv(Mes/glom)] and surface density of the peripheral glomerular basement membrane per glomerulus [Sv(PGBM/glom)]. Vv(Mes/glom) was estimated by point counting and Sv(PGBM/glom) by an intercept counting method [5]. Electron micrographs at ×12 000, obtained via a systematic uniform random sampling protocol, were used to estimate glomerular basement membrane (GBM) width using the orthogonal intercept method [10] and fractional volume of mesangial matrix per glomerulus [Vv(MM/glom)] and fractional volume of mesangial cells per glomerulus [Vv(MC/glom)] using point counting [4, 5]. Reference values for glomerular structural parameters were derived from 76 (33M/43F) age- and sex-matched normal living kidney transplant donors as we previously published [5]. They were 37.6 ± 12.1 years old (19–64 years). The mean ± 2 SD of the measurements in these donors defined the reference range for glomerular structures.

Statistical analyses

Statistica (10.0, Statsoft) and Microsoft Office Excel (2003, Microsoft) were used for statistical analyses. Data are expressed as mean ± SD, unless otherwise specified. Inter-group comparisons were done by Student's t-test, or χ². Values for P < 0.05 were considered statistically significant. AER values were logarithmically transformed. T1D subjects ranked by AER values were distributed into two matched sets by sequentially assigning every other subject to one of the sets (Table 1). One of these sets, the T1D-model (T1D-M), was arbitrarily chosen to derive structural–functional relationship models using standard multiple regression analysis with AER or GFR as dependent variables and Vv(Mes/glom), GBM width and Sv(PGBM/glom) as predictors. Intercept was allowed in the model. The tolerance threshold for inclusion of predictors in the model was set to 0.0001. To verify the robustness of these models, multiple regression analyses equations derived from the T1D-M were tested in the other T1D set, the T1D-test (T1D-T) set. Briefly, predicted AER_log values from the T1D-T patients were calculated by multiple regression analyses equations derived from the T1D-M subjects. Total variance, regression variance, error variance and coefficient of determination and adjusted coefficient of determination were accordingly calculated from the observed and predicted values of AER_log and GFR (Figure 1). The F-test was used to determine the statistical significance of the regression models. Piecewise linear regression analyses was performed with AER or GFR as dependent variables and Vv(Mes/glom), GBM width and Sv(PGBM/glom) as predictors. The quasi-Newton method was used by the software to minimize the loss function. Breakpoints were estimated by the software. The analyses were repeated after patients who were on angiotensin-converting enzyme inhibitors or angiotensin receptor blockers were removed from the analyses.

Table 1.

Demographic, clinical and glomerular structural parameters

	T1D (total), n = 160	T1D-M, n = 80	T1D-T, n = 80	P-value (T1D-M versus T1D-T)
Age (years)	35 (19–64)	34 (20–64)	38 (19–57)	NS
Age at diabetes onset (years)	12 (0–39)	11 (0–33)	13 (2–39)	0.03
T1D duration (years)	22 ± 9	23 ± 9	22 ± 9	NS
HbA1c (%)	8.9 ± 1.9	8.7 ± 1.9	9.1 ± 1.9	NS
SBP (mmHg)	124 ± 12	125 ± 11	124 ± 12	NS
DBP (mmHg)	74 ± 8	74 ± 7	74 ± 8	NS
AER (μg/min)	30 (2–4630)	30 (2–3868)	31 (2–4630)	NS
GFR (mL/min/1.73 m2)	99 ± 29	100 ± 28	98 ± 31	NS
Vv(Mes/glom)	0.36 ± 0.13	0.36 ± 0.13	0.36 ± 0.13	NS
Vv(MM/glom)	0.20 ± 0.09	0.20 ± 0.08	0.21 ± 0.09	NS
Vv(MC/glom)	0.10 ± 0.04	0.11 ± 0.04	0.10 ± 0.03	NS
GBM width (nm)	578 ± 153	580 ± 156	576 ± 151	NS
Sv(PGBM/glom)	0.09 ± 0.03	0.09 ± 0.03	0.09 ± 0.03	NS

Data are mean ± SD or median range.

T1D, type 1 diabetes; T1D-M, type 1 diabetes model group; T1D-T, type 1 diabetes test group; HbA_1c, hemoglobin A_1c; SBP, systolic blood pressure; DBP, diastolic blood pressure; AER, albumin excretion rate; GFR, glomerular filtration rate; Vv(Mes/glom), fractional volume of mesangium per glomerulus; Vv(MM/glom), fractional volume of mesangial matrix per glomerulus; Vv(MC/glom), fractional volume of mesangial cells per glomerulus; GBM, glomerular basement membrane; Sv(PGBM/glom), surface density of the peripheral glomerular basement membrane per glomerulus.

Reference range for glomerular structural parameters: Vv(Mes/glom): 0.20 ± 0.03; Vv(MM/glom): 0.09 ± 0.02; Vv(MC/glom): 0.08 ± 0.02; GBM: 332 ± 46 and Sv(PGBM/glom): 0.13 ± 0.02.

FIGURE 1:

Observed versus predicted values of albumin excretion rate (AER) (top) and GFR (bottom) for T1D-M (black circles) and T1D-T (white circles) superimposed. T1D-T predicted values are calculated from models obtained in the T1D-M group.

RESULTS

The demographic, clinical and classical DN glomerular structural parameters of the entire T1D cohort and comparisons between T1D-model (T1D-M) and T1D-test (T1D-T) groups are shown in Table 1. All parameters, except for age at diabetes onset that was slightly greater in T1D-T, were similar between T1D-M and T1D-T (Table 1).

Vv/(Mes/glom), Vv/(MM/glom) and Vv/(MC/glom) were increased by 80, 122 and 25%, respectively, in these T1D patients as compared with normal controls. GBM width was increased, on average, by 74%, while Sv(PGBM/glom) was decreased by 69% in the patients with T1D (all P < 0.001).

Correlations between renal function and glomerular structural parameters in the entire type 1 diabetic cohort

AER correlated directly with Vv(Mes/glom) (r = 0.78, P = 0.001), Vv(MM/glom) (r = 0.75, P = 0.001), Vv(MC/glom) (r = 0.49, P = 0.0001) and GBM width (r = 0.67, P = 0.001), and inversely with Sv(PGBM/glom) (r = −0.71, P = 0.001). GFR correlated inversely with Vv(Mes/glom) (r = −0.52, P = 0.0001), Vv(MM/glom) (r = −0.52, P = 0.0001), Vv(MC/glom) (r = −0.34, P = 0.0001) and GBM width (r = −0.36, P = 0.0001), and directly with Sv(PGBM/glom) (r = 0.46, P = 0.0001).

These classical glomerular structural parameters were also interrelated. For example, Vv(Mes/glom) correlated directly with Vv(MM/glom) (r = 0.95, P = 0.001), Vv(MC/glom) (r = 0.66, P = 0.001) and GBM width (r = 0.57, P = 0.0001), and inversely with Sv(PGBM/glom) (r = −0.79, P = 0.001).

Robustness of structural–functional relationship models

Multiple regression analyses—AER

Using standard multiple regression analysis, Vv(Mes/glom), GBM width and Sv(PGBM/glom) explained 70% of AER variance in the T1D-M cohort (R²a = 0.70, P = 0.0000001) through the equation:

$$\text{AE}{\text{R}}_{\log }=0.51+2.7\times \text{Vv}(\text{Mes/glom})+0.002\times \text{GBM}\text{width}-7.5\times \text{Sv}(\text{PGBM/glom})$$ (1)

with each of these glomerular structural parameters being independent predictors of AER. To test the robustness of Equation (1) in T1D patients, Equation (1) was applied to data from the T1D-T cohort and the regression coefficient calculated (see ‘Statistical Analyses’). Thus, 63% of AER variance in the T1D-T group was explained by Vv(Mes/glom), GBM width and Sv(PGBM/glom) (R²a = 0.63, P = 0.000001) using this equation, confirming that T1D-M and T1D-T groups followed very similar structural–functional relationship models for AER. Multiple regression analysis in T1D-M and T1D-T groups combined explained 68% of AER variance (P = 0.0000001) with GBM width and Sv(PGBM/glom) as independent predictor variables.

Multiple regression analyses—GFR

Vv(Mes/glom), GBM width and Sv(PGBM/glom) explained 32% of GFR variance in the T1D-M cohort (R²a = 0.32, P = 0.0001) through the equation:

$$\text{GFR}=123-105\times \text{Vv(Mes/glom)}+0.005\times \text{GBM}\text{width}+126\times \text{Sv}(\text{PGBM/glom})$$ (2)

with Vv(Mes/glom) being the only independent predictor of GFR (P = 0.004). Equation (2) applied to data from the T1D-T cohort explained, 21% of GFR variance by Vv(Mes/glom), GBM width and Sv(PGBM/glom) (R²a = 0.21, P = 0.00002), indicating similar GFR structural–functional relationships in T1D-M and T1D-T groups. Multiple regression analysis in the T1D-M and T1D-T groups combined explained 27% of GFR variance (P = 0.0000001) with Vv(Mes/glom) the only independent predictor variable.

Piecewise linear regression analyses—AER and GFR

Because the piecewise linear regression models developed in T1D-M patients resulted in comparable variance explanations for AER and GFR in the T1D-T cohort, the data from these two groups were combined for the following analyses.

Vv(Mes/glom), GBM width and Sv(PGBM/glom) explained 83% of AER variance through the following equations.

If AER ≤ 42 µg/min:

$$\text{AE}{\text{R}}_{\log }=0.12+1.9\times \text{Vv(Mes/glom)}+0.0006\times \text{GBM}\text{width}+0.6\times \text{Sv}(\text{PGBM/glom})$$ (3)

if AER ≥42 µg/min:

$$\text{AE}{\text{R}}_{\log }=1.51+1.85\times \text{Vv(Mes/glom)}+0.0008\times \text{GBM}\text{width}-6.9\times \text{Sv}(\text{PGBM/glom})$$ (4)

with the model breakpoint at AER of 42 µg/min. Vv(Mes/glom) was the only statistically significant independent predictor of AER values ≤42 μg/min (P = 0.002), while Vv(Mes/glom) (P = 0.0006), GBM width (P = 0.02) and Sv(PGBM/glom) (P = 0.01) were all independent predictors of AER for values above this breakpoint.

Vv(Mes/glom), GBM width and Sv(PGBM/glom) explained 66% of GFR variance in the combined T1D cohort through the following equations.

If GFR ≥99 mL/min/1.73 m²:

$$\text{GFR}=102+44\times \text{Vv(Mes/glom)}-0.009\times \text{GBM}\text{width}+100\times \text{Sv}(\text{PGBM/glom})$$ (5)

If GFR ≤99 mL/min/1.73 m²:

$$\text{GFR}=107-49\times \text{Vv(Mes/glom)}-0.01\times \text{GBM}\text{width}-27\times \text{Sv}(\text{PGBM/glom})$$ (6)

with the breakpoint at a GFR of 99 mL/min/1.73 m². Vv(Mes/glom) was the only statistically significant independent predictor of GFR values ≤99 mL/min/1.73 m² (P = 0.04), while there were no statistically significant independent predictors for GFR values greater than this breakpoint.

All analyses were repeated after exclusion of subjects who were receiving renin–angiotensin system inhibitors (n = 28). This did not result in any meaningful differences compared with the analyses which included the entire cohort (n = 160) (data not shown).

DISCUSSION

This is the first study addressing the robustness of DN structural–functional relationship models in a large cohort of T1D patients. Such robust models with strong renal functional predictive power may be considered representative of how classical diabetic glomerulopathy lesions relate to progression of albuminuria and loss of GFR through most of the natural history of DN in T1D patients. It is important to consider that the structural functional relationships described herein are based on the study of random sections through 3 of about 1.5 million glomeruli [11]. Given variations in glomerular number among persons [11, 12], in lesion severity among glomeruli and between different levels in the same glomerulus, and given the imprecision in GFR measurements [13] and the day-to-day variability of AER [14], it is remarkable that >80% of AER variability and >65% of GFR variability are explained by glomerular structural variables alone. The robust models derived from these studies speak strongly to the diffuse nature of diabetic glomerulopathy.

This study confirms our observations in a far smaller T1D cohort that piecewise linear regression models explain more of GFR and AER variances [15]. This mirrors the natural history of DN which, clinically silent for many years, is followed by progressive renal function decline in the later stages. The breakpoints introduced for AER and GFR by the piecewise linear regression models represent estimates of transition from an earlier more clinically silent phase to a more accelerated phase of disease progression. There are weaker relationships between the DN glomerular structural parameters and renal function in the earlier stages of DN. In fact, the absence of statistically significant independent glomerular structural predictors for GFR values above 99 mL/min/1.73 m² in the piecewise linear regression model contrasts with the strong glomerular predictors for GFR values below this breakpoint. Indeed, patients with normal GFR may have glomerular structural parameters ranging from normal to quite abnormal values. Although the GFR breakpoint is in the ‘normal range’, it is noteworthy, given that in the earlier years of T1D GFR averages ∼140 mL/min/1.73 m² [9, 16], the breakpoint of 99 mL/min/1.73 m² represents, on average, a nearly 30% loss of GFR. One could hypothesize that variations among individuals in glomerular number [11, 12], glomerular hypertrophy [17] and compensatory hemodynamic mechanisms [18] could contribute to the weaker structural functional relationships before these breakpoints. However, there are levels of severity of lesions where virtually all persons will have clinical findings of DN. Of note, the AER breakpoint is several-fold above normal [19], and at least twice as high as the upper limits of the accepted normoalbuminuric range [20] and cannot address variability in tubular protein reabsorption [21].

This study was designed to test the robustness of DN structural functional models by aligning the subjects by AER and selecting alternate subjects resulting in two groups (‘model’ and ‘test’), with very similar demographic, clinical, renal functional and renal structural characteristics. However, this strategy cannot fully address heterogeneity among patients. Thus, among normoalbuminuric T1D patients, those with reduced GFR had worse DN glomerular lesions than those with normal or high GFR [22]. The design of the current study presumably distributed these subjects evenly between the ‘model’ and the ‘test’ groups. Nonetheless, the robustness and reproducibility of these models could be documented despite the underlying heterogeneity in the functional manifestations of DN lesions.

It is recognized that not all glomerular structural parameters were assessed in this study. Thus, we did not include measures of podocyte number [23], foot process width [24], podocyte detachment [25], endothelial cell fenestration [25] or glomerular–tubular junction abnormalities [6], nor were vascular lesions and global glomerulosclerosis addressed [26]. These quite time-consuming measures may have added to the pathophysiologic understanding of altered glomerular permeability in DN and, perhaps to the precision of glomerular structural explanation of functional abnormalities. Perhaps more importantly, the fact that glomerular volume was not systematically measured did not permit total filtration surface per glomerulus (TFS/glom) to be calculated. This parameter was previously shown to be closely correlated in T1D patients with a wide range of GFR [27, 28]. Moreover, in a study of a much smaller cohort of T1D patients with GFRs >40 mL/min/1.73 m² where TFS/glom was included in the piecewise regression analyses, classical DN glomerular lesions alone accounted for 95% of AER variability and 78% of GFR variability [15]. Albeit subject to overfitting, this smaller study suggests that the present study could be underestimating the ability of classical glomerular lesions to account for renal functional abnormalities in T1D patients without advanced GFR loss. Nonetheless, the present much larger study confirms that robust structural–functional models that explain most of the variability in AER and GFR can be developed by using these classical diabetic glomerulopathy parameters alone.

In our smaller studies referred to above, we also found that when the volume fraction of cortex was interstitium, the fraction of cortical tubules which were atrophic and the percentage of globally sclerotic glomeruli which were included with the glomerular parameters in the multiple regression analyses they collectively added only ∼5% to the GFR explanatory models [6]. However, studies by others have demonstrated the importance of tubulointerstitial lesions at later disease stages, i.e. in patients GFR from ∼40 mL/min/1.73 m² to nearly ESRD [29, 30]. Our studies do not address these late stages of DN, because, in such advanced cases, the high frequency of global glomerulosclerosis makes glomerular morphometry impractical. Moreover, the mechanisms driving disease progression at later stages are not necessarily specific to DN. Our 5-year sequential biopsy studies at earlier stages of DN in T1D patients transitioning from normoalbuminuria to microalbuminuria and from microalbuminuria to proteinuria found that only increases in mesangial fractional volume were associated with these functional changes while cortical interstitial fractional volume did not change over this interval [31]. Moreover, there were no differences in cortical interstitial fractional volume between normoalbuminuric T1D patients progressing to proteinuria and ESRD and those remaining normoalbuminuric after over 10 years of follow-up [32].

Thus, although it is likely that at more advanced stages of DN, tubular, interstitial and vascular changes, in addition to classical diabetic glomerulopathy lesions, become increasingly important, most of the functional changes from normoalbuminuria to overt proteinuria and almost 70% of GFR loss are best explained by classical diabetic glomerulopathy lesions in T1D patients. It is also important to consider that this study was carried out in research volunteers, while many studies have been conducted in diabetic patients undergoing clinical biopsies, which are overrepresented by atypical clinical presentations which make it difficult to extrapolate their findings to the more typical natural history of this disease [33]. Moreover, our findings in T1D may not be directly extrapolatable to type 2 diabetic (T2D) patients where greater heterogeneity of lesions in relation to albuminuria has been described in Caucasian and Japanese cohorts [34–36]. Thus, while many T2D patients with albuminuria have typical diabetic glomerulopathy lesions [34–36], many others have a paucity of renal structural abnormalities for their degree of albuminuria [36] or have tubulointerstitial, vascular and/or global glomerulosclerotic lesions out of proportion to the glomerular lesions [36, 37].

In summary, these studies demonstrate the strong and robust relationships of glomerular structure and function in T1D patients with GFR ≥40 mL/min/1.73 m². These studies also reinforce the concept that nonlinear models better reflect the natural history of DN, where for long periods of time lesions may develop in clinical silence, later followed by renal functional abnormalities in association with more advanced diabetic glomerulopathy lesions. Given their close association with albuminuria and GFR, these studies provide a rationale for considering reduction in the late progression of diabetic glomerulopathy lesions as an intermediate surrogate for DN early intervention trials [10].

CONFLICT OF INTEREST STATEMENT

The authors do not have any financial relationships with companies that may have a financial interest in the information contained in this manuscript.