Perspectives in Diabetes
The Absence of a Glycemic Threshold for the
m
Development of ~ o n g - ~ e rComplications:
The Perspective of the Diabetes Control and
The Diabetes Control and Complications Trial Research Group
The Diabetes Control and Complications n i a l (DCCT)
demonstrated a reduction in the development and progression of the long-term complications of IDDM with
intensive therapy aimed a t achieving glycemic control
as close t o the nondiabetic range as possible. The
DCCT subsequently showed that the total lifetime
exposure t o glycemia was the principal determinant of
the risk of retinopathy and that there was a continuous
nonlinear relationship between this risk and the mean
level of HbA,, (DCCT Research Group, Diabetes
44:968-993,1995). In contrast, other authors, based on
a retrospective study (Krolewski e t al., N Engl J Med
332:1251-1255, 1995), have suggested that a glycemic
threshold for microalbuminuria and for retinopathy
exists a t an HbA,, level of -8%, below which there is no
further appreciable reduction in risk. In this perspective, we examine whether the DCCT data demonstrate
such a glycemic threshold for the development of
retinopathy, nephropathy, or neuropathy. In the DCCT,
1,441 patients with IDDM were randomly assigned t o
intensive (n = 711) or conventional (n = 730) therapy
and followed for a mean of 6.5 years. Retinopathy was
assessed every 6 months by stereoscopic fundus photography; albumin excretion was measured annually in
a 4-h collection; and neuropathy was assessed with a
standardized protocol performed at baseline and a t 5
years. Glycosylated hemoglobin was measured quarterly. Episodes of severe hypoglycemia were ascertained using standardized procedures. The risks (hazard rates) of retinopathy progression and of developing microalbuminuria and neuropathy were found to be
continuous but nonlinear over the entire range of glycosylated hemoglobin values in the intensive, conventional, and combined treatment groups. These nonlinear relationships describe a constant relative risk gra-
dient in which proportional reductions in HbA,, are
accompanied by proportional reductions in the risk of
complications. Although the magnitude of the absolute
risk reduction declines with continuing proportional
reductions in HbA,,, there are still meaningful further
reductions in risk as the HbA,, is reduced toward the
normal range. When the instantaneous risks for different complications associated with different HbA,, values are compounded over time, there are substantial
differences in the cumulative incidence of patients
experiencing a complication for patients with HbA,,
values of 6 vs. 7 vs. 8% or higher. In fact, no HbA,,
threshold could be identified, short of normal
glycemia, below which there was no risk of the development or progression of these complications. Furthermore, as the HbA,, was reduced proportionately,
the proportional rate of decline in the relative risk for
each of these complications was similar for HbA,, levels S8.0% and for levels >8%.In contrast, although the
absolute risk of severe hypoglycemia in the intensive
treatment group increased as the HbA,, decreased, the
relative risk gradients were significantly less for HbA,,
levels 18.0% than for levels 28%.
These extensive prospective DCCT data do not support the conjecture that a glycemic threshold for the
development of complications exists a t an HbA,, of 8%
or that an HbA,, goal of 8% is maximally beneficial. In
the DCCT, as HbA,, was reduced below 8% there were
continuing relative reductions in the risk of complications, whereas there was a slower rate of increase in
the risk of hypoglycemia. Therefore, the DCCT continues t o recommend implementation of intensive therapy with the goal of achieving normal glycemia as
early as possible in as many IDDM patients as is safely
possible. Diabetes 45:1289-1298, 1996
From t h e Diabetes Control and Complications Trial (DCCT) Research
Group. A con~pletelisting of the DCCT Research Group is published in
Archit~esof Ophtha~lmology113:49-51, 1995.
Address correspondence and reprint requests to The DCCT Research
Group, Box NDICIDCCT, Bethesda, MD 20892.
Received for publication 14 March 1996 and accepted in revised form
6 May 1996.
AER, albuinin excretion rate; DCCT, Diabetes Control and Complications Trial; ETDRS, Early Treatment of Diabetic Retinopathy Study; MIM,
multiplicative intensity model; PH, proportional hazards; SNPDR, severe
nonproliferative diabetic retinopathy.
he Diabetes Control and Complications Trial
(DCCT) was a multicenter randomized controlled clinical trial that demonstrated that,
compared with conventional diabetes therapy,
intensive therapy reduced by 35 to 75%the development
and progression of the long-term comp~icationsof IDDM
The goal Of intensive
was to achieve
glycemic levels in the nondiabetic range (HbA,, <6.05%);
DIABETES, VOL. 45, OCTOBER 1996
T
1289
the mean HbA,, actually achieved with intensive therapy
was 7.2% during the study, -2 units lower than that
achieved with conventional therapy. These reductions in
HbA,, were accompanied by a threefold increase in the
risk of severe hypoglycemia with intensive versus conventional treatment (2,3).
The primary analysis of the DCCT results, as mandated
by the study design, compared the long-term outcomes of
the two randomly assigned treatment groups. The DCCT
also reported a detailed epidemiological analysis of the
association between glycemic levels and risk of retinopathy progression observed in the DCCT (4). This analysis
demonstrated a continuously increasing risk of retinopathy progression with increasing mean glycosylated hemoglobin level. Earlier analyses also showed that the risk of
severe hypoglycemia within the intensive treatment
group increased apparently exponentially as the HbA,,
was reduced (2).
Based on the initially published DCCT results (2),
some have recommended that an HbA,, of 7-8% should
be a goal for treatment that would achieve a maximal
benefit-to-risk ratio, weighing the reductions in the risk
of complications with the increased risk of hypoglycemia
(5). Further, a recently published study (6), based on a
retrospective analysis of urine albumin measurements
and glycosylated hemoglobin levels measured 1 and 3
years before the renal assessment, reported that the risk
of microalbuminuria "was almost flat" or nearly constant
for levels of HbA, i 10.1%(estimated to be comparable to
an HbA,,. &I%). This was interpreted to suggest that a
biological threshold for risk existed at an HbA,, of -8%.
These authors concluded that little or no clinical benefit
would result from a further decrease below this level (6).
The article, an accompanying editorial (7), and a letter to
the editor that suggested a similar glycosylated hemoglobin threshold for progression of retinopathy (a), recommended that therapy of IDDM be directed at achieving
HbA,, levels i8.1%, as opposed to the HbA,, goals of
intensive therapy as implemented in the DCCT.
These recommendations prompted us to compare the
associations of HbA,, values 58% vs. those >8%with the
risk of development of nephropathy (rnicroalbuminuria)
and neuropathy, with progression of retinopathy, and with
the incidence of severe hypoglycemia. These analyses
were performed within the intensive and conventional
treatment groups of the DCCT and for both groups combined and are based on extensive DCCT data collected
prospectively during >9,000 patient-years of observation.
RESEARCH DESIGN AND METHODS
In the DCCT, 1,441 subjects were recruited at 29 clinical centers during
1983-1989 and were followed for an average of 6.5 years. Patients were
recruited either within a primary prevention cohort (n = 726) consisting of
patients with 1-5 years' duration of IDDM, no retinopathy detected by sevenfield stereoscopic fundus photography, and urinary albumin excretion <40
mu24 h, or within a secondary intervention cohort. The secondary intervention cohort comprised 715 patients with 1-15 years' duration of IDDM, very
mild to moderate nonproliferat ive retinopathy, and urinary albumin excretion
of C200 nlgL24 h (1). Of the secondary intervention cohort, 90% had albumin
excretion <40 mu24 h. Within each cohort, patients were randomly assigned
to receive either conventional or intensive treatment, totaling 730 and 711
subjects, respectively, in the combined cohorts. A complete description of
the baseline characteristics of the patients ha.? been reported previously (2).
Treatment and follow-up. Conventional and intensive treatments as practiced in the DCCT have been described in detail elsewhere (9). Conventional
therapy consisted of one or two daily injections of insulin with daily selfmonitoring but without specific glycemic goals. Intensive therapy involved
three or more daily injections of insulin or use of an external insulin pump
with frequent self-monitoring of blood glucose and had the goal of achieving
glycemic control a s close to the nondiabetic range a s safely possible.
Outcome measures. HbA,, was measured monthly among intensively
treated subjects and every 3 months for conventionally treated subjects,
with an estimated reliability coefficient of 99.2% and a normal (nondiabetic)
range of 4.05-6.05% (10). A total of 37,058 quarterly HbA,, values were collected. The mean of the quarterly HbA,, values is used as a summary measure of glycemic exposure for individual subjects during the DCCT.
Seven-field stereoscopic fundus photographs were obtained every 6
months (18,388 assessments) and were graded at a central facility (1 1). The
earlier DCCT report ('2) used the interim grading scale from the Early Treatment of Diabetic Retinopathy Study (ETDRS). Subsequent DCCT publications (11,12) used the final ETDRS scale (13), which is considered a more
sensitive and reliable indication of retinopathy severity. The final ETDRS
scale is used herein; therefore, the results differ slightly from those presented in the prior report (2). Sustained onset of retinopathy among patients
in the primary prevention cohort consisted of the development of three or
more microaneurysms or more severe lesions on two successive 6-month
fundus photographs. Sustained progression of retinopathy consisted of a
three or more step progression from baseline on two successive 6-month
assessments. Severe nonproliferative diabetic retinopathy (SNPDR; level 53
or higher on the final EDTRS scale) was defined by severe retinal hemorrhages, venous beading, or moderately severe intraretinal microvascular
abnormalities.
A 4-h standardized urine collection was performed annually (9,065
assessments) for the assessment of urinary albumin excretion rate (AER),
expressed a s milligrams per 24 h (14). Microalbuminuria was defined as an
AER 240 mgL24 h, advanced n~icroalbuminuriaas an AER 2100 mgI24 h, and
albuminuria a s an AER 2300 mu24 h. Sustained microalbuminuria was
defined as an AER 240 mu24 h on two successive annual evaluations. Seventy-six patients who entered the trial with microalbuminuria were
excluded from analyses of the developn~entof microalbuminuria and
advanced microalbuminuria.
A standardized clinical neurological examination was performed at baseline, 5 years, and study end and was accompanied by electrophysiological
assessment of nerve condurtion velocities (15). Autonomic neuropathy was
assessed at baseline and every 2 years (16). Confirmed clinical neuropathy
was defined as the presence of a history or physical examination consistent
with a peripheral somatosensory neuropathy, accompanied by either abnormal nerve conduction or autonomic neuropathy test results (15). Ninety-two
patients entered the trial with confirmed clinical neuropathy at baseline and
were excluded from the analyses of the development of neuropathy at 5
years. Among the remainder, 1,161 patients completed the 5-year neurological assessment.
Hypoglycemia. Severe hypoglycemia was defined as an episode characterized by symptoms consistent with hypoglycemia in which the patient
required assistance from another person and the blood glucose was <50
mg/dl andlor the clinical manifestations were reversed by oral carbohydrate.
subcutaneous glucagon, or intravenous glucose CJ). A total of 3,788 such
episodes were ascertained. Subjects were instructed to report immediately
all episodes of suspected severe hypoglycemia; all were interviewed regarding the episodes using a standard series of questions. In addition, at the quarterly follow-up visit, subjects were asked about the occurrence of any hypoglycemia. Episodes were further characterized by the symptoms experienced by the patient during each episode. Severe hypoglycemia with
coma/seizure consisted of those episodes in which the patient experienced
either coma andlor seizure as reported by the patient o r another person.
Statistical analysis. The relationship between glycemic exposure during
the DCCT and the various study outcomes was assessed using the updated
mean HbA,, at the time of each outcome assessment: annually for nephropathy, every 6 months for retinopathy, and at 5 years for neuropathy. Crude
rates within 1112th percentiles of the distribution of HbA,,, values, standardized to the number per 100 patient-years, were computed a s the ratio of the
number of events (x) to the patient-years of exposure (y) within each percentile. With more than 12 categories, some had no events. The standard
error of the log crude rate is 112 (17).
For retinopathy, nephropathy, and recurrent episodes of hypoglycemia.
the risk gradient was estimated from a Poisson regression model (17,18)
using the updated log (mean HbA,,.) as a time-dependent covariate. These
models describe the instantaneous absolute risk of an event, expressed as a
rate per 100 patient-years, a s a function of the updated mean HbA,,, termed
the risk gradient. Additional proportional hazards (PH) models (19) that
DIABETES, VOL. 45, OCTOBER 1996
inherently adjust for the background variation in risk over time were also
used. PH models of retinopathy progression were stratified-adjusted for the
baseline level of retinopathy; models of nephropathy onset and progression
were adjusted for the log(AER) at baseline and stratified by primary versus
secondary cohort. A generalization of the PH model for recurrent events, the
multiplicative intensity model (MIM) (19), was used for the analysis of hypoglycemia. Since these PH (MIM) models provided similar results to those of
the simpler F'oissor~ models, only the Poisson models are presented. The
logistic regression model (18) was used to describe the association between
the updated log(mean HbA,,) at .5 years and the risk (odds ratio) of confirmed clinical neuropathy at 5 years. Models in the combined groups were
adjusted for intensive versus conventional treatment by further stratification in the PI3 models or by adding an additional term in the Poissori and
logistic models.
All ana1yst.s used the natural log of the updated mean HbA,,, which provided a better fit than using the mean HbA,, itself (4). These regression
models express the log(risk) as a linear function of the log(HbA,,), i.e., a
straight line is used to describe the relationship between the log(risk) versus the log(HbA,,). To assess the validity of this log-log model, a nonparametric (model-free) spline function was fit to the raw data with these log
transformations. The results were strongly linear, indicating that this model
is appropriate.
From a linear log-log relationship, the risk gradient can be expressed as
the percentage reduction in risk for a fixed 10%reduction in the HbA,,, computed as (0.9P - 1) * 100, where p is the coefficient (slope) for log(HbA,,).
Such models describe a constant relative risk relationship over the range of
HbA,,. A continuous nonlinear absolute risk relationship for the risk versus
HbA,,., termed the simple exponential risk gradient, is obtained when a
straight line fit to the log(risk) versus the log(HbA,,) is presented in the original scales of' measurement (untransformed).
A change point is defined as a value of HbA,, below and above which
there is a statistically significant difference in the slope or risk gradient for
the association with the risk of complirations. The presence and significance of a change point was assessed through segmented log(risk) versus
log(HbA,,.) regression models that included a linear predictor of the form p,
(x) + p2 ( x - c) I (x > c), where ;r is the log(mean HbA,,.), I(.) is the (0,l) indicator function, and c is the =sumed change point. With c = log@), this model
provides an estimate of the risk gradient for HbA,,, values 28% (p,) and the
risk gradient for values >8%(8, + p,). To test whether a change point exists
at the specified value (c), the improvement in fit of the segmented regression
model over the simple exponential model was assessed by a likelihood ratio
test comparing this model with a model containing the p, effect alone, which
is also a test that p, = 0. A test for an optimal change point selected by a
search among a range of values from 6 t o 9.9%in increments of 0.1%was performed using a 2 df likelihood ratio test of the segmented model at the optimal value compared to the simple exponential model (20).
A threshold is defined as a change point below which the risk, if any, is
constant, or the risk gradient is zero. A test of a threshold at a specified
change point (c) was assessed by a Wald test that p, = 0 (20).
Poisson regression analyses within the intensive plus conventional
groups combined were adjusted for treatment group. To describe the risk
gradients in the figures, the treatment group effect was absorbed into the
assuming half the patients were in each group.
intercept (a)
These risk gradients describe the instantaneous hazard rate (A per year)
of events at any point in time as a function of the current updated log(mean
HbA,,) (x) of the form A(x) = exp(a + px). For an individual patient with a
constant log(mean HbA,,) of x during the study, the cumulative incidence of
an event is estimated from the integrated cumulative hazard over the 9 years
of follow-up in the DCCT as 1 - exp [-9A(.r)].
All results nominally significant at P < 0.05 are indicated.
RESULTS
Intensive versus conventional treatment effects.
The distribution of the mean HbA,, over the total period
of follow-up of each subject, ranging from 4 to 9 years, in
the conventional and the intensive treatment groups, is
shown in Fig. 1. (A similar figure, but incorporating the
mean HbA,, up to each semi-annual visit, is presented in
Ref. 4.) In the conventional treatment group, the median
value at 1,he final visit was 9.02 and only 20.3%of patients
(.n = 148:) had a mean HbA,, of 8 or less. In the intensive
treatment group, the median of the values at the final
DIABETES, VOL. 45, OCTOBER 1996
Glycosylated Hemoglobin (%)
FIG. 1. Spline-smoothed estimates of the distribution of the mean
HbA,, level for each subject during the DCCT, separately for 711
patients in the intensive group (-) and 715 patients in the conventional group (- - -).The density estimate is the integrand such that
the total area under each curve is 1.0.
visit was 7.07, and 83.4%of patients (n = 593) had a mean
HbA,, of 8 or less. In the combined conventional plus
intensive groups, approximately equal numbers of
patients had a mean HbA,, 1 8 (n = 741) and >8 (n = 700).
Thus, maximal power to distinguish a difference in risk
over the range of HbA,, values 1 8 versus that among values >8 is provided by an analysis among the combined
treatment groups.
Previously, the DCCT showed that intensive treatment,
compared to conventional treatment, reduced the risk of
retinopathy by 63% and of microalbuminuria by 39% in
the combined cohorts (2). The less striking effect of
intensive treatment on the risk of microalbuminuria,
compared with retinopathy, is attributed to the greater
degree of random within-patient variation for this outcome measurement. Analyses using less variable outcomes (14) revealed that intensive treatment reduced the
risk of developing sustained microalbuminuria by 60%
and of developing advanced microalbuminuria by 51%.
Therefore, the relationship between nephropathy and
HbA,, was explored using these other more stable manifestations of nephropathy as well as microalbuminuria.
Exponential risk gradients. For each complication,
Table 1 presents the percentage reduction in the instantaneous hazard rate (the risk gradient) associated with a
10% reduction in HbA,,. The risk gradient value is
obtained from the slope of the linear relationship of the
log of risk versus the log of the mean HbA,,, which is a
simple exponential model. Values are presented for
analyses among the intensive and conventional treatment group patients separately and also for an analysis
of the combined intensive plus conventional treatment
groups. The risk gradients for all complications were of
similar magnitude in the intensive and conventional
groups and were not significantly different between the
treatment groups for any complication. Thus, it was reasonable to examine the HbA,, risk gradient without
regard to treatment.
For each retinal, renal, and neurological con~plication
assessed, there was a significant reduction in risk (the
TABLE 1
Relative risk reductions associated with a 10%lower mean HbA,, in the intensive and conventional treatment groups and in the combined groups
Intensive therapy
-
--
Conventional therapy
% risk
reduction
95% CI
reduction
Combined groups
%&k
reduction
95% CI
-
-
% risk
95% CI
--
---
-
Retinopathy (final ETDRS)
Sustained onset, primary cohort only
Sustained progression (23 steps)
SNPDR (>level 53), secondary cohort only
Nephropathy
Microalbuminuria (240 mu24 h)
Sustained microalbuminuria (240 mu24 h)
Advanced nucroalbuminuria (2100 mu24 h)
Albuminuria (2300 mu24 h), secondary cohort
Neuropathy
Confirmed clinical neuropathy
Data for the combined groups are from a Poisson regression model aausted for intensive vs. conventional treatment. * P < 0.001.
t P < 0.01. $P < 0.05.
hazard rate) associated with a 10% reduction in HbA,,
over t h e entire range of HbA,,. The risk gradient for
microalbuminuria was l e s s steep t h a n t h a t of t h e o t h e r
renal manifestations owing to t h e greater degree of ran-
d o m within-patient variation. Results from PH models
were similar to t h o s e presented i n t h e table.
The linear regression of t h e log of t h e risk (hazard rate)
of sustained progression of retinopathy versus the log of
Log Glycosylated Hemoglobin (log %)
0.0
6.0
Glycosylated Hemoglobin (%)
0.0
5.5
6.0
8.5
7.0
Glycosylated Hemoglobin (%)
7.5
r
8.0
I
5
6
7
8
B
10
I1
12
Glycosylated Hemoglobin (%)
FIG. 2. The absolute risk of sustained retinopathy progression (hazard rate per 100 patient-years) in the combined treatment groups as a
function of the updated mean HbA,, during follow-up in the DCCT estimated from a Poisson regression model with 95% confidence bands:
a = -13.45,
(coefficient for log HbA,,) = 4.68, and offset = log(112 year). A: log(rate) vs. log(HbA,,). B: rate vs. HbA,, over the range
observed in the trial. C: r a t e vs. values of HbA,, 58%.D: cumulative incidence (probability) over 9 years of treatment vs. HbA,,.
DIABETES, VOL. 45, OCTOBER 1996
Log Glycosylated Hemoglobin (log %)
Glycosylated Hemoglobin (%)
Glycosylated Hemoglobin (%)
FIG. 3. The absolute risk of microalbuminuria (hazard rate per 100
patient-years) in the combined treatment groups as a function of
the updated mean HbA,, during follow-up in the DCCT estimated
from a Poisson regression model with 95% confidence bands: a =
-9.18; 6 (coefficient of log HbA,,) = 2.79, and offset = log(1 year).
A: log(rate) vs. log(HbA,,). B: rate vs. HbA,, over the range observed
in the trial. C: cumulative incidence (probability) over 9 years of
treatment vs. HbA,,.
the mean HbA,, is shown in Fig. 2A for the combined treatment groups. As the log of the HbA,, decreased, the log of
the risk also decreased in a strongly linear fashion, as indicated by the log of the crude rate within 1112th percentiles
of the HbA,, values. Note also that among the lower levels
of HbA,,, it was difficult to estimate the shape of the relaDIABETES, VOL. 45, OCTOBER 1996
tionship precisely because the rates were estimated with
less precision, as indicated by the larger standard error
bars, owing to the smaller nuntbers of events at lower
HbA,, values. Nevertheless, this figure shows that there is
a continuing decline in risk with lower HbA,,.
Figure 2B presents the same relationship expressed in
the original units of measurement showing the absolute
risk (rate per 100 patient-years) versus the updated mean
HbA,,. From the estimated linear log-log relationship, for
each 1 M reduction in HbA,,, such as from 9 to 8.1 or
from 7 to 6.3, there was a constant 39%reduction in risk.
As the HbA,, values were reduced proportionately, the
reductions in absolute risk became smaller because the
risk relationship was based on a constant relative or percentage reduction in risks. Thus, we observed a continuous simple exponential nonlinear risk relationship. Figure 2C presents the same risk gradient, but only over the
range of HbA,, values 58%. With a rescaling of the y-axis,
this clearly shows the continuing reduction in risk as the
HbA,, values were reduced below 8% and the absolute
risk became smaller.
Figures 2 8 and 2C describe the instantaneous hazard
rate of retinopathy progression as a function of the
updated HbA,,. For any given value of HhA,,, assumed
held constant over time, the cumulative incidence of
retinopathy progression can then be estimated by cornpounding these instantaneous risks over the 9 years of
follow-up in the DCCT, as shown in Fig. 2 0 over the cornplete range of HbA,, values. These figures show that
small differences in the instantaneous risks (Fig. 2B) correspond to larger differences in the total probability
(cumulative incidence) over a period of many years.
Similarly. Fig. ;3A presents the estimated linear relationship between the log of the risk (hazard rate) of
developing microalbuminuria and the log of the current
HbA,, within the combined intensive plus conventional
treatment groups. Across the range of HbA,, values, the
relationship is linear, but again it is difficult to estimate
precisely the nature of the relationship among lower values of the HbA,, because of imprecision in the estimates
of the rates.
Figure 3B presents the exponential relationship
between the risk of developing microalbuminuria and the
current mean HbA,,, without the log transformations.
For each 10% reduction in the HbA,,, there was a 25%
reduction in risk. Figure 3C presents the estimated cumulative incidence as a function of a constant mean HbA,,
value over the 9 years of follow-up in the DCCT.
Threshold and change point models for complications. Relative risk gradients associated with a 10%
reduction in HbA,, were obtained from segmented Poisson regression models with a change point at an HbA,, of
8% (Table 2). These models provide separate estimates
for the slope of the regression of the log(risk) on the
log(HbA,,) for HbA,, values 58 and values >8. From these
log-log regression model slope estimates, separate estimates of the relative risk gradient were calculated for a
10%reduction in the updated mean HbA,, over the range
of HbA,, values <8 and over the range of HbA,, values >8.
In general, all of the complications examined, including retinopathy progression, microalbun~inuria, sustained microalbuminuria, and confirmed clinical neu-
TABLE 2
Relative risk reductions associated with a 10% lower mean HbA,, among HbA,, values 58 vs. values 93% estimated from a segmented
(change point) model
--
--
HbA,,
58%
-- -
--
Co~nplication
-
% risk reduction
95% CI
- - -- - -
P
- - --
- -
% risk
HbA,,
18% -
- -
-
-
reduction 95%CI
-
-
-- -
P
P for HbA,,
<8 vs. >8
-
Sustained retinopathy progression
Intensive
Conventional
Intensive and conventional
Microalbuminuria
Intensive
Conventional
Intensive and conventional
Sustained microalbuminuria
Intensive
Conventional
Intensive and conventional
Confirmed clinical neuropathy
Intensive
Conventional
Intensivc and conventional
49
69
53
(27-65)
(29-87)
(37-66)
0.0003
0.006
<0.0001
37
34
35
(17-53)
(2641)
(28-42)
0.002
<0.0001
<0.0001
0.46
0.055
0.052
23
21
25
(2-40)
(-31-52)
(8-39)
0.036
0.37
0.008
33
24
26
(11-50)
(1334)
(16434)
0.006
<0.0001
<0.0001
0.55
0.87
0.97
43
58
49
(2-67)
(-50-88)
(19-68)
0.041
0.18
0.005
44
33
35
(17-62)
(1735)
(2346)
0.005
0.0002
<0.0001
0.97
0.47
0.40
30
32
32
(-19-58)
(-70-73)
(-5-56)
0.19
0.41
0.082
35
28
29
(-17-64)
(11-43)
(1342)
0.16
0.00'3
0.0008
0.87
0.93
0.90
ropathy, demonstrated similar associations with HbA,,
level within the HbA,, 58% range and the >8%range. The
risk reductions ranged from 21 to 49% for a 10% lower
mean HbA,, and were similar for the intensive and conventional treatment groups. In all instances, there was a
further reduction in risk with reductions in HbA,, <8%,
and in most cases, this additional reduction itself was
statistically significant. However, in no case was there a
significant difference between the risk gradients evaluated for HbA,, 58% and for HbA,, >8%, thus failing to
establish a significant change point at an HbA,, of 8% and
rejecting the hypothesis of a threshold at an HbA,, of 8%.
In no instance did the two-segment (change point)
regression model provide a significantly better fit to the
data than did the simple exponential regression model
presented in Table 1 and in Figs. 2 and 3.
In further analyses of retinopathy progression using
proportional hazards models, stratified-adjusted for
treatment group and baseline retinopathy severity, there
was again no difference in the risk gradients for HbA,,
18% and >8%. In analyses of the sustained onset of
retinopathy in the primary cohort and of the development of SNPDR among patients in the secondary cohort
(Table 2), there were no differences between the risk gradients for 18% vs. >8%in either treatment group or in the
groups combined. These analyses again indicate a significant continuing relative reduction in risk of sustained
retinopathy progression among intensive group patients
with HbA,, 58%. The relative risk reduction was not
significantly less than that observed among values >8%.
In general, PH models of retinopathic and nephropathic outcomes yielded the same results as those presented in Table 2 from the simpler Poisson models. In no
instance did the segmented model provide a significant
improvement over the simple exponential model. We
searched for evidence of a significant change point or a
threshold using PH models for microalbuminuria and
sustained retinopathy over the range of HbA,, from 6.0 to
9.9% in increments of O.l%, and found none.
1294
Hypoglycemia. The current HbA,, value, measured
monthly in the intensive treatment group and quarterly in
the conventional treatment group, was more strongly
associated with the risk (incidence) of all episodes of
hypoglycemia (including recurrent episodes) than was
the updated mean HbA,,. Among patients in the intensive
and conventional groups, analysis of the risk gradients for
the incidence of all severe hypoglycemia and for episodes
with coma/seizure, expressed as the percentage ifzcrease
in risk per 10% lower current HbA,,, revealed that the risk
gradients were significantly higher in the conventional
than in the intensive group. For each 10% reduction in the
HbA,,, there was a 26% increase in the risk of all severe
hypoglycemia (95% CI 22-29%) in the intensive treatment
group versus a 54%increase in risk (95% CI 49-60%) in the
conventional group (P < 0.0001 for intensive versus conventional). For hypoglycemia with coma/seizure, the risk
gradients were likewise markedly different between the
intensive and conventional treatment groups (18 vs. 48%
increase in risk, respectively, per 10% reduction in HbA,,;
P < 0.0001). Thus, we did not combine the treatment
groups for the examination of a change point.
For the intensive and conventional treatment groups,
the relative risk gradients associated with a 10% reduction in HbA,, obtained from segmented regression models with a change point at an HbA,, of 8% are presented
in Table 3. For episodes of severe hypoglycemia in the
intensive treatment group, the risk gradient among HbA,,
values 18% (18% greater risk per 10% lower HbA,,) and
that among values >8% (60% greater risk per 10% lower
HbA,,) were each significant at P < 0.0001. Furthermore,
the risk gradient for hypoglycemia for HbA,, values 58%
was significantly less than that for values >8% (18 vs.
60%, P < 0.0001). Similar results were observed in the
conventional treatment group, in which the risk gradients over these two segments were also significantly different (43 vs. 72%,P = 0.005).
For episodes with comaJseizure in the intensive group,
the risk gradient among HbA,, values 58% (7% greater
DIABETES. VOL. 45. OCTOBER 1996
TABLE 3
Relative risk increases in severe hypoglycen~iaassociated with a 10%lower current" HbA,, among HbA,, values 58 vs. values >8 estimated from a segmented (change point) Poisson model
HbA,, 18%
% risk increase
95% CI
- -
Severe hypoglycemic events
- - - - - - - - - - - - -- -
-
P
HbA,, >8%
% risk increase 95% CI
-
-
- -
- -- -
- -
-
P
P for HbA,,
18 vs >8
-
-
All events requiring assistance
Intensive
Conventional
Coma/seizure only
Intensive
Conventional
18
43
(14-23)
(34-53)
<0.0001
<0.0001
60
72
(44-79)
(58-87)
<0.0001
<0.0001
<0.0001
0.005
7
44
(-0.3-16)
(27-63)
0.062
<0.0001
61
52
(32-98)
(32-76)
<0.0001
<0.0001
0.003
0.64
*The current HbA,, is the current monthly HbA,, in the intensive treatment group, quarterly in the conventional treatment group.
risk per 1.0%lower HbA,,) was again significantly lower
(P = 0.003) than that anlong values >8% (61% greater risk
per 10% lower HbA,,). Moreover, among HbA,, values
<8%, this risk gradient in the intensive group was low and
was nominally not significantly different from zero ( P =
0.062), although this may in part be a function of the
smaller number of such episodes than all episodes of
severe hypoglycemia. In the conventional group, the risk
gradients over these two segments were not significantly
different (44 vs. 52%,P = 0.64).
These analyses indicate that in the intensive treatment
group, the risk of severe hypoglycemia increased more
slowly with reductions in HbA,, 18%, compared to the
rate at which the risk increased for reductions in HbA,,
in the range of >8%. In contrast to the results with longterm complications, a two-parameter segmented model
provided a significantly better description of the trend in
the risk of hypoglycemia as a function of the HbA,, than
the simple exponential curve previously presented (Fig.
5B in Ref. 2).
a reduction in risk of retinopathy progression of 6.57
cases per 100 patient-years (10.78 - 4.21). In contrast, a
reduction from 8 to 7.2% yields a reduction in risk of 0.95
cases per 100 patient-years (2.43 - 1.48). Although each
represents a 10% reduction in HbA,, and each results in
the same proportionate (39%) reduction in the risk of
retinopathy progression, the magnitude of the reduction
in absolute risks is much greater in the former than in the
latter case. All of these associations use the current
HbA,, value; however, if we use the mean change from
baseline to current HbA,, in the model, we obtain virtually identical risk reductions in retinopathy progression
for 10%lower HbA,,.
The recent suggestion that decreasing HbA,, below
8% is not accompanied by further reduction in risk (6,s)
(the threshold hypothesis) is refuted by these analyses
of the extensive DCCT data. Moreover, our analyses of
the risks of retinopathy, nephropathy, and neuropathy
within the intensive, conventional, and combined
groups in the DCCT revealed no evidence that an ZIbA,,
of 8% or any other value represented a significant
DISCUSSION
threshold or change point in the relative risk relationThe DCCT, with 9,065 patient-years of prospective fol- ship (Table 2). In most instances, the relative risk gradilow-up and 4,835 patient-years with a mean HbA,, 58% ent over the segment of HbA,, values 58% was signifiand using precise and frequent measures of glycemia and cantly different from zero, rejecting the hypothesis that
complications over time, provides an excellent opportu- a threshold exists at an HbA,, of 8%. In all instances, the
nity to assess the relationships between glycemic expo- overall relative risk gradient was significant and no difsure and the risk of complications. We have demon- ference was found between the risk gradients over the
strated previously that the dominant predictor of the risk two segments of HbA,, values 58% versus values >8%,
of retinopathy is the total lifetime exposure to hyper- demonstrating no discernible change point at an HbA,,
glycemia as reflected by the level of HbA,, and the dura- of 8%. Thus, for each complication, the DCCT data indicate a continuous relative risk relationship over the
tion of diabetes (4).
Herein we show that there is a linear relationship range of HbA,, values, wit,h n o apparent glycemic
between the log of the absolute risk (instantaneous haz- threshold or change point.
ard rate) of retinopathy and nephropathy and the log of
Furthermore, the fact that the intrinsic risk relationthe current HbA,, over the entire range of HbA,, values ship between levels of glycemia and complications is
(Figs. 2A and 3A). This equates to a nonlinear relation- nonlinear implies that two straight lines could be fit to
ship between the untransformed risk of complications the data in Figs. 2B or 3B (without the log transformaand the untransformed level of HbA,,, a relationship that tion), with different slopes over the range of HbA,, values
is based on a constant relative risk in both the intensive 18% and for values >8%,as was done by Krolewski et al.
and the conventional treatment groups (Figs. 2B and 3B). (6). However, such an analysis obscures the fact that
For a proportionate reduction in HbA,,, there is a con- there is a constant relative risk relationship over the
entire range of HbA,, values.
stant proportionate reduction in risk.
This nonlinear relationship means that the absolute
Several other studies, including a large (n = 3,250)
reduction in risk is greatest when high values of HbA,, cross-sectional study of IDDM and its con~plications
are reduced. A reduction in HbA,, from 11 to 9.9% yields (21), the prospective Wisconsin Epidemiologic Study of
DIABETES, VOL. 45, OCTOBER 1996
129.5
Diabetic Retinopathy ,(22,23), the prospective Stockholm Diabetes Intervention Study (24), and others in
Colorado (25) and Berlin (26), have reported relationships between HbA,, and various complications that
agree with our findings and do not support the existence of glycemic thresholds for the early stages of
retinopathy, nephropathy, and neuropathy studied in the
DCCT. However, some of these studies have suggested
the possibility of a glycemic threshold for the development of proliferative retinopathy (21,24). There were
too few such events during the DCCT for our data to
address this possibility directly. Analyses of severe nonproliferative diabetic retinopathy, the stage a few steps
short of proliferative retinopathy on the ETDRS scale
(13), failed to demonstrate any glycemic threshold in
the DCCT (Table 2).
Features of the Krolewski et al. (6) study likely contributed to their failure to demonstrate evidence of risk
reductions in microalbuminuria at HbA,, values 4%.
One
important statistical factor is the influence of random
variation or noise. Only one to three random urine albumin-to-creatinine ratios were measured. It is well known
that levels of albumin in the microalbuminuric range are
highly variable over time within patients, whereas albuminuria, once developed, is more stable. Further, the
patients who had reached clinical albuminuria (proteinuria) were eliminated from the Krolewski et al. (6) analysis. Thus, they attempted to correlate a noisy outcome
(microalbuminuria measured in spot urine speci~nens)
with HbA,,, weakening the sensitivity of their analysis
and excluding the patients of greatest clinical interest
(those with albuminuria).
The DCCT and some of the studies cited above
(22-24,26) were able to describe a risk relationship with
microalburninuria (and other complications) because
periodic longitudinal measurements in large numbers of
patients can override the influence of random withinpatient variation over time. Also, in the DCCT, the slope
of the risk relationship between glycemia (HbA,,) and
microalbuminuria (Fig. 3A) is not as large as that for sustained retinopathy (Fig. 2A). This suggests that retinopathy may be more sensitive to the effects of glycemia than
is microalbuminuria, as measured in the DCCT.
The matter of greatest clinical interest is the magnitude of further reductions in risk of complications with
HbA,, values 58%. For example, from the regression
model in Table 1 and Fig. 2B, the estimated risk of
retinopathy progression is 2.43 cases per 100 patientyears at an HbA,, of 8%,which is reduced to 1.30 at an
HbA,, of 7% and t o 0.63 at an HbA,, of 6%. However,
each of these rates per 100 patient-years is in fact an
instantaneous risk (hazard rate), which can be interpreted (approximately) as the percentage of patients
among those still a t risk who will first experience the
event in a year's time. When compounded over a 9-year
period of follow-up as in the DCCT (Fig. 2D), these rates
correspond to 20% cumulative incidence of patients
developing retinopathy progression when held a t an
HbA,, of 8%,versus 11%at an HbA,, of 7%, and only 5.5%
at an HbA,, of 6%.
Likewise, from Fig. 3B, the estimated risk of microalbuminuria is 3.41 cases per 100 patient-years at an HbA,,
of 8%;2.35 at an HbA,, of 7%, and 1.53 at an HbA,, of 6%.
When compounded over a 9-year period of follow-up
(Fig. 3C), these hazard rates correspond to 26% cumulative incidence of patients developing rnicroalbuminuria
when held at an HbA,, of 8%over 9 years, versus 19%at
an HbA,, of 7%, and 13%at an HbA,, of 6%.
When projected over longer periods of treatment,
these differences in the cumulative incidence of retinopathy progression or microalbuminuria diverge exponentially for different values of HbA,,. Therefore, when the
absolute instantaneous risks associated with HbA,, values <8%,albeit progressively smaller, are applied to hundreds of thousands of patients and are compounded year
by year over a lifetime of diabetes, there are substantial
differences in the numbers of patients affected at the different HbA,, levels over the range of 6 to 8%.
It is also important to note that 83% of intensively
treated subjects in the DCCT achieved a mean HbA,, 58%
during the trial, the median value among all subjects
being 7.07%. When compared with conventional treatment with a median of 9.02%,this reduction in HbA,, with
intensive therapy led to dramatic reductions in the risks
of complications, including a 70%reduction in the risk of
sustained retinopathy progression and a 39%reduction in
the risk of rnicroalbuminuria. However, from Figs. 2B and
3B, it is estimated that an intensive treatment with an
average HbA,, of 8%, compared with conventional treatment, would achieve only a 43% reduction in the risk of
sustained retinopathy progression and a 28%reduction in
the risk of microalbuminuria,
It could be argued that the absolute further gains in
preventing complications by lowering HbA,, below 8%
would not be worth the effort, risk, and cost required.
For example, treatment of retinopathy with photocoagulation at the earliest appropriate time does offer a
practical and relatively low-risk treatment (27) for the
large residue of retinopathy that would accrue if all
patients with IDDM were treated by only seeking an
HbA,, goal of 8%, as recently suggested (5-8). However,
many patients do not reach the ophthalmologist in time
(28) or have the severity of their retinopathy recognized
(29). In addition, photocoagulation both is costly and
can itself cause loss of vision (27). Similar conclusions
apply to neuropathy, which all too often leads to foot
ulcers and amputation (30) because preventive measures are frequently ignored in practice (31). Only intensive therapy offers the chance of maximally decreasing
the risk of development and progression of all of the
diabetes-specific complications.
If an HbA,, of 8%were adopted as a standard, the greatest concern would be the thousands of patients who
ndght develop renal failure and be encumbered with the
morbidity, mortality, and expense associated with endstage renal disease and its treatment (32). There is ample
evidence that elevated albumin excretion presages ultimate renal insufficiency (33,34). The DCCT has demonstrated maximal reduction in the risk of microalbuminuria and albuminuria when therapy is aimed at lowering
the HbA,, as much as possible. Decreasing the development and progression of the earlier stages of nephropathy is very likely to decrease andlor delay the development of end-stage renal disease. Given the catastrophic
DIABETES, VOL. 45, OCTOBER 1996
nature of end-stage renal disease to the patient and the REFERENCES
1. DCCT Research Group: The Diabetes Control and Conlplications Trial
huge cost to society, it is imperative that physicians rec(DCCT): design and methodologic considerations for the feasibility
ommend the most effective current therapy aimed at the
phase. Diabetes 35530-545, 1986
lowest HbA,, that can be safely achieved. From a practi2. DCCT Research Group: The effect of intensive treatment of diabetes on
cal point of view, aiming for an HbA,, goal of 8% is likely
the development and progression of long-tern1 complications in insulinto result in values that are considerably higher. In the
dependent diabetes mellitus. N Engl J Med 329:977-986, 1993
3. DCCT Research Group: Adverse events and their association with treatDCCT, the actual mean HbA,, achieved was -1% higher
ment regimens in the Diabetes Control and Complications Trial. Diabdes
than the goal of <6.05%.
Cave 18:1415-1427, 1995
A major incentive for other authors to select HbA,,
4. DCCT Research Group: The relationship of glycemic exposure (HbA,,)
goals higher than the target used in the DCCT was the
to the risk of development and progression of retinopathy in the Diabelief that higher HbA,, goals would substantially lower
betes Control and Complications Trial. Diabetes 44:968-983, 1995
5. Sipperstein MD, Browner WS: The case for moderate rather than tight
the risk for severe hypoglycemia (5-8). However, our
glucose control in insulin-dependent diabetes mellitus. Cliu Diabetes
data reveal, in contrast to the relationship between
13:7-9, 1995
HbA,, and the risk of developing complications, that the
6. Krolewski AS, Laffel LMB, Krolewski M, Quinn M, Warram JH: Glycosyrisk of severe hypoglycemia does not increase proporlated hemoglobin and the risk of microalbuminuria in patients with
insulin-dependent diabetes n~ellitus.N Engl J Med 332:1251-1255, 1995
tionally as HbA,, is reduced over the range of values
7. Viberti G: A glycemic threshold for diabetic complications? N Engl J Med
18% as it does over the range of values >8%. Although
332: 1293-1294, 1995
the risk of severe hypoglycemia continues to increase at
8. Warram JH, Manson JE, Krowleski AS: Glycosylated hemoglobin and the
lower HbA,, values with intensive therapy, the risk grarisk of retinopathy in insulin-dependent diabetes mellitus. N Engl J Med
dient flattens substantially. Therefore, as the HbA,, lev332:1305-1306, 1995
9. DCCT Research Group: Implementation of treatment protocols in the Diaels are reduced below 8% with intensive therapy, there
betes Control and Complications Trial.Diabetes Care 18361-375, 1995
is a continuing proportionate reduction in the risk of
10. DCCT Research Group: Assessment of hemoglobin Alc in the Diabetes
complications, whereas the risk of severe hypoglycemia
Control and Complications Trial (DCCT). Clin Chem 33~2267-2271,
increases at a slower rate. The mechanism for this flat1987
tening of the risk curve for hypoglycemia is unknown. 11. DCCT Research Group: Progression of retinopathy with intensive versus
conventional treatment in the Diabetes Control and Complications Trial.
However, it may be that the risk for all hypoglycemia
Ophthalmology 102:647-661, 1995
(which includes recurrent episodes) tends to flatten
12. DCCT Research Group: The effect of intensive diabetes treatment on
because patients who experience severe reactions are
progression of diabetic retinopathy in insulin-dependent diabetes mellimore likely to practice preventive behavior that ameliotus: the Diabetes Control and Complications Trial. A w h Ophthalmol
113:3651, 1995
rates their risk for hypoglycemia without substantially
13. Early Treatment Diabetic Retinopathy Study Research Group: Fundus
affecting their HbA,,.
photographic risk factors for progression of diabetic retinopathy:
Given the demonstrated association between glycemia
ETDRS Report Number 12. Ophthalmology 98:823-833, 1991
and HbA,,, in both the conventional and intensive treat- 14. DCCT Research Group: Effect of intensive therapy on the development
ment groups, it is tempting to advocate any therapy that
and progression of diabetic nephropathy in the Diabetes Control and
Complications Trial (DCCT). Kidney Int 47:1703-1720, 1995
achieves the target HbA,, of intensive therapy. However,
15.
DCCT
Research Group: The effect of intensive diabetes therapy on the
since intensive therapy as implemented in the DCCT and
development and progression of neuropathy in the diabetes control and
other controlled clinical trials (24) is the only therapy
complications trial. Ann Intern Med 122561-568, 1995
that has been tested directly and demonstrated both to 16. Schumer M, Pfeifer MA, Crum D, Burton G. Burton C, the DCCT
Research Group: Diabetic autonomic neuropathy-part I: autonomic nerlower glycemic levels and to prevent or delay long-term
vous system data analysis by a computerized central unit in a multicencomplications, the DCCT Research Group continues to
ter trial. A m J Med 85 (Suppl. 5A):137-143. 1988
recommend it as the first choice for most patients with
17. Breslow NE, Day NE: Statistical Methods i n Cancer Research. II. The
IDDM. Until other therapies are proven equally effective
Design and Analysis of Cohort Studies. Lyon, France, International
and at least as safe as intensive therapy, intensive therapy
Agency for Research on Cancer (London, Oxford Univ. Press), 1987, p.
120-142
should remain the treatment of choice for IDDM.
In conclusion, the recent assertion that a glycemic 18. Agresti A: Categorical Data Analysis. New York, Wiley, 1990, p. 8 4 9 1 ,
190-195
threshold exists at an HbA,, of 8% (6,8) is not supported 19. Fleming
TR, Harrington DP: Counting Processes and Survival Analysis.
by the best available data. Further, the therapeutic recNew York, Wiley, 1991, p. 125-153
ommendation that the target level of HbA,, be set at 8% 20. Ulm K: A statistical method for assessing a threshold in epidemiological
studies. Stat Med 10:341-349, 1991
could well do harm. Therefore, the DCCT continues to
recommend implementation of intensive therapy with 21. EURODIAB IDDM Complications Study Group: Microvascular and acute
complications in IDDM patients: the EURODIAB IDDM Conlplications
the goal of achieving normal glycemia as early as possiStudy. Diabetologia 37:278-285, 1994
ble in as rnany IDDM patients as is safely possible.
22. Klein R, Klein BEK, Moss SE, Cruickshanks KJ: Relationship of hyperACKNOW1,EDGMENTS
The DCCT is sponsored by the Division of Diabetes,
Endocrinology and Metabolic Diseases of the National
Institutes of Diabetes and Digestive and Kidney Diseases,
the National Institutes of Health, through cooperative
agreemen.ts and a research contract. Additional support
was provided by the National Heart, Lung and Blood
Institute, the National Eye Institute, and the National
Institute for Research Resources.
DIABETES, VOL. 45, OCTOBER 1996
glycemia to the long-term incidence and progression of diabetic
Med 1549169-2178, 1994
retinopathy. Arch I n t m ~
23. Davis MD: Diabetic retinopathy. Diabetes Care 1518441874, 1992
24. Riechard P: Are there any glycemic thresholds for the serious microvascular diabetic complications? J Diabetes Complications 9:25-30, 1995
25. Chase P, Jackson WE, Hoops SL, Cockerham RS, Archer PG, O'Brien D:
Glucose control and the renal and retinal complications of insulin-dependent diabetes. JAMA 251: 1155-1 160, 1989
26. Danne T, Weber B, Hartmann R, Enders 1, Burger W, Hovener G: Longterm glycemic control has a nonlinear association to the frequency of
background retinopathy in adolescence with diabetes: follow-up of the
Berlin Retinopathy Study. Diabetes Care 17:1390-1396, 1994
27. Ferris FL: How effective are treatments for diabetic retinopathy? J M 4
269:1290-1291, 1993
28. Witkin SR, Klein R: Ophthalmologic care for persons with diabetes.
JAM4 251:2534-25.77, 1984
29. Sussman EJ, Tsiaras WG, Soper KA: Diagnosis of diabetic eye disease.
JAM4 247:3231-3234, 1982
30. Bild DE, Selby JV,Sinnock P, Browner WS, Braveman P, Showstack JA:
Lower-extremity amputation in people with diabetes. Epiden~iologyprevention. Diabetes Care 1224-31, 1989
31. G~unfeldC: Diabetic foot ulcers: etiology, treatment, and prevention.
Adu I ) t t e m Med 37:103-132, 1991
32. Perneger TV, Brancati FL, Whelton PK, Klag MJ: End-stage renal disease
attributable to diabetes mellitus. Atln Intern Med 121:912-918, 1994
33. Mogensen CE, Christensen CK: Predicting diabetic nephropathy in
insulin-dependent patients. N Etlgl J Med 311:89-93, 1984
34. Viberti GC, Jarrett RJ, Mahmud U, Hill RD, Argyropoulos A, Keen H:
Microalbuminuria a s a predictor of clinical nephropathy in insulindependent diabetes mellitus. Lojlcet i:1430-1432, 1982
DIABETES, VOL. 45, OCTOBER 1996
Perspectives in Diabetes
Diabetogenic T-cell Clones
Kathryn Haskins and Dale Wegmann
boxypeptidase H, ICA69 (islet-cell antigen 69), and several gangliosides (1). Perhaps one of the biggest dividends of this work is the basis that has been laid for diagnosing at-risk populations through analysis of their
autoantibody titers; as more reliable assays are developed for defined p-cell autoantigens such as insulin,
GAD, and ICA69, mass screening of potential diabetic
patients will become more feasible (2). However, no
direct role for autoantibodies in the pathogenesis of diabetes has been established.
During the mid-80s, increasing interest focused on cellmediated autoimmunity and the nature of the infiltrate in
the islets, especially on the role of T-lymphocytes (or
T-cells). Histological analysis of islet lesions revealed
that T-cells were a predominant component of the infiltrate and that both CD4' and CD8' T-cells were present
(3). A good deal of work, summarized in several recent
reviews (4-6), has demonstrated the importance of
T-cells in the pathogenesis of diabetes: disease was prevented by treatments that included neonatal thymectomy, antibodies to T-cell molecules, and immunosuppressant drugs that affect T-cells, and most convincingly,
disease could be adoptively transferred with splenocytes
from diabetic mice.
Early adoptive-transfer experiments, in which fractionhe two most distinctive features of type I diabetes, or IDDM, as an autoimmune disease are ated spleen cells were transferred into neonatal or subthe presence of autoantibodies in the sera of lethally irradiated adult recipients, indicated that both
patients and the mononuclear cellular infiltra- CD4' and CD8' T-cells were required to induce diabetes
tion of islets known as insulitis. From about the mid-70s (7,8), but the relative importance of each subset has been
to the mid-80s, much attention was focused on the more difficult to assess. In general, CD4' T-cells seem to
humoral aspects of autoimmunity as researchers tried to be primarily involved in the initiation of the disease
define 13-cell autoantigens through the isolation and char- process and in recruiting other cells (including CD8'
acterization of autoantibodies arising spontaneously in T-cells) to the pancreas. The role of CD8' T-cells in p-cell
sera of patients and animals with IDDM or as a result of destruction has been more controversial. The recent
immunization with various islet and pancreatic tissue development of the NOD-scid mouse has allowed a more
preparations. This approach succeeded to a large extent detailed analysis of the role of CD4' and CD8' T-cells in
in that a number of important autoantibody reactivities which it has been observed that the degree of CD8 depenwere identified, including those to insulin, GAD, car- dency of disease transfer was donor-age dependent (9).
Thus, when spleen cells from overtly diabetic donors
were depleted of CD8' T-cells, there was decreased diabetes
incidence and delayed time to onset; however,
From the Barbara Davis Center for Childhood Diabetes, Departments of
when 6- to 8-week-old donors were used, there was an
Immunology and Pediatrics, LTniversityof Colorado Health Sciences Center, Denver, Colorado.
absolute dependence on both CD4' and CD8' T-cells.
Address correspondence and reprint requests to K. Haskins, DepartThese results suggest that CD8 T-cells are required for
ment of [mmunology, Box B184, University of Colorado Wealth Sciences
development of a pathogenic T-cell response to islet
Center, 4200 E. 9th Ave., Denver, CO 80262.
cells, but once a CD4 response develops, CD8' T-cells are
Received for publication 28 March 1996 and accepted in revised form
27 June lL996.
not absolutely necessary.
APC, antigen-presenting cell; BDC, Barbara Davis Center for ChildWith the advent of cloned T-cell lines, we now have at
hood Diabetes; HSP, heat shock protein; IL, interleukin; IFN-y, y-interhand the tools to begin dissecting the individual contriferon; M:HC, major histocompatibility complex; Th, T-helper; TNF, tumor
necrosis factor.
butions of different T-cell subsets and specificities. For
The role of T-cells in the pathogenesis of IDDM has
been an area of much interest, and investigators have
recently acquired new tools for studies on T-cells with
the advent of T-cell clones that are reactive with islet
antigens. Derived from NOD mice, diabetogenic T-cell
lines and clones have for the most part been CD4' and
T-helper 1 (Th1)-like in their cytokine production.
Some CD8' cytotoxic clones have also been reported,
although these have generally not transferred diabetes in the absence of CD4' T-cells. The T-cell clones
that have been described can also be separated on the
basis of their antigen reactivity. While many of the
T-cell lines and clones described react with islets, isolated islet cells, or islet membrane preparations, others have known antigen specificities, reacting with
defined islet cell proteins such as insulin, GAD, and
heat shock proteins. Particularly in the case of
insulin-reactive clones, diabetogenicity has also been
demonstrated. In light of the many possible T-cell
reactivities that may arise from the islet lesion, the
question of whether there is a dqminant initiating
antigen is a particularly intriguing one. Diabetes
4531299-1305,1996
DIABETES, VOL. 45, OCTOBER 1996
instance, information on the relative contributions of
CD4' and CD8' T-cells to P-cell destruction, as well as
interactions between these two populations, could be
provided by analysis of representative examples of each
of these types of cells in the form of islet-specific T-cell
clones. In this review, we describe T-cell clones that have
recently become available, with particular emphasis on
those that are diabetogenic. Although there have been
numerous reports of T-cell lines and clones reacting with
islets and islet-derived antigens, very few of these T-cells
have been well characterized with respect to pathological activity. This becomes an important issue when one
considers the relevance of human T-cell clones and
whether they are pathogenic.
ISLET-REACTIVE T-CELL CLONES
Barbara Davis Center for Childhood Diabetes
(BDC) panel o f islet-specific diabetogenic T-cell
clones. Although there have been a few reports of isletreactive T-cell lines isolated from human peripheral blood
lymphocytes (10,ll) or from the BB rat (12), nearly all of
the cloned T-cell lines described within the last 10 years
have come from the NOD mouse. The first report of a
pathogenic T-cell clone came from the Barbara Davis Center for Childhood Diabetes (BDC) and was described by
Haskins et al. (13). A CD4' T-cell clone, BDC-2.5, was
derived from a newly diabetic NOD mouse and cultured in
the presence of mouse islet cells and irradiated splenocytes as antigen-presenting cells (APCs). When assayed
with islet cells and NOD APCs, this clone proliferated and
made interleukin (1L)-2; it did not respond to either islet
cells or APCs alone, which suggests that it recognized an
islet cell antigen in the context of the NOD class I1 molecule, I-Ag7.In vivo, the BDC-2.5 clone was indicated to be
pathogenic by virtue of the fact that it mediated islet but
not pituitary graft destruction. A second clone, BDC-2.4,
did not display specificity for islet cells in vitro and did
not affect islet grafts. Haskins et al. (14) subsequently produced a panel of cloned T-cell lines, all CD4' and derived
from spleen and lymph node cells of newly diabetic
female NOD mice, that responded in vitro to islet cells in
the presence of NOD APCs by proliferating and making
IL-2. Further analysis indicated that these clones were all
of the T-helper 1 (Thl) phenotype in that they produce the
cytokines IL-2, y-interferon (IFN-y), and tumor necrosis
factor (TNF), but not IL-4 (15), and that their T-cell receptor variable and joining regions were heterogeneous (16).
In vivo, it was found that every islet-specific clone that
was tested could mediate islet graft destruction in (CBA
X N0D)Fl recipients, animals that are not prone to developing spontaneous diabetes but are rendered diabetic
with streptozotocin and then transplanted with NOD
islets (14,17). The diabetogenic nature of these clones
was confirmed by studies in which two or three injections
of two of the T-cell clones, BDC-2.5 and BDC-6.9, were
administered intraperitoneally to young NOD recipients;
mice receiving these two clones (but not controls receiving the non-islet-specific clone BDC-2.4) rapidly developed diabetes andlor extensive insulitis (18).
Over the last few years, extensive studies have been
carried out to characterize the in vivo activity of this
panel of islet-specific T-cell clones. In the first studies, in
which the T-cell clones were adoptively transferred into
very young (2- to %week-old) NOD recipients (la), we
observed that efficiency of disease induction appeared to
correlate with the age of the mice. Extensive insulitis
developed in all of the animals that received islet-specific
T-cell clones, but as mice approached 3 weeks of age, the
incidence of overt diabetes, indicated by hyperglycemia,
decreased. Further investigation of the kinetics of disease induction revealed that in unirradiated NOD mice,
there is a very narrow age window in which disease can
be efficiently transferred with T-cell clones (19). In the
three age-groups studied, it was found that four different
diabetogenic T-cell clones caused disease with high efficiency (75%incidence of overt diabetes with three of four
clones, 67% overall) in the youngest (8- to 14-day-old)
group, with lower efficiency in the 15- to 18-day-old
group, and not at all in the third age-group (19- to 28-dayold). These results, together with similar findings by Bendelac et al. (7), who investigated disease induction with
diabetic splenocytes in three age-groups of NOD recipients, suggested that there is a developn~entalevent in the
NOD mouse after 3 weeks of age that confers resistance
to T-cell transfer of disease.
In another set of studies, the question was whether the
T-cell clones could adoptively transfer diabetes to animals not prone t o developing spontaneous disease, e.g.,
F1 crosses between NOD mice and other strains. Four
diabetogenic clones were used in transfer experiments
in seven different NOD F1 strain combinations, including three in whch there was positive expression of the
class I1 I-E. Our results showed that although there was
variability among different clones in different Fl recipients, all four clones were able to transfer diabetes to
three or more F1 combinations (20). However, there was
one strain combination, the (C57lL X NOD)Fl, that was
highly resistant to disease transfer, an interesting observation in light of work by McDuffie and coworkers
describing three diabetes-resistant NOD congenic lines
that were produced from a C57L X NOD cross and the
evidence that three separate and independent genes are
involved in conferring resistance in these lines (M.
McDuffie, Abstract, IXth International Congress of
Immunology, July 1995). Furthermore, disease transfer
studies in the NOD congenic mice with the diabetogenic
T-cell clone BDC-6.9, indicated that clone-induced diabetes was delayed in two of the mouse lines compared
with NOD controls (100% of controls became diabetic
with BDC-6.9 110 days after the first injection of cells).
In the third line, however, there was a significant degree
of protection: in 100%of the mice that received BDC-6.9,
no overt diabetes was observed in the 16day period
before the animals were killed (M. McDuffie and K.H.,
unpublished observations).
We have also conducted studies using islet-specific
T-cell clones to investigate the relative roles of CD4' and
CD8' T-cells in the induction of diabetes. In a continued
collaborative effort with Lafferty and coworkers (21,22)
to investigate the ability of the clones to mediate islet
graft destruction, it was found that the T-cell clone BDC6.9 could migrate to the graft site when injected intraperitoneally into (CBA X N0D)Fl mice that had received
NOD islet transplants and that islet destruction took
DIABETES, VOL. 16, OCTOBER 1996
place whether or not CD8' T-cells were present. This
work, along with an earlier description of diabetes transfer with the T-cell clones in young NOD mice (18), suggested that at least in the case of an activated T-cell
clone, CD4' T-cells were sufficient for disease induction.
More recently, a better model has become available for
exploring the individual contributions of different T-cell
subsets:: the NODILt-scid/scid (NOD-scid) mouse
described by Christianson et al. (9). The NOD-scid
mouse, like other scid mice, does not have T- or B-cells
and is not prone to diabetes. However, adoptive transfer
of splerlocytes from diabetic NOD mice to NOD-scid
mice will reproducibly result in full-blown diabetes, characterized by hyperglycemia and severe insulitis, within a
3- to 4-week period (9,15). Using this immunodeficient
mouse ;is a recipient, we demonstrated that transfer of
the islet-specific CD4' T-cell clone BDC-6.9 caused
destruction of pancreatic p-cells and development of diabetes without help from host B-cells, CD4' T-cells, or
CD8' T-cells (15). However, a second islet-specific T-cell
clone, I3DC-2.5, could not transfer diabetes or insulitis
into NOD-scid mice. Although it readily induces diabetes
in young unmanipulated NOD mice, BDC-2.5 could only
induce disease in NOD-scid mice with a cotransfer of
CD8-enriched T-cells from diabetic spleen. In young NOD
mice receiving the VP4' T-cell clones BDC-2.5 or BDC6.9, imrnunohistochemical staining of pancreatic lesions
showed the presence of CD4+,CD8+,Vp4', and MAC-1'
cells within the infiltrate, as in the infiltrates in lesions of
spontaneously diabetic female NOD mice. In NOD-scid
mice that received BDC-6.9, there were CD4+Vp4' T-cells
and a large population of MAC-1' cells in islet lesions. In
NOD-scid recipients of cotransferred BDC-2.5/CD8'
splenic T-cells, there was a small population of CD4'
T-cells and a larger population of CD8' T-cells in the
islets, whereas no infiltrate was detectable in recipients
of CD8 splenocytes or BDC-2.5 alone. These results indicate that there may be a role for CD8' T-cell help in diabetes induced by some islet-specific CD4' T-cell clones.
Other islet-reactive T-cell clones. That there is considerable diversity in the types and activities of T-cell
lines and clones produced from NOD mice is indicated
by reports from other investigators. Unlike the BDC
clones, which were derived from spleen and lymph node
T-cells and are all CD4+,almost all other T-cell lines that
have been described were isolated from the islet lesions
of NOD mice and included CD8' as well as CD4' T-cells.
For example, Reich et al. (23) described two cloned Tcell lines derived from NOD islets, one CD4' and the
other CD8+,that responded to NOD islet cells and that
could in combination (but not singly) induce extensive
insulitis in 7- to 8-week-old sublethally irradiated NOD
or (NOD X BALB/c)Fl recipients (23). Another report
from N'agata et al. (24) described the isolation of cytotoxic C:D8 T-cell lines by culturing islets from 20-weekold female NOD mice in IL-2; these cells were capable of
specific killing of islet cells, but no analysis was made of
their activity in vivo.
Nakano et al. (25) isolated from islet infiltrates of NOD
mice CD4' T-cell clones that responded to islet cells from
various strains of mice in the context of NOD APCs,
showed diverse usage of T-cell receptor V and J segDIABETES, VOL. 45, OCTOBER 1996
ments, and transferred insulitis into I-E' transgenic NOD
mice (25). In a report by Pankewycz et al. (26), T-cell
lines were propagated from pancreatic islets of prediabetic (2-month-old) NOD mice and yielded a CD4' clone
that responded to islet cells and induced disease in
female NOD mice. Shimizu et al. (27) isolated CD4' and
CD8' T-cell clones from islets obtained from irradiated
nondiabetic male NOD mice that had received a transfer
of splenocytes from diabetic females. These clones proliferated and made IFN-y in response to islet cells in the
presence of NOD APCs, and the CD8 clones also killed
islet cells. Two of the CD4' T-cell clones were diabetogenic in irradiated NOD male recipients, particularly in
the presence of splenic CD8' T-cells from diabetic mice.
In a report on the isolation of CD4' and CD8' T-cell
clones from islets of diabetic female NOD mice, Nagata
and Yoon (28) found that CD4+,but not CD8+,T-cell lines
could be isolated from animals < l o weeks of age and
concluded that CD4' T-cells may be involved primarily in
the initiation of disease, whereas CD8' T-cells are latestage effectors. A later study by this group indicated that
the CD8' CTL lines could induce disease in 7- to 8-weekold sublethally irradiated NOD recipients, but only in the
presence of CD4' T-cells (29). To isolate CD4' and CD8'
T-cell lines from islet infiltrates, Maugendre et al. (30)
used an interesting approach in which culture wells were
precoated with antibodies to T-cell receptor Vp6 or Vp8;
lines with both phenotypes were found to be cytotoxic
with CD3-coupled P815 targets or yeast artificial chromosome (YAC) cells, but islet cell specificity was not
determined. One VpGICD4 line could induce insulitis in
neonatal recipients, but only when injected together with
CD8' T-cells. Another novel approach to deriving isletspecific T-cell lines was used by Wegmann et al. (31), who
used syngeneic islet grafts in NOD recipients to recruit
autoreactive T-cells. Several lines with islet specificity
were obtained, and clones from one line, Nl, were shown
to be diabetogenic in NOD-scid recipients. Gelber et al.
(32) reported on the isolation of lines and clones that
were reactive with various fractions produced from
extracts of a p-cell insulinoma and that were in some
cases diabetogenic. Most recently, Healey et al. (33)
described CD4' T-cell lines that were reactive with
extracts from a rat insulinoma and that included both
Thl (IFN-y-secreting) T-cells and Th2 (IL-4-secreting)
T-cells. The Thl lines could induce insulitis and diabetes,
whereas irljections with the Th2 line resulted in only a
mild peri-islet infiltrate (33).
The diabetogenicity of these various CD4' and CD8' Tcell lines and clones is apparently quite variable from
group to group. In some reports, the pathogenesis
described was in the form of insulitis, not overt diabetes
(23,25). In several studies, including our own, CD4' T-cell
clones alone induced disease (18,25,27,31),but in others,
only a combination of CD4' and CD8' clones resulted in
pathogenesis (23,29,30).In the case of islet-reactive CD8'
T-cells, investigators generally have not observed disease
transfer when CD8' clones are used alone, but only when
they are cotransferred with CD4' lines or spleen cells.
However, Wong et al. (34) have recently reported on CD8
T-cell clones isolated from 7-week-old female NOD mice
and cultured on islets expressing the R7-1 costimulatory
TABLE 1
Islet-reactive T-cell clones
Origin of T-cells
CD41CD8
Antigen specificity
-
Spleen and lymph nodes
of diabetic females
Islet infiltrates of diabetic
females
Islet infiltrates of 20-week-old
nondiabetic females
Islet infiltrates of 7- to 11week-old nondiabetic mice
Islet infiltrates of 8-week-old
nondiabetic females
Infiltrated islets of an irradiated
male recipient
Islet infiltrates of ~ 1 0and
- >10week-old nondiabetic females
Infiltrated islet isografts
Spleen and lymph nodes of 30- to
40-day-old nondiabetic females
Lymph nodes of diabetic females
Islet infiltrates of 7-week-old
nondiabetic females
Investigators
-
Islet cells,
P-granule membrane
Islet cells
Insulitis and diabetes
Insulitis
Haskins et al. (13-22),
Bergman and Haskins (35)
Reich et al. (23)
Islet cells
ND
Nagata et al. (24)
Islet cells
Insulitis
Nakano et al. (25)
Islet cells
Insulitis and diabetes
Pankewycz et al. (26)
Islet cells
Diabetes (CD4 clones only)
Shimizu et al. (27)
Islet cells
Islet cells
Insulinoma extracts
Diabetes (CD8 with
cotransfer of CD4 T-cells)
Insulitis and diabetes
Insulitis
Nagata and Yoon (28)
Nagata et al. (29)
Wegmann et al. (31)
Gelber et al. (32)
Islet cells
Islet cells
Diabetes and insulitis
Diabetes
Healey et al. (33)
Wong et al. (34)
molecule. These CD8' T-cell lines and clones were
H-2K"restricted and were found to proliferate and be
cytotoxic to NOD islets in vitro. In adoptive-transfer
experiments to NOD- or CB17-scid recipients, diabetes
developed rapidly and in the absence of CD4' T-cells.
Antigen specificity of islet-reactive clones. All of
the T-cell clones described above (and summarized in
Table 1) are termed islet-reactive because they respond
to islets or islet cells in T-cell assays in vitro. Purification
of unknown antigens from islet tissue is not a simple
task, and in most cases, little is known about the antigen
targets for these T-cells. Several labs have reported CD4'
NOD T-cell lines or clones that react to islet cells from
multiple mouse strains in a NOD class 11-restricted manner (14,25,27), whereas CD8' T-cells respond to islet
cells alone (23,27). The work of Shimizu et al. (27) indicated that the antigen for their CD4' T-cell clones could
not be detected in conditioned media from islet cells,
but live APCs with formaldehyde-fixed islet cells did
stimulate responses, suggesting that a membrane-associated protein was responsible for antigenicity.
Evidence for an undefined membrane-associated antigen was further provided by studies with the BDC panel
of CD4' islet-specific T-cell clones. From an extensive
analysis of the subcellular location of islet antigens for
these clones, Bergman and Haskins (35) found that a
fraction that was highly enriched in p-cell granules was
antigenic for every islet-specific T-cell clone tested from
the panel, and we subsequently localized that activity to
the p-granule membrane. More recent work has indicated
that the antigenic activity for at least two T-cell clones
can be isolated by anion exchange chromatography of
detergent-solubilized p-cell membranes and further purified by size-exclusion chromatography, yielding an active
fraction that falls into a molecular weight range of 50-80
kDa (B. Bergman, J.L. McManaman, K.H., unpublished
observations).
1302
In vivo activity
-
Because of the unique islet cell response of the T-cell
clone BDC-6.9, Haskins and colleagues were also able to
take a genetic-mapping approach to identifying the antigen for one of the BDC clones (36). It was found that
BDC-6.9 reacts only to islet cells from the NOD mouse,
from its related strains NOR and NON mice, and from
SWR mice, whereas most of the BDC clones, including
BDC-2.5, react to islet cells from all of the inbred mouse
strains tested. We produced an (NOD X BALB/c) X
BALB/c backcross and looked for correlations between
chromosomal loci mapped by microsatellite markers and
the presence or absence of antigenicity in islets from
each backcross mouse. The result was that a locus in the
telomeric region of mouse chronlosome 6 was found to
have a 100%correlation with the presence of the islet antigen in the backcross (36).
T-CELL CLONES REACTIVE TO DEFINED ISLET ANTIGENS
Insulin-specific T-cell clones. As more information has
become available with regard to p-cell autoantigens
defined by autoantibody reactivity, investigators in
recent years have begun to ask whether these autoantigens are also T-cell targets. Because insulin comprises
-80% of the p-cell protein, as well as being critical in prevention and control of diabetes, its role as an autoantigen
is of great interest. The first report of insulin-specific
T-cell lines came from Wegmann e t al. (37), who
described CD4+T-cell lines and clones isolated from the
islet infiltrates of unimmunized 7- and 12-week-old NOD
mice. Of T-cell clones isolated from the older mice, about
half were found to react with insulin; insulin reactivity
was also observed in clones from a line derived from the
7-week-old mice, but at a much lower frequency (14%).
Because these clones were initially cultured with islet
cells as antigen and because their use of T-cell receptor
V p regions was diverse, it was deemed unlikely that the
insulin reactivity was a result of in vitro selection.
DIABETES, VOL. 45, OCTOBER 1996
TABLE 2
T-cell clones reactive with defined islet antigens
Origin of T-cells
CD4/CD8
Antigen specificity
In vivo activity
Investigators
Islet infiltrates of 4- and 7week-old nondiabetic fenlales
3-month-oldfemale
Lymph nodes of 6-week-old
or diabetic females
CD4
Insulin
Insulitis and diabetes
CD4
CD4
HSPm
Insulitis
GAD67
ND
Wegmann et al. (37),
Daniel et al. (38)
Elias et al. (40)
Elliott et al. (44)
--
--
-
--
Further characterization of the insulin-specific T-cell
clones uras provided in a later report in which the functional properties and epitope specificity of the clones
were investigated (38). It was found that of six clones
derived from 12-week-oldNOD mice, five were Thl-like in
that they secreted IFN-y and no IL-4; one clone, however,
secreted both cytokines. To varying degrees, all of these
T-cell clones showed diabetogenic potential in young
(~14-day-old)NOD recipients, causing diabetes andlor
extensive insulitis. One clone was also able to transfer
diabetes into the lymphocyte-deficient NOD-scid mouse.
To determine which insulin epitopes were antigenic for
the insdin-specific T-cell clones, overlapping 15-amino
acid synthetic peptides from both isoforms of mouse
insulin were tested, and interestingly, all of the six clones
and a number of uncloned insulin-reactive T-cell lines
responded only to the peptide encompassing residues
9-23 of the insulin B chain. This restricted antigen specificity for peptide B-(9-23) is different from insulin-specific
T-cells derived from non-diabetes-prone mouse strains
immunized with insulin and reactive to either the insulin
A chain or a combinatorial A chain-B chain determinant;
it may be that the B-(9-23) peptide is unique in terms of
being autoantigenic. The importance of the B-(9-23) peptide has been further confirmed by a recent study by
Daniel and Wegmann (39) in which it was shown that this
was the dominant epitope for >300 insulin-specific T-cell
clones and that intranasal and subcutaneous administration of this peptide resulted in significant delay in onset
and decrease in incidence of diabetes compared to that in
mice given a control peptide from tetanus toxin.
CD4' insulin-specific T-cell lines have also been successfully produced from NOD mice that were immunized
with va.rious forms of insulin, including bovine, ovine,
and porcine insulin as well as A or B chains from bovine
insulin, and both A and B chain determinants were found
to be antigenic (A. Cooke, unpublished observations).
The unique autoantigenic nature of the insulin B-(9-23)
peptide was again demonstrated by the isolation of a diabetogenic CD4' T-cell clone from an NOD mouse immunized with this peptide (D. Healey and A. Cooke, personal
communication).
T-cell clones reactive t o other p-cell antigens. Other
p-cell proteins have also been identified as T-cell antigens.
T-cell reactivity to heat shock protein (HSP) was reported
by Elias et al. (40), and HSP,, reactive T-cell lines were
found to induce disease in NOD mice. Further, vaccination with an HSP,, peptide was found to protect against
development of diabetes in NOD mice (41). Work from
two laboratories indicated that T-cell reactivity to GAD,,
or several GAD,, peptides could be detected in splenoDIABETES, VOL. 45, OCTOBER 1996
cytes from unprimed NOD mice and that an antigen-specific tolerance could be induced with GAD,, (42,43).
Elliott et al. (44) immunized 4-week-old NOD mice with
the 67-kDa GAD isoform (GAD,,), and found that this
form of GAD also prevented or delayed onset of disease.
Two T-cell clones were isolated that proliferated in
response to NOD APCs and GAD, but the diabetogenic
potential of these clones was not determined. More
recently, Wegmann and coworkers have isolated NODderived CD4' T-cell clones reactive with GAD,, peptides
in vitro and diabetogenic in young NOD recipients (D.W.,
unpublished observations). Cooke and coworkers have
also isolated T-cell clones from NOD mice with other
known antigen specificities, including CD8' CTL clones
specific for both GAD isoforms and a CD4' Thl-like clone
reactive with carboxypeptidase H (A. Cooke, personal
communication). Published reports of T-cell clones reactive with defined islet antigens are summarized in Table 2.
CONCLUDING COMMENTS
The recent report of CD8' T-cell clones with diabetogenic
activity in scid mice and the several published reports of
diabetogenicity with CD4' T-cell clones provide definitive
evidence that either CD4' or CD8' T-cell clones are capable of mediating p-cell destruction. Therefore, there are
probably at least two distinct mechanisms by which
T-cells can destroy p-cells. One of the most straightforward is that of CD8' T-cell-mediated cytolysis. The observations that CD8' T-cell clones have potent cytolytic
activity in vitro and that they are capable of relatively
more rapid induction of diabetes than is generally
observed with CD4' T-cell clones suggest that direct
cytolysis is the mechanism of in vivo @-celldestruction by
CD8' T-cells. The importance of this mechanism to the
overall process of pathogenicity will depend on the relative contributions of CD8' and CD4' T-cells in this
process. With regard to CD4' T-cell clones, it has been
demonstrated that T-cell clones specific to either P-granule-associated proteins or insulin are capable of inducing
overt diabetes in NOD-scid mice. Although CD4' T-cells
are capable of directly lysing class I1 major histocompatibility complex (MHC)-positive cells in the presence of
appropriate antigen, the current consensus is that the
p-cell does not express MHC class 11, even in the environment of a local inflammatory response. Instead of a direct
mechanism of action, there is a growing body of evidence
that indicates that CD4' T-cells destroy p-cells by an indirect mechanism in which the T-cell is activated by antigen/APCs to secrete soluble effector molecules that promote destruction of the p-cells. Compelling evidence for
such a mechanism was provided by the results of experi-
ments in which CD4' T-cell clones, specific for insulin in
the presence of NOD APCs, were found to be capable of
destroying islet allografts and rat or human xenografts
(D.W., R. Gill, unpublished observations). Because these
T-cell clones cannot recognize antigen on a110 or xeno
MHC, the results indicate that p-cell destruction by CD4'
clones can occur as a consequence of recognition of the
antigen on a cell other than the target P-cell, a finding that
is much in keeping with the observations of Lo et al. (45)
in studies on antigen presentation in chimeric mice.
There has been increasing interest in the role of
cytokines in T-cell pathogenesis, particularly in those produced by Thl T-cells versus those produced by Th2
T-cells. The current paradigm is that cytokines such as
IL4 and IL-10, produced by Th2 T-cells, have the capacity
to both counteract the actions of and downregulate the
production of Thl products, such as IFN--y, leading to an
amelioration of the destructive process. Although this is
an attractive concept and there have been experiments in
which administration of Th2 T-cell cytokines to NOD mice
have resulted in protection from diabetes (46,47), it is also
true that manipulations such as transgene-mediated production of ILlO by p-cells have resulted in acceleration of
the development of diabetes in NOD mice (48). The in
vivo activities of islet-specific T-cell clones of either Thl
or Th2 phenotypes have just begun to be assessed, and
thus far, the results provide little clarification. In the first
report of Th2 T-cell lines from NOD mice, Healey et al.
(33) observed that in comparison with a Thl line that
rapidly caused diabetes and extensive insulitis in neonatal NOD recipients, animals injected with a Th2 line did
not become diabetic within the 30-day period of the
experiment and had a nondestructive peri-ductal infiltrate. In a set of experiments by Katz et al. (491, in which
mice transgenic for the T-cell receptor of the diabetogenic
CD4' T-cell clone BDC-2.5 (13) were used as donors for
transfer of in vitro activated spleen cells to recipient
mice, it was observed that spleen cells activated under
Thl-promoting conditions transferred diabetes with good
efficiency. On the other hand, spleen cells activated under
Th2-promoting conditions were not capable of inducing
diabetes but were also unable to protect mice from the
actions of Thl cells. In a third investigation, it was found
that both Thl-like and Th2-like insulin-specific T-cell
clones could transfer diabetes and that a clone that could
be switched to either a Thl or a Th2 phenotype in vitro
was diabetogenic under both conditions (D.W., unpublished observations). Thus, it is not yet clear that Th2 Tcells can protect against diabetes in the NOD mouse.
Finally, there is the issue of autoantigen specificity of
T-cell clones. Diabetogenicity has been demonstrated for
T-cell lines and clones reactive to undefined islet antigens
and, in a few instances, to defined f3-cell antigens. It is
apparent that T-cell reactivities to a variety of islet cell proteins arise during the development of disease, and some of
them have been detected in or isolated from mice as young
as 4 weeks. There is extensive evidence from the studies
of Bergman and Haskins (35, unpublished results) and a
strong indication from the work of Shimizu et al, (27) as
as from the findings with human
'lone' (lo),
that an important antigen or group of antigens is associated with the p-granule membrane. In these studies, no
1304
antigenicity was found in soluble supernatant or subcellular cytosolic fractions. On the other hand, T-cell specificities have now been demonstrated for several known p-cell
proteins, and in some instances, treatment with these antigens under tolerizing conditions can ameliorate or inhibit
disease. The important question that remains to be
answered is whether there is any one T-cell autoantigen
that is dominant in the initiation of disease and whether
identification of that antigen will have therapeutic implications. The induction of antigen-specific tolerance in
autoimmune disorders is being investigated by many labs,
and evidence to date suggests that at least in the case of
IDDM, there may be numerous antigens, including insulin
and GAD, administered by various routes, that can prevent
or inhibit disease in highly controlled, genetically identical
mice. There are also many studies indicating that a wide
variety of relatively nonspecific methods-including treatment with antibodies to several T-cell molecules and
inhibitors or upregulators of cytokines-are efficacious in
animals (6). Thus, it is difficult to predict at this time
whether an antigen-specific approach will be better in the
highly diverse human population than other approaches
that are broader in spectrum. To return to the first part of
the question regarding the importance of identifying an initiating antigen, it is improbable that every possible
autoantigen in the islet p-cell would have the same potential for invoking a primary immune response; therefore, it
seems likely that one antigen or group of antigens is
involved in the initiation of disease. Identification of this
antigen could contribute much in terms of gaining an
understanding of how the disease process begins. Part of
what has been gained from investigating islet antigens
such as insulin and GAD lies in the fact that autoantibodies to these proteins have some value as predictive markers of disease. Since the ultimate approach to treating
IDDM would be the rescue of islet function before autoimmune destruction begins or becomes advanced, improving the early diagnosis of IDDM is a high priority goal for
clinicians and researchers, and it is in this regard that the
identification of a dominant disease-initiating antigen is
the most important.
REFERENCES
1. Gianani R, Eisenbarth GS: Humoral autoimmunity. In Qpe I Diabetes:
Molecula~;Cellular and Clinical Zmmu~n~ology.
Eisenbarth GS, Lafferty
K, Eds. New York, Oxford Univ. Press. In press
2. Eisenbarth G, Verge C, Allen H, Rewers M: The design of trials for prevention of IDDM. Diabetes 42341-947, 1993
3. Miyazaki A, Hanafusa T, Yamada K, Miyagawa J, wino-Kurihara H,
Nakajima H. Nonaka K, Tarui S: Predominance of T lymphocytes in pancreatic islets and spleen of prediabetic non-obese diabetic (NOD) mice:
a longitudinal study. Clin Exp Immunol60:622-630, 1985
4. Kikutani H, Makino S: The murine autoimmune diabetes model: NOD and
related strains. Ad?, Immunol 51:285-322, 1992
5. Bach JF: Insulin-dependent diabetes mellitus as an autoimmune disease.
Endo Reviews 15:516-542,1994
6. Peterson JD, Haskins K: In~munotherapyof animal models of insulindependent diabetes mellitus. Curr Opiti Endocrinol Diabetes 217-23,
,
* ~ . u -
7, Bendelac A, Carnaud C, Boitard C, Bach JF: Syngeneic transfer of autoimmune diabetes from diabetic NOD mice to healthy neonates: requirement
for both L3T4+and Lyt-2+T cells. J Exp llfed 166:823-832, 1987
8. Miller BJ, Appel MC, O'Neil JJ, Wicker LS: Both the LYT-2' and L3T4' T
cell subsets are required for the transfer of diabetes in nonobese diabetic
mice, Jlm,nwnol
140:52-58, 1988
9. Christianson SW, Shultz ~ ~ , ' ~ e EH:
i t e Adoptive
r
transfer of diabetes into
DIABETES, VOL. 45, OCTOBER 1996
inlmunodeficient NOD-scidlscid mice: relative contributions of CD4'
and CD8' T-cells from diabetic versus prediabetic NOD.NON-Thy-la
donors. Diabetes 4244-55, 1993
10. Roep RO, Arden SD, deVries RR, Hutton JC: T-cell rlones from a type-1
diabetes patient respond to insulin secretory granule proteins. Nature
345632-634, 1990
11. Durinovic-Bello I, Steinle A, Ziegler AG, Schendel DJ: HLA-DQrestricted, islet-specific T-cell clones of a type I diabetic patient. Diabetes 43:1:318-1325, 1994
12. Prud'homme GJ, Fuks A, Colle E, Guttmann RD: Isolation of T-lyrnphocyte lines with specificity for islet cell antigens from spontaneously diabetic (insulin-dependent) rats. Dinbetes 33401-803, 1984
13. Haskins K, Portas M, Bradley B, Wegmann D, Lafferty K: T-lymphocyte
clone specific for pancreatic islet antigen. Diabetes 37:1444-1447. 1988
14. Haskins K, Portas M, Bergman B, Lafferty K, Bradley B: Pancreatic isletspecific T-cell clones from nonobese diabetic mice. Proc Natl Acad Sci
USA 86:8000-8004, 1989
15. Peterson JD, Haskins K: Transfer of diabetes in the NOD-scid mouse by
CD4 T-cell clones. Diabetes 45:328-336, 1996
16. Candeias S, Katz J, Benoist C, Mathis D, Haskins K: Islet-specific T-cell
clones from nonobese diabetic mice express heterogeneous T-cell receptors. Pmc Natl Acad Sci USA 88:6167-6170, 1991
17. Bradley BJ, Wang Y, Lafferty KJ, Haskins K: In viva activity of an isletreactive T-cell clone. J Autoimm~rnity3:449456, 1990
18. Haskins K, McDuffie M: Acceleration of diabetes in young NOD mice
with a CD4+islet-specific T cell clone. Science 249:1433-1436, 1990
19. Peterson JD, Pike B, Dallas-Pedretti A, Haskins K: Induction of diabetes
with islet-specific T-cell clones is age dependent. Immunology
85:455-460, 1995
20. Peterson JD, Pike B, McDuffie M, Haskins K: Islet-specific T cell clones
transfer diabetes to NOD F1 mice. J. Immunol 153:2800-2806, 1994
21. Bradley W,Haskins K, La Rosa FG, Lafferty W: CD8 T cells are not
required for islet destruction induced by a CD4' islet-specific T-re11
clone. Diabetes 41:1603-1608, 1992
22. Haskire K, Bradley RJ, Lafferty KJ: Islet-graft destruction by a CD4+ Tcell clone is not CD8 T-cell dependent. Diabetes Nulr Metab 5 (Suppl.
1):lll-114, 1992
23. Reich EP, Sherwin RS, Kanagawa 0,Janeway CA: An explanation for the
protective effect of the MHC class I1 I-E molecule in murine diabetes.
Natwre 341:326-328, 1989
24. Nagatii M, Yokono K, Hayakawa Ivl, Kawase Y, Hatamori N, Ogawa W,
Yonezawa K, Shii K, Baba S: Destruction of pancreatic islet cells by
cytotoxic T lymphocytes in nonobese diabetic mice. J Iinmunol
1431155-1162, 1989
25. Nakano N, Kikutani H, Nishimoto H, Kishimoto T: T cell receptor V gene
usage of islet p-cell-reactive T cells is not restricted in non-obese diabetic mice. J Exp Med 173:1091-1097, 1991
26. Pankewycz 0 , Strom TB, Rubin-Kelley VE: Islet-infiltrating T cell clones
from non-obese diabetic mice that promote or prevent accelerated onset
diabetes. E u r J Immunol21:873-879, 1991
27. Shimizu J , Kanagawa 0, Unanue ER: Presentation of p-cell antigens to
CD4' and CD8' T cells of non-obese diabetic mice. .J Im?nunol
151:1723-1730, 1993
28. Nagata M, Yoon JW: Studies on autoimmunity for T-cell-mediated betacell destruction: distinct difference in beta-cell destruction between
CD4' and CD8' T cell clones derived from lymphocytes infiltrating the
islets of NOD mice. Diabetes 41:998-1008, 1992
29. Nagata M, Santamaria P, Kawamura T, Utsugi T, Yoon J: Evidence for the
role of CD8' cytotoxic T cells in the destmction of pancreatic p-cells in
nonot~esediabetic mice. J Immunol 152:2042-2050, 1994
30. Maugendre D, Lagrand B, Olivi M, Bedossa P, Bach .JF, Carnaud C: Establishment of T-cell lines from infiltrated NOD islets by stinlulation of specific T-cell receptor VP segments. J Autoimmunity 6:423436, 1993
DIABETES, VOL. 45, OCTOBER 1996
31. Wegmann DR, Shehadeh N, Lafferty KJ, Norbury-Glaser M, Gill RG,
Daniel D: Establishment of islet-specific T-cell lines and clones from islet
isografts placed in spontaneously diabetic NOD mice. J Autoimmu~fity
6:517-527, 1993
32. Gelber C, Paborsky L, Singer S, McAteer D, Tisch R, Jolicoeur C, Buelow
R, McDevitt H, Fathman CG: Isolation of nonobese diabetic mouse Tcells that recognize novel autoantigens involved in the early events of
diabetes. Diabetes 43:33-39, 1994.
33. Healey D, Ozegbe P, Arden S, Chandler P, Hutton J, Cooke A: In vivo
activity and in vitro specificity of CD4' Thl and Th2 T cells derived from
the spleens of diabetic NOD mice. J Clin Invest 952979-2985, 1994
34. Wong FS, Visintin I, Wen L, Flavell RA, Janeway CA: CD8 T cell clones
from young nonobese diabetic (NOD) islets can transfer rapid onset of
diabetes in NOD mice in the absence of CD4 cells. J Exp Med 183:67-76,
1996
35. Bergman B, Haskins I<: Islet-specific T cell clones from the NOD mouse
respond t o p-granule antigen. Diabetes 43:197-203, 1994
36. Dallas-Pedretti A, McDuffie M, Haskins K: A diabetes-associated T cell
autoantigen maps to a telonleric locus on mouse chromosome 6. Proc
Natl Acad Sci LISA 92:1386-1390, 1995
37. Wegmann DR, Norbury-Glaser M, Daniel D: Insulin-specific T cells are a
predondnant component of islet infiltrates in pre-diabetic NOD mice.
E u r J Irt~munol24:1853-1857. 1994
38. Daniel D, Gill RG, Schloot N, Wegmann D: Epitope specificity, cytokine
production profile and diabetogenic activity of insulin-specific T cell
clones isolated from NOD ndce. E u r J Immunol25:1056-1062, 1995
39. Daniel D, Wegmann DR: Protection of nonobese diabetic mice from diabetes by intranasal or subcutaneous administration of insulin peptide B(9-23). Proc Null Acad Sci USA 93:956-960, 1996
40. Elias D, Markovits D, Reshef T, van der Zee R, Cohen IR: Induction and
therapy of autoimmune diabetes in the non-obese diabetic (NODLt)
mouse by a 65-kDa heat shock protein. Proc Natl Acad Sci D3A
87:1576-1580, 1990
41. Elias D, Reshef T, Birk OS, van der Zee R, Walker MD, Cohen 1R: Vaccination against autoimmune mouse diabetes with a T-cell epitope of
the human 66-kDa heat shock protein. Proc Natl Acad Sci LISA
88:3088-3091, 1991
42. Tisch R, Yang X-D, Singer SM, Liblau RS, Fugger L, McDevitt HO:
Immune response to glutamic acid decarboxylase correlates with insulitis in non-obese diabetic mice. Nature 366:72-75, 1993
43. Kaufman DL, Clare-Salzler M, Tian J, Forsthuber T, Ting GSP, Robinson
P, Atkinson, MA, Sercan EE, Tobin AJ,Lehmann PV: Spontaneous loss of
T-cell tolerance t o glutamic acid decarboxylase in murine insulin-dependent diabetes. Nature 366:69-72, 1993
44. Elliott JF, &in H-Y, Bhatti S, Smith DK, Singh RK, Dillon T, Lauzon J,
Singh B: Immunization with the larger isoform of mouse glutamic acid
decarboxylase (GAD67) prevents autoimmune diabetes in NOD mice.
Diabetes 431494-1499, 1994
45. Lo D, Reilly CR, Scott B, Liblau R, McDevitt HO, Burkly LC: Antigen-presenting cells in adoptively transferred and spontaneous autoimmune
diabetes. E u r J Immunol23: 1693-1698, 1993
46. Rapoport IW,Jaramillo A, Zipris D, Lazarus AH, Serreze DV, Leiter EH,
Cyopick P, Danska JS, Delovitch TL: Interleukin 4 reverses T cell proliferative unresponsiveness and prevents the onset of diabetes in nonobese
diabetic mice. J E.rp Med 17837-99, 1993
47. Pennline KJ, Roque-Gaffney E, Monahan M: Recombinant human ILlO
prevents the onset of diabetes in the nonobese diabetic mouse. Clin
I m n ~ u n oImmu?lopathol
l
71:169-175, 1994
48. Wogenson L, Lee M, Sarvetnick N: Production of interleukin-10 by islet
cells accelerates immune-mediated destruction of P cells in nonobese
diabetic mice. J Exp Med 179:1379-1384, 1994
49. Katz JD, Benoist C, Mathis D: T helper cell subsets in insulin-dependent
diabetes. Science 268:1185-1188, 1995