To simplify description of the results, all noted differences between diagnostic groups were significant unless otherwise indicated (see figures and tables for P values). Differences reported across diagnostic groups were significant in both sexes. All IR, IGF-1R, and IRS-1 phosphorylation sites are numbered according to the human sequences. IR amino acid numbers are those for the isoform found in the brain (IR-A; see Discussion). Demographic, autopsy, neuropathological features, and cognitive data on all subjects are summarized in Supplemental Table 1.
Total basal levels of insulin and IGF-1 signaling molecules in the cerebellar cortex and HF are normal in AD.
In contrast to some earlier reports (28
) on the HF and/or frontal cortex in AD, Western blotting and qIHC showed that total non-phosphospecific levels of signaling molecules in the IR/IGF-1R→IRS→PI3K pathway (IRβ, IGF-1Rβ, IRS-1, IRS-2, Akt1, GSK-3β, mTOR, and ERK2) were normal in the cerebellar cortex and HF of AD cases compared with normal controls (referred to herein as N cases) matched in sex, age, and low postmortem interval (PMI; 6–12 hours). There was, however, a trend toward elevated IRS-1 in the HF (P
If insulin signaling is impaired in AD, then it would be evident only in the activation states of its signaling molecules, especially in response to applied insulin. We tested such responses using an ex vivo stimulation protocol with which we previously showed that nicotinic, neuregulin-1, and NMDA signaling is intact in human postmortem tissue obtained 6–11 hours after death on average (73
). Such findings are not surprising, since neurons in thick brain sections from postmortem N and AD cases with PMIs up to 8 hours can be kept alive in culture media at least 3 weeks without significant loss in numbers, morphology, or measures of viability (75
). To validate the ex vivo protocol for insulin signaling studies on postmortem tissue, we tested it first on normal brain tissue.
Insulin signaling mechanisms are intact in low-PMI brain tissue.
Ex vivo tests on human HF slices from 5 N cases with PMIs of 5–19 hours demonstrated that 0.1–100 nM insulin evoked clear, reliable, dose-dependent activation of the insulin signaling pathway under study. Increasing doses caused increasing tyrosine phosphorylation (pY) of the IR kinase regulatory domain (Y1150, Y1151), IRβ binding (i.e., recruitment) of IRS-1, and activation of such downstream molecules as Akt1 pS473 and ERK2 pT185/pY187 (Figure A). The functional integrity of the postmortem tissue was further indicated by its strong responsiveness to glutamate-induced glucose uptake (see below).
Ex vivo stimulation is a valid method for studying insulin signaling in postmortem HF.
Since responses at 10 nM approached those at 100 nM, we limited further testing to the 1-nM dose, close to physiological levels of brain insulin (76
), and the 10-nM dose, commonly used in studies on insulin signaling in peripheral tissues (58
). Tests on rat HF showed that the magnitude of insulin signaling responses did not diminish significantly with PMIs as long as 16 hours (Figure , B–D). In human HF from cases with mean PMIs of 6 hours, the magnitude of the same insulin signaling responses was as large as in the rat HF (compare Figure , C and D, and Figure , A and E).
At near-physiological doses (1 nM), insulin and IGF-1 activate different IRS signaling pathways.
Near-physiological doses of brain insulin and IGF-1 selectively activate their cognate receptors.
When tested in adult mammalian brains, levels of extracellular IGF-1 range 1.12–2.38 ng/ml (0.14–0.31 nM) (78
). Corresponding extracellular data are not available for insulin, but the upper limit can be estimated from total insulin levels in the mammalian brain, commonly ranging 0.2–8 ng/g wet weight (76
). Given that the density of brain tissue (1.05 g/cc) is about that of water (1.05 g/ml), extracellular brain insulin would normally be no higher than 0.19–7.6 ng/ml (0.033–1.31 nM). Thus, a 1-nM dose of insulin or IGF-1 is probably close to, or somewhat above, physiological levels of those hormones in the brain. A 10-nM dose is supraphysiological for both hormones.
Consistent with the view that 1 nM insulin or IGF-1 exert physiological effects, ex vivo tests at that dose revealed selective signaling effects of these hormones. In rat HF, 1 nM insulin activated IRβ and induced IRβ binding to IRS-1, but did not activate IGF-1Rβ or induce IGF-1Rβ binding to IRS-1 (Figure , C and D, and Supplemental Figure 2). This finding was replicated in HF and cerebellar cortex of N cases, where it was also found that 1 nM IGF-1 did not activate IRβ or induce IRβ binding to IRS-1 (Figure , C and G, Figure D, Figure D, and Supplemental Figure 3). In contrast, 10 nM doses of insulin or IGF-1 activated each other’s receptors, though not maximally (Figure ). The same results were found in AD cases (see below) and applied to all activation sites studied in the IR, namely Y1150/1151 in its kinase domain and Y960 in its IRS-1 binding domain (79
), and in the IGF-1R, namely Y1135/1136 and Y1131 in its kinase domain (81
Ex vivo stimulation revealed IRS-1–associated insulin resistance and IRS-2–associated IGF-1 resistance in the cerebellar cortex of AD cases.
Ex vivo stimulation revealed IRS-1–associated insulin resistance and IRS-2–associated IGF-1 resistance in HF of AD cases.
Near-physiological doses of brain insulin and IGF-1 signal via different IRS isoforms.
This was demonstrated in ex vivo tests on cerebellar cortex and HF of N cases. At the 1-nM doses shown to selectively activate their cognate receptors, insulin stimulated IRβ binding to IRS-1, not IRS-2, whereas IGF-1 stimulated IGF-1Rβ binding to IRS-2, not IRS-1 (Figure , E–H, Figure , and Figure ). As that result predicted, 1 nM insulin activated IRS-1, not IRS-2, whereas 1 nM IGF-1 activated IRS-2, not IRS-1 (Figure , I–L, Figure , and Figure ). Thus, as expected, 1 nM insulin induced IRS-1, not IRS-2, binding to PI3K p85α, whereas 1 nM IGF-1 induced IRS-2, not IRS-1, binding to PI3K p85α (Figure , M–P, Figure , and Figure ). These dichotomous responses to 1 nM insulin and IGF-1 (seen to a lesser degree at 10 nM doses) were also found in AD cases of both brain areas studied (Figures and ). At near-physiological doses, then, these insulin and IGF-1 signaling pathways did not converge upstream of PI3K. Our tests of resistance to insulin and IGF-1 thus focused on results with 1-nM doses of these hormones.
Insulin resistance associated with IRS-1 dysfunction occurs in the cerebellar cortex and more markedly in the HF of AD cases.
Ex vivo responses to 1 and 10 nM insulin were tested in 8 pairs of N and AD cases from the set of cases characterized in Supplemental Table 1. Members of each pair were well matched for age (N, 85.5 ± 7.9 years; AD, 84.2 ± 5.3 years; mean ± SD), sex (6 female, 2 male in both groups), and PMI (N, 6.02 ± 2.6 hours; AD, 5.92 ± 2.7). None of the N or AD cases had a history of diabetes. The same cases were studied for comparison of the cerebellar cortex and HF. For testing insulin resistance, the antigen panel was extended to IRS-1 pY941
in rodents), which is critical for activating the regulatory subunit (p85) of PI3K (47
Testing IRβ pY1150/1151 distinct from the homologous sequence in IGF-1Rβ pY1135/1136 was accomplished by first immunoprecipitating each receptor with an antibody to nonhomologous regions of the 2 receptors and then immunoblotting with an antibody to the shared phosphospecific region (see Methods). The phosphorylation levels of all molecules were expressed as ratios of phosphorylated to total antigen levels and were thus independent of neuronal numbers in the samples studied.
In the cerebellar cortex and HF of both the AD and N cases, insulin induced activation of IRβ (pY1150/1151 and pY960) and IRS-1 (total pY and pY612) as well as IRβ binding of IRS-1 and IRS-1 binding of PI3K p85α (Figures and and Supplemental Tables 2 and 3). In AD cases, however, the percent increase in these insulin responses above baseline levels was less than in N cases at all levels of the insulin signaling pathway studied (Figure and Tables and ). Except for activation of IRβ pY960 in the cerebellar cortex, the reduced responsiveness to 1 nM insulin in both structures was modest at the level of the IR, but moderate to strong with respect to IRS-1 and its interactions with PI3K p85α (Figure and Tables and ). While there were marked reductions in IRS-1 activation and binding of IRβ and PI3K p85α in response to 1 nM insulin in AD, there were no such reductions in IRS-2 responses to that dose of insulin (Tables and ).
Direct demonstration of insulin and IGF-1 resistance in the cerebellar cortex (A–L) and HF (M–X) of AD cases without diabetes.
Difference in cerebellar cortex signaling responses to insulin stimulation between AD cases and matched controls
Difference in hippocampal formation signaling responses to insulin stimulation between AD cases and matched controls
Insulin resistance in AD cases was more advanced in the HF than in the cerebellar cortex. While cerebellar responses to 1 nM insulin were significantly lower at all levels of the signaling pathway tested, such responses to 10 nM insulin were reduced to a lesser degree and were often insignificant (Figure and Table ). Insulin resistance in cerebellar cortex was thus ameliorated at the 10-nM dose. However, that was not the case in the HF, where significant insulin resistance was seen at both 1 and 10 nM insulin (Table ). At the 1-nM dose, moreover, the HF displayed greater reductions in responsiveness below the IR, as seen in insulin-induced IRS-1 pY (90% reduced in the HF vs. 42% in cerebellar cortex), IRS-1 pY612 (86% vs. 54%), IRS-1 pS616 (90% vs. 68%), and IRS-1 binding to PI3K p85α (96% vs. 76%) (compare Tables and ).
The more pronounced insulin resistance below IR seen in the HF compared with the cerebellar cortex was associated with increased basal levels of IRS-1 pY and IRS-1 pS. Total basal IRS-1 pY and IRS-1 pY612
in the AD cases were normal in the cerebellar cortex (Table ), but were highly elevated in the HF, along with elevated basal levels of IRS-1 bound to PI3K p85α (Table ). These conditions are known to attenuate IRS-1 signaling (82
). Total basal IRS-1 pS in AD was also normal in the cerebellar cortex (Table ), but highly elevated in the HF (Table ), another condition known to attenuate insulin signaling (42
IGF-1 resistance associated with IRS-2 dysfunction is severe in both the cerebellar cortex and the HF of AD cases.
In the same samples that showed insulin resistance in AD, IGF-1 resistance was discovered in the IGF-1R→IRS-2→PI3K pathway. Responses to 1 and 10 nM IGF-1 were reduced at all tested levels of that pathway (Figures –) and were nearly always greater in the HF than the cerebellar cortex (Tables and ). In other respects, however, IGF-1 resistance clearly differed from insulin resistance. First, stimulus-induced receptor activation, as shown by IGF-1Rβ pY1135/1136 and IGF-1Rβ pY1131 levels, was strongly reduced and was not significantly ameliorated at 10 nM IGF-1, even in the cerebellar cortex (Tables and ). Second, while the AD cases showed marked reductions in 1 nM IGF-1–induced activation of IRS-2 and of IRS-2 binding to IGF-1Rβ and PI3K p85α, they showed no such corresponding effects of IGF-1 on IRS-1 (Tables and ). The increase in total basal levels of IRS-2 in both brain areas studied was thus probably associated with resistance to IGF-1, not insulin. Since we lacked clues to the proximal causes of brain IGF-1 resistance, we did not pursue that phenomenon further. Instead, we next focused on the many clues to the potential proximal causes of brain insulin resistance.
Difference in cerebellar cortex signaling responses to IGF-1 stimulation between AD cases and matched controls
Difference in hippocampal formation signaling responses to IGF-1 stimulation between AD cases and matched controls
IRS-1 pS616 and IRS-1 pS636/639 are candidate biomarkers of brain insulin resistance.
While total basal levels of IRS-1 pS were normal in the cerebellar cortex, basal levels of IRS-1 pS616
and IRS-1 pS636
were elevated there and in the HF of AD cases, unlike IRS-1 pS312
levels. These elevations, which were also seen in cases of mild cognitive impairment (MCI; see below), were among the few basal abnormalities in insulin signaling molecules consistently associated with insulin resistance (compare Tables and ). Consequently, IRS-1 pS616
and IRS-1 pS636
are candidate biomarkers of brain insulin resistance, especially since their elevation is a feature of insulin resistance in peripheral tissues (43
). The cause of such elevations was suggested by further study of the HF, where basal elevations in IRS-1 pS616
and IRS-1 pS636/639
in AD were accompanied by basal elevations in activated forms of kinases directly or indirectly phosphorylating IRS-1 at S616 and/or S636. These activated kinases were Akt1 (pS473
), mTOR (pS2448
), and ERK2 (pT185
), basal levels of which were elevated above normal levels in AD cases by 193%, 107%, and 179%, respectively (Table and Supplemental Table 3B). Such activation was not triggered by increased downstream insulin signaling via non–IRS-1 pathways, because 1 and 10 nM insulin in AD induced markedly less activation of these kinases and of both GSK-3β pY216
and GSK-3β pS9
(activated and suppressed forms, respectively; Table ).
Candidate biomarkers IRS-1 pS616 and IRS-1 pS636/639 and their activated kinases are commonly and markedly elevated in HF neurons of AD cases without diabetes.
To test the generality and cellular locus of IRS-1 pS elevations in AD and to identify their likely causes and consequences, we studied the relatively large UPenn and ROS cohorts described above. We began with our discovery cohort from UPenn, a set of 24 N and 24 AD cases without a history of diabetes matched pairwise for sex, age within 5 years, and PMI within 5 hours (Supplemental Table 1).
qIHC was chosen for this phase of our study, since it allows selective quantification of neuronal (as opposed to glial) proteins in a precisely defined anatomical field and simultaneous processing of many cases. We focused on hippocampal field CA1, given its relatively high levels of IRs (64
) and IRS-1 (84
); its vulnerability to AD pathology (61
); and its large size, encompassing as many as 3,200 neuronal profiles per 6-μm section. The findings necessarily reflect chronic antigen levels, given that the PMIs were on the order of hours, not minutes.
To study IRS-1 pS species in the context of other changes that may occur in the IR→PI3K signaling pathway of AD cases, a large set of insulin signaling and regulating molecules was quantified in CA1 using the antibodies and IHC conditions shown in Supplemental Table 6. Since antibodies to IRβ pY1150/1151 also recognize IGF-1Rβ pY1135/1136, we refer to the antigen as IR/IGF-1Rβ pY hereafter. In the gray matter of CA1, all signaling molecules tested were restricted to pyramidal neurons, except for activated mTOR in a small set of glial cells in AD cases. We measured cytoplasmic differences between N and AD cases, since the cytoplasm was the site at which the molecules tested were most consistently detected in AD. Measures of cytoplasmic antigen levels are detailed in Table and Supplemental Methods.
CA1 neuronal levels of insulin signaling and regulating molecules in ROS cases
The heat map in Figure summarizes the relative cytoplasmic levels of neuronal insulin signaling and regulating molecules in the CA1 of the UPenn cohort. Since the data are based on qIHC, they capture only basal levels of the molecules tested. Total levels of the signaling molecules were variably altered in AD. Levels of IRβ, PTEN, and Akt1 were unaffected. IRS-1 itself was increased, and GSK-3β was decreased. Activation states of these and related molecules, however, were abnormal in a large percentage of the AD cases. Levels of IR/IGF-1Rβ pY and IRβ pY960
were reduced 13% and 21%, respectively, in contrast to normal levels of basal IR pY in Western blots on the HF as a whole (Table ). Activated forms of IRS-1 were higher in AD (IRS-1 pY612
, 162%; IRS-1 pY941
, 73%), but its suppressed forms were much higher in the same cases (IRS-1 pS616
, 1,564%; IRS-1 pS636/639
; 259%). IRS-1 pS312
was also elevated (1,093%), perhaps reflecting a compensatory process, since such phosphorylation can promote IRS-1 function in mice despite the conclusion of cell-based studies that IRS-1 pS312
is suppressive (48
). Elevated basal IRS-1 pS and IRS-1 pY may impair insulin’s ability to further increase those IRS-1 species, helping explain why insulin’s effect on them was typically blunted in AD.
CA1 pyramidal cells in AD display marked elevation in cytosolic levels of IRS-1 pS species and their activated kinases.
Activated forms of downstream molecules were also elevated in AD: Akt1 pS473
(103%), Akt2 pS474
(166%), PKCζ/λ pT410/403
(248%), mTOR pS2448
(843%), IKKα/β pS176/180
(85%), and JNK1/2 pT183
(31%). Suppressed GSK-3α/β pS21/9
was elevated 25%. Reliable IHC reactions were not obtained for GSK-3 α/β pS279/216
) or ERK2 pT185
. These increases were not the result of antigen compression caused by cell atrophy, because the average CA1 neuronal size in AD cases was not different from that of matched N cases in the UPenn cohort (Supplemental Table 1).
Nitrotyrosine, a marker of inflammatory and oxidative stress associated with insulin resistance in T2D (88
), was elevated in AD cases along with both total and oligomeric Aβ plaque load. No alterations were found in levels of protein phosphatases (i.e., PP2A, PP2B, and PTP1B) known to act on the IR, IRS-1, and Akt (89
IRS-1 pS616 and IRS-1 pS636/639 are elevated in HF neurons of MCI and AD, regardless of APOE ε4 status.
All the significant findings made in our discovery cohort were next tested in CA1 of the ROS cohort, consisting of 30 N, 29 MCI, and 31 AD cases (Supplemental Table 1). The 3 diagnostic groups did not differ significantly in age, sex ratio, PMI, or years of education. A few cases in each group had a history of T2D, but the results were the same with or without them. In the AD cases, the mean size of CA1 neurons was 27% less than normal, which may account for the higher total IRS-1 and GSK-3α/β pS21/9 in the AD cases, but antigen compression is unlikely to account for the 144%–699% increases in IRS-1 pS616, IRS-1 pS636/639, IRS-1 pY612, and IRS-1 serine kinases (Table ).
In addition to replicating the findings in the UPenn cohort (Table ), testing the ROS cohort provided information not obtained in the discovery cohort. First, those cases carrying 1 or 2 copies of APOE ε4 did not differ significantly from noncarriers in levels of the tested insulin signaling or regulating molecules. Second, reliable IHC detection of phosphatidylinositol-triphosphate (PIP3), which was achieved only in the ROS cohort, showed that this was reduced in AD (Table ). Third, compared with N cases, MCI cases showed (a) reduced IR/IGF-1Rβ pY, but not IRβ pY960
; (b) elevated IRS-1 pS616
and IRS-1 pS636/639
without elevated IRS-1 pS312
, IRS-1 pY612
, or IRS-1 pY941
; (c) lower levels of PIP3 and total GSK-3β; and (d) increased nitrotyrosine (Table , Figure , Supplemental Figure 5, D–F and J–L, and Supplemental Figure 6, M–O). These findings were made not only in our MCI group as a whole, but also in its amnestic (n
= 12) and nonamnestic (n
= 17) subgroups. Since MCI cases, especially amnestic cases, have a higher than normal risk of developing AD (93
), our data suggest that brain insulin resistance may begin at a predementia stage of AD.
Key insulin signaling molecules seen immunohistochemically in CA1 neurons of N, MCI, and AD cases of the ROS cohort.
Contrary to first impressions (Supplemental Figure 5, M–O, and Supplemental Figure 6, A–L), MCI cases showed no significant increases in Akt1 pS473
, Akt2 pS474
, GSK-3α/β pS21/9
, IKKα/β pS176/180,
, or PKCζ/λ pT410/403
(Table ). Whether the same is true for activated levels of ERK known to phosphorylate IRS-1 at S616 and S636 (50
) could not be determined in the qIHC studies, because as noted above, we were unable to get reliable IHC reactions with antibodies specific to activated ERK.
Neuronal IRS-1 pS616 and IRS-1 pS636/639 are negatively correlated with basal activation of IR/IGF-1Rβ and positively correlated with basal activation of IRS-1 serine kinases and oligomeric Aβ plaque load.
Correlational analyses were run on the 90 ROS cases to assess the relationship of IRS-1 pS616 and IRS-1 pS636/639 to relevant insulin signaling and regulating molecules in CA1 (Table ). Consistent with a role for these IRS pS species in brain insulin resistance, their basal levels were negatively correlated with basal levels of activated IR/IGF-1Rβ pY, IRβ pY960, and PIP3 (r = –0.26 to –0.35). Consistent with a role for IRS-1 serine kinases in elevated basal IRS-1 pS616 and IRS-1 pS636/639, basal levels of these IRS-1 species were positively correlated with activated levels of mTOR (r = 0.32–0.38), JNK (r = 0.0.46–0.65), and PKCζ/λ (r = 0.55–0.65) as well as with levels of activated Akt1 (r = 0.46–0.56) indirectly phosphorylating IRS-1 via mTOR. Basal levels of these activated IRS-1 serine kinases were in turn positively and often highly correlated with total and oligomeric Aβ plaque loads (Supplemental Table 7).
Correlation of candidate markers of brain insulin resistance with basal levels of insulin signaling or regulating molecules in CA1 neurons
Insulin by itself has no effect on HF glucose uptake.
Searching for the physiological consequences of brain insulin resistance, we looked for evidence that insulin-induced glucose uptake is reduced in AD. Among the molecules mediating such uptake in peripheral tissues are activated Akt2 (pS474
), deactivated AS160 (i.e., TBC1D4; pT642
), and activated GluT4 (pS488
). Basal levels of these markers in CA1 of the UPenn cases were uninformative, however, being elevated for Akt2 pS474
, normal for AS160 pT642
, and reduced for GLUT4 pS488
(Figure ). More informative were ex vivo tests on the 8 matched pairs of N and AD cases, which showed that neither 1 nor 10 nM insulin affected HF levels of GLUT4 pS488
or AS160 pT642
in either group (Supplemental Figure 7). Since it is still possible for insulin to stimulate glucose uptake independent of AS160 (95
) and perhaps GLUT4 pS488
, we used the ex vivo stimulation paradigm to test the effect of 1 and 10 nM insulin on glucose uptake in HF slices and synaptosomes. The latter preparations were included to test predominantly neuronal tissue. Insulin had no effect on basal (i.e., non–depolarization-induced) [3
H] glucose uptake in whole or synaptosomal tissue of either N or AD cases (Figure ). As a positive control, we tested the ability of 10 μM glutamate to induce glucose uptake; 1 μM glycine was also added to facilitate NMDAR activation, in light of reports that NMDA (97
) and depolarization (99
) trigger such uptake in neural tissue. This stimulus readily evoked glucose uptake in both HF preparations, but the magnitude of the uptake was reduced in AD by 68% in tissue slices and 72% in synaptosomes (see Figure ).
Insulin by itself does not affect glucose uptake in HF slices (A and B) and HF synaptosomes (C and D) of N or AD cases.
Basal activation states of neuronal insulin signaling molecules are closely related to cognitive ability.
Using the available neuropsychological data on the ROS cohort (Supplemental Table 1), we focused on the relationship of neuronal insulin signaling and regulating molecules in CA1 to episodic memory, given the close association of CA1 atrophy in AD to this type of memory (72
). Linear regression analyses revealed that, apart from Akt 2, basal activation states of insulin signaling and regulating molecules in CA1 were highly related to episodic memory (Table ). The relationships were positive for molecular forms driving insulin signaling (IR/IGF-1Rβ pY, IRβ Y960
, and PIP3) and negative for those attenuating such signaling (IRS-1 pS, GSK-3 pS21/9
, IKKα/β pS176/180
, JNK1/2 pT183
, mTOR pS2448
, and PKCζ/λ pT410/403
) or likely to do so (chronic IRS-1 pY612
and IRS-1 pY941
; see Discussion). The same pattern of relationships was found with respect to working memory and an index of global cognition.
Linear regression prediction of episodic memory scores by CA1 neuronal insulin signaling and regulatory variables adjusted for age, sex, and years of education
The density of neurons with detectable cytoplasmic levels of IRS-1 pS616 showed the strongest association with cognitive ability (Table and Figure ). Its correlations with episodic memory, working memory, and global cognition were –0.66, –0.52, and –0.63, respectively. In a linear regression model, IRS-1 pS616 levels in CA1 adjusted for age, sex, and education together accounted for 47% of the variance in episodic memory scores (parameter estimate, –0.137; Table and Figure A). Inclusion of neurofibrillary tangle (NFT) densities and Aβ plaque load in the model did not notably diminish the association of IRS-1 pS616 levels with episodic memory scores, which remained highly significant (parameter estimate, –0.14). This suggests that the contribution of IRS-1 pS616 levels in CA1 to cognitive dysfunction in AD is independent of Aβ plaques and NFTs.
The density of CA1 neurons displaying cytoplasmic IRS-1 pS616 is inversely associated with episodic memory (A), working memory (B), and global cognition (C).