Address correspondence and reprint requests to H. Uri Saragovi, Department of Pharmacology and Therapeutics, Oncology/Cancer Center, McGill University, Lady Davis Institute-Jewish General Hospital, 3755 Cote St. Catherine, F-223, Montreal, Quebec, Canada, H3T 1E2. E-mail: Uri.Saragovi@mcgill.ca
A direct correlation between disease progression and reduced expression of TrkA receptor in cholinergic neurons has been documented in neurocognitive pathologies including Alzheimer's disease. We investigated whether reduced expression of TrkA protein might also correlate with the level of cognitive impairment in age-associated cognitive impairment. Quantitative and qualitative measurements of TrkA protein levels in the cortex and nucleus basalis of aged rats that had been well-characterized behaviorally as ‘unimpaired’, ‘mildly impaired’ or ‘fully impaired’ demonstrated significant changes in TrkA expression. In the mildly impaired cognitive state phenotypic silencing of TrkA was detected in neurons expressing TrkA at high density but before cholinergic atrophy or loss of TrkA+ neurons was detected. In the fully impaired cognitive state a significant loss in TrkA+ cholinergic neurons together with a more significant phenotypic silencing of TrkA expression then took place. These data suggest that TrkA+ cholinergic cells are associated with cognition, TrkA could be a biomarker of the cognitive state and phenotypic loss of TrkA precedes neuronal loss and probably sensitizes cells to death. We speculate that neurotrophic deficits may be a shared mechanism for cognitive decline in aging and Alzheimer's disease.
Cholinergic projection neurons of the basal forebrain nucleus basalis (NB) provide the major source of cholinergic innervation to the cerebral cortex (Mesulam et al. 1983) and play a key role in memory and attention (Coyle et al. 1983; Bartus 2000). During normal aging, these neurons can undergo slow atrophy and degeneration, while in Alzheimer's disease (AD) and Down's syndrome neuronal degeneration can occur faster and be more extensive.
These findings suggest that reduced cortical–NB cholinergic transmission may contribute to symptoms involving memory loss (Whitehouse et al. 1981; DeKosky et al. 1992; Bartus 2000). However, the pathogenic mechanisms underlying NB degeneration and memory loss in aging, AD and Down's syndrome are still not fully understood. In some way AD and Down's syndrome are a form of ‘faster aging’ of the brain and it has been speculated that perhaps some mechanisms may be shared.
One possible shared mechanism may be deficiencies involving neurotrophin growth factors and their receptors, in particular nerve growth factor (NGF). NGF is responsible for the survival of cholinergic neurons, maintenance of the cholinergic phenotype and function of adult cholinergic neurons (Eide et al. 1993) and acute antagonism of NGF receptors in the adult rat results in cholinergic degeneration (Sofroniew et al. 1993; Debeir et al. 1999).
The NGF receptor that is clearly associated with neuronal survival is termed TrkA, a receptor tyrosine kinase (Kaplan and Miller 1997). In the CNS, TrkA is expressed almost exclusively by a subset of cholinergic cortical and NB neurons (Sobreviela et al. 1994; Chao et al. 1998; Sofroniew et al. 2001). NGF is secreted by target neurons in the cortex, binds TrkA at the nerve terminals and the activated TrkA receptors are retrogradely transported towards the cellular nucleus in the NB, during which time they signal through downstream effector pathways (Neet and Campenot 2001). The survival/growth signals arising from activated TrkA receptors at the nerve terminals or traveling along the axon differ from the survival/growth signals arising from activated TrkA that has reached the cell body (Saragovi et al. 1998; Zhang et al. 2000). In such a manner, NGF supports the maintenance and phenotype of a subset of adult cholinergic neurons (for reviews see Sofroniew et al. 2001; Lad et al. 2003).
Indeed, reduced expression or activity of TrkA receptors is documented in diverse neurocognitive pathologies, i.e. human mild cognitive impairment (MCI) (Mufson et al. 2000, 2002; Counts et al. 2004), AD (Mufson et al. 1989, 2000), human Down's syndrome (Sendera et al. 2000) and in a transgenic mouse model of Down's syndrome (Cooper et al. 2001). Phenotypic loss of TrkA in these pathologies is not due to extensive cellular loss because the number of NB perikarya expressing several other cholinergic markers is unchanged in MCI and mild AD compared with no cognitive impairment (NCI)[e.g. choline acetyl transferase (ChAT), vesicular acetylcholine transporter, the p75 neurotrophin receptor and even the expression of mature NGF] (Gilmor et al. 1999; Mufson et al. 2003; Counts et al. 2004).
These data are interpreted as suggesting that many cholinergic NB neurons undergo atrophy by phenotypic silencing of TrkA during the earliest stages of cognitive decline. In turn, this could lead to frank cell loss of the cholinergic subset dependent on trophic support mediated by TrkA and to advanced stages of disease. Together, the evidence described above supports the notion that disorders leading to memory loss may share deficiencies of TrkA as a mechanism. However, whether or not this is true for age-associated cognitive impairment is not known.
To address the question of whether reduced expression of TrkA protein correlates with the level of cognitive impairment in aging (as it does in other neurocognitive pathologies, i.e. MCI, AD and Down's syndrome), we performed quantitative and qualitative measurements of TrkA protein levels in cortical and NB regions of aged rats that had been well-characterized behaviorally as ‘unimpaired’ (UI), ‘mildly impaired’ (MI) or ‘fully impaired’ (FI).
The data demonstrate quantitative and qualitative changes in TrkA expression in brain. In MI rats there is phenotypic silencing of TrkA which reduces detection of cortical and NB neurons expressing TrkA at high density, preceding detectable cholinergic atrophy or cellular loss. In FI rats there is more extensive reduction in the detection of cortical and NB neurons expressing TrkA at high density, together with significant cholinergic neuronal cell loss. These data suggest that phenotypic loss of TrkA precedes neuronal loss and probably sensitizes cells to death. The rationale of using aged rats as a model for neurotrophic protection in MCI/AD and of using TrkA agonists to delay cholinergic neuronal loss is discussed.
Materials and methods
Aged (24-month-old) rats were characterized behaviorally in a Morris water maze as cognitively UI, MI or FI. This was done to avoid the potential confounding effects of studying a global and heterogeneous population of aged rats. All procedures were approved beforehand by the Animal Care Committee of McGill University and followed the guidelines of the Canadian Institutes of Health Research. Young adult (∼6-month-old) rats were used as controls.
Aged rats (n = 120) were required to find a submerged platform in a 1.4-m diameter pool of white, non-toxic colored water using only distal and spatial clues available in the testing room. Throughout, all tests were always carried out in the same room and set up. The center of the escape platform (15 cm diameter) was located 45 cm from the pool wall. Animals were tested in 15 trials over 5 consecutive days (three trials per day with an intertrial time of 20 min) with the platform 2 cm below the water. At the end of the testing periods, all animals were given three trials in which the platform was raised 2 cm above the water to exclude visual deficits as the cause of poor performance, as described by Rowe et al. (2003). To exclude motor deficits as the cause of poor performance the swim speeds and distances were recorded using a video-tracking system (HVS Image, Buckingham, UK) and to exclude thigmotaxis and amotivational status as the cause of poor performance concentric zones of the pool defined by the software program (HVS Image) were used to average swim patterns during days 2–5 of the acquisition phase (Graziano et al. 2003; Rowe et al. 2003). The latency to locate the escape platform was used to segregate rats into aged cognitively UI and aged cognitively impaired groups. The full behavioral data for the rats reported here were published in a previous study (Bruno et al. 2004).
Details for the rats used in this study have been published elsewhere (Bruno et al. 2004). Aged rats whose individual mean latencies to locate the platform were within 0.5 SD of the latency of young adult rats (n = 10, used as baseline) were considered cognitively UI (n = 14). Aged rats whose individual average latencies to locate the platform were >2 SDs from those of young adult rats on days 2–5 of the acquisition phase were considered cognitively FI (n = 31). Rats displaying an intermediate performance between UI and impaired (i.e. rats with individual average latencies <2 SD and >0.5 SD from those of young adult rats) were deemed to be cognitively MI (n = 23).
Tissue sections from frontal cortex and NB were obtained as described by Mufson et al. (1997) and quantitative western blotting was performed as described previously (Maliartchouk and Saragovi 1997; Mufson et al. 1997). Briefly, the tissues were sonicated in ice-cold homogenization buffer (20 mm Tris, 1 mm EGTA, 1 mm EDTA, 10% sucrose, pH 7.4) containing protease inhibitors [2 mg/mL leupeptin, 0.01 U/mL aprotinin, 1 mg/mL pepstatin A, 1 mg/mL antipain, 2.5 mg/mL chymostatin, 10 mm benzamidine, 0.1 mm phenylmethylsulfonyl fluoride, 0.4 mg/mL N-alpha-p-tosyl-L-lysine chloromethyl ketone (TPCK), 0.4 mg/mL N-tosyl-L-phenylalanyl chloromethyl ketone (TLCK), 0.4 mg/mL soybean trypsin inhibitor, 0.1 mm sodium fluoride, 0.1 mm sodium orthovanadate]. After further solubilization with non-ionic detergent (1% NP-40) fractions were prepared by centrifugation (14 000 g at 4°C) and the protein concentration determined (protein assay, Bio-Rad Laboratories, Hercules, CA, USA). Sample proteins (10–30 µg/lane) were separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) and transferred to polyvinylidene fluoride membranes (Immobilon P, Millipore Corporation, Bedford, MA, USA). Membranes were blocked in Tris-buffered saline/0.1% Tween-20/bovine serum albumin for 1 h at room temperature (23°C) followed by incubation with rabbit anti-TrkA RTA antiserum (1 : 2000) or mouse anti-β-actin (1 : 20 000; Chemicon, Temecula, CA, USA). After several rinses in Tris-buffered saline/0.1% Tween-20/bovine serum albumin the blots were incubated for 1 h at room temperature with horseradish peroxidase-conjugated secondary antibody (1 : 10 000; Sigma, St Louis, MO, USA or Bio-Rad Laboratories). Immunoreactive (IR) proteins were visualized by enhanced chemiluminescence (Amersham Biosciences, Piscataway, NJ, USA) and quantified by densitometry. TrkA bands were normalized to β-actin because it is unchanged in the aged brain (Hemby et al. 2003). Quantification was performed from four independent experiments, each with four independent gels, using three independent samples taken from NB or cortex from at least two rats in each group.
Rats were anesthetized and perfused transcardially with a mixture of 4% paraformaldehyde, 0.5% glutaraldehyde and 15% saturated picric acid in 0.1 m phosphate buffer, pH 7.4, for 30 min followed by the same fixative without glutaraldehyde for an additional 30 min. The tissue was subsequently post-fixed for 4 h in the latter fixative mixture and left at 4°C with a solution of 30% sucrose in 0.1 m phosphate-buffered saline. Sections (35 µm thick) between coordinates from bregma (0.5 and −3.0 mm, which comprise the parietal region of the cerebral cortex and the NB area) (Paxinos and Watson 1986) were then cut using a sledge microtome (Sliding Microtome, SM2000 R; Leica Microsystems, Quebec, Canada) equipped with a freezing stage. Before immunostaining, the free-floating sections were rinsed and incubated in phosphate buffer + 0.05% Triton X-100 for 30 min and then for 1 h in phosphate buffer + 0.05% Triton X-100 containing 1% bovine serum albumin. Serial sections were then immunostained with antibodies to ChAT or TrkA. All dilutions and washes were performed with phosphate buffer + 0.05% Triton X-100 containing 1% bovine serum albumin. For ChAT staining, sections were incubated overnight at 4°C with a sheep anti-ChAT antibody (1 : 4000) in the presence of donkey serum (1 : 500). After washing, the tissue was incubated with biotin-conjugated donkey anti-sheep antibody (1 : 500, Jackson Laboratories, Bar Harbor, ME, USA) followed by more washes and incubation with avidin conjugated with horseradish peroxidase (ABC kit, Vector Laboratories, Burlingame, CA, USA). For TrkA staining, sections were incubated overnight at 4°C with anti-TrkA rabbit antiserum RTA (a kind gift of L. Reichardt) in the presence of goat serum (1 : 500). After washing, the tissue was incubated with biotin-conjugated goat anti-rabbit antibody (1 : 500, Jackson Laboratories) followed by more washes and incubation with avidin conjugated with horseradish peroxidase (ABC kit, Vector Laboratories). In all cases, after additional washing the sections were incubated in 0.6% diaminobenzidine (Sigma) for 15 min at room temperature followed by 0.01% H2O2 as catalyst. After washing, the sections were mounted on gelatin-coated glass slides, dehydrated and coverslipped with Entellan (Merck, Darmstadt, Germany). Replacing the primary antibody with the same dilution of normal serum obtained from the same source served as a negative control.
For neurons in lamina V of the parietal cortex (bregma coordinates: lateral, 4.7–5.7 mm; ventral, 3.0–4.2 mm) we measured the relative density of ChAT-IR or TrkA-IR fibers. For neurons in the NB (lateral, 2.3–3.1 mm; anteroposterior, −1.2 to −1.8 mm; ventral, 6.3–7.3 mm) we measured the number of cell bodies, size of the somata and relative intensity of the ChAT-IR and TrkA-IR neurons. Data were analysed using a BH-2 Olympus microscope connected to an image analysis system (MCID Elite; Imaging Research, St. Catharines, Ontario, Canada) as described previously (Debeir et al. 1999). Briefly, four images per area, from both hemispheres, were taken from five different slices per rat (50 images per rat for each region, basalis and cortex). For lamina V of the parietal cortex and the NB, five animals were used in each group. For quantification of lamina V of the parietal cortex, 80 000 µm2 total area per rat were analysed. For the NB, a total of 50 000 µm2 were analysed per rat. Results are expressed per 1000 µm2. For analyses of cell size, a total of 10 000 ChAT-IR cells per rat were studied. Previous reports have demonstrated that vesicular acetylcholine transporter-IR boutons decrease in aged FI animals compared with aged UI animals (Bruno et al. 2004) and thus we do not show those data here.
Quantification of immunostaining intensity
Staining intensities were arbitrarily assigned scores as 4+ (very high), 3+ (high), 2+ (medium) and 1+ (low). The data are presented in two alternative forms. First, the percent of the total TrkA-IR cells counted for each group are scored as per their intensity. Second, the actual number of TrkA-IR cells counted for each group are scored as per their intensity. Two scorers (one blinded and the other not blinded to cognitive status) independently generated all of the counts. While the counts were very similar, only the data from the blinded scorer are presented in this study.
Statistical analyses were performed using commercial software (Systat 10.0; SPSS Inc., Chicago, IL, USA). Quantitative western blotting was analysed by anova with post-hoc testing. Immunohistochemistry data were subjected to univariate and multivariate anova. Between-group comparisons were made using Tukey's test. All probability values were two-tailed and a level of 0.05 was considered significant.
Behavioral characterization and segregation of aged rats
Aged rats were tested in the Morris water maze and mean latency times were obtained (Fig. 1). The aged rats were then segregated based on their mean latency time, as described in Materials and methods, as cognitively UI, MI and FI. On day 5 of the testing, the mean latency time for the young adult rats was 8.3 ± 1.2, while for the aged rat groups UI = 17 ± 2.1, MI = 40.7 ± 3.9 and FI = 92.9 ± 4.
Quantification of TrkA levels in nucleus basalis and cortex of aged unimpaired, mildly impaired and fully impaired rats
Nucleus basalis and cortical TrkA protein levels in aged rats were evaluated quantitatively using samples taken from UI, MI or FI rats (Fig. 2a). After normalizing the blots against protein-loaded and β-actin internal control, the mean ± SEM TrkA protein levels were calculated relative to the expression of Trk in the corresponding region of the young adult brain (Fig. 2b).
In NB and cortex quantitative western blots showed that TrkA protein in the aged UI rat was decreased by 10–15% compared with the young adult rat (p < 0.005) (quantification not shown). Amongst the aged rat groups, TrkA protein expression was highest in the UI group.
The UI group had significantly more TrkA protein in NB than the MI group (p < 0.001), with an overall reduction of ∼20% in the latter. The MI group had significantly more TrkA protein than the FI group (p < 0.001), with an overall reduction of ∼45% in the latter.
In cortex there were no significant differences between the UI and MI groups. However, both the UI and MI groups had significantly more TrkA protein than the FI group (p < 0.001), with an overall reduction of ∼30% in the latter.
Localization of TrkA expression in nucleus basalis
Fixed cryosections from the NB were immunostained for TrkA. Representative low-power pictures are shown in Fig. 3. Image analyses for all the pictures [out of >42 taken for each aged rat group (UI, MI and FI)] were used to count TrkA-IR somata in a total area of 50 000 µm2 NB per rat and the data were averaged per 1000 µm2 ± SD (Fig. 4a). The NB of the young adult rat had significantly more TrkA-IR cells than the NB of all the aged groups (p < 0.005). A comparison across the aged groups showed that, on average, the UI group had significantly more TrkA-IR cells than the MI and FI groups (p < 0.05). However, while the MI group had more TrkA-IR cells than the FI group, the difference was not statistically significant.
The trend in these data is consistent with the data from the quantitative western blots. Thus, the TrkA-IR cell counts were studied further based on the relative intensity of cellular immunostaining, using pictures from sections that were processed and immunostained simultaneously. Staining intensities were arbitrarily assigned scores of 4+ (very high), 3+ (high), 2+ (medium) and 1+ (low). The data are presented in two alternative forms. First, the percent of the total TrkA-IR cells counted for each group are scored as per their intensity (Fig. 4b) and second, the actual number of TrkA-IR cells counted for each group are scored as per their intensity (Fig. 4c).
A comparison of the young adult versus the aged UI rats showed that there were no differences in the percentage of TrkA-IR cells with the 4+ intensity. However, in the aged MI rat the 4+ intensity was statistically lower than that in aged UI rats. Moreover, in the aged FI rat the 4+ intensity was statistically lower than that in aged MI rats. The percentage of cells with the 1+ low intensity score increased significantly for the aged FI rats versus all other groups. There were no statistical differences in the percentage of TrkA-IR cells for the intermediate intensities 3+ and 2+ across all groups of rats (Fig. 4b).
A comparison of the actual number of TrkA-IR cells broken down by intensity shows that the main change is a significant decrease of high TrkA density-expressing neurons in the FI rat brain. For example, the young adult rat had an average of 43 TrkA-IR cells/1000 µm2 (Fig. 4a). Of those 43 somata, 19 scored 4+, nine scored 3+, seven scored 2+ and eight scored 1+ (Fig. 4c). In marked contrast, the aged FI rat had an average of 26 TrkA-IR cells/1000 µm2 (Fig. 4a) and, of those, five were 4+, five were 3+, six were 2+ and 10 were 1+ (Fig. 4c). For the aged MI rats there was also a decrease of high TrkA density-expressing neurons but it was not statistically different when compared with aged UI rats.
These data indicate that the aged rat had less TrkA-expressing neurons than the young adult rat and that within the aged rats there was a direct correlation between lower numbers of TrkA-expressing neurons and lower cognitive status in FI and MI. In addition, the phenotypic reduction in the number of TrkA+ neurons in aged FI and MI rats was mostly detectable as a reduction of cells expressing high density TrkA.
Expression of choline acetyl transferase in nucleus basalis
Sections from the NB were immunostained for ChAT. Representative low power pictures are shown in Fig. 5. Image analyses for all the pictures [>48 taken for each aged rat group (UI, MI and FI)] were used to count the average number of ChAT-IR cells per 1000 µm2 in aged rat UI, MI and FI groups. The UI and MI aged rat groups were not statistically different from each other. In contrast, the FI aged rat group had significantly fewer ChAT-IR neurons (p < 0.005) than the UI and MI aged rat groups (Fig. 6a).
The mean area of ChAT-IR neurons, representative of mean neuronal cell size, was calculated. The size of ChAT-IR cells was not statistically different for UI and MI aged rat groups. However, the ChAT-IR neurons in the FI aged rat group were of smaller size (p < 0.005) than the UI and MI aged rat groups (Fig. 6b).
These data suggest that there was a direct correlation between lower number of cholinergic neurons and lower cognitive status only for the aged rat FI group but not for the MI group. In addition, there was atrophy or shrinkage of cholinergic neurons in the FI group compared with MI and UI.
Our results confirm previous findings that the aged rats have less overall TrkA density than the young adult rats in the NB and cortex. Our novel findings are that the TrkA phenotype and relative TrkA density correspond to a defined cognitive state within the aged rat UI, MI and FI groups. The UI aged rat has a relatively high number of TrkA-expressing neurons, ∼50% of which express TrkA at high density. The MI aged rat also has a relatively high number of TrkA-expressing neurons, ∼35% of which express TrkA at high density. Lastly, the FI aged rat has a relatively low number of TrkA-expressing neurons (with few neurons expressing at high density and many expressing at very low density) and a significant loss of cholinergic cells.
It is important to note that the reduction in detection of high density TrkA+ neurons in MI aged rats occurs in the context of no loss of cholinergic cells or ChAT as a cholinergic marker, whereas in the FI aged rats there is loss of cholinergic cells and ChAT markers. Consistent with our data, previous reports (Bruno et al. 2004) demonstrated that the density of vesicular acetylcholine transporter-IR neurons is significantly reduced in aged FI rats compared with aged UI rats. The size of cholinergic neurons is also reduced significantly in aged FI rats compared with aged UI rats, whereas synaptophysin (as a measure of generic neuronal, not just cholinergic, connections) is not altered.
Thus, there is a direct correlation between lower numbers (or low density) of TrkA-expressing neurons and lower cognitive status. While these data do not necessarily indicate causality, the results suggest that changes in the TrkA phenotype in the NB and cortex precede cholinergic atrophy and lead to mild cognitive decline. We speculate that the neurons of cognitively MI rats are sensitized to subsequent death and they would eventually progress to losing TrkA+ cells/density more extensively, leading to detectable cholinergic atrophy and loss and full cognitive impairment.
Timeline of TrkA loss, atrophy and cellular death
Our data are helpful in developing a hypothetical mechanism to account for cholinergic cellular loss. First, the high TrkA-expressing cells shift phenotype progressively to medium and then to low density causing MCI. Thereafter, the low TrkA-expressing cells die causing full cognitive impairment. This would be consistent with the data obtained for the MI aged rat group, which has the same number of TrkA+ cells as the UI group but expresses lower TrkA density.
In the FI group, which has a lower number of TrkA+ cells than the UI group, only the high TrkA density cell bodies are ‘missing’. This could suggest that the high TrkA-expressing cells die outright but we favor the notion that, in the FI aged rat group, the high TrkA-expressing cells are shifted to the low TrkA-expressing population whereas the long-standing low TrkA-expressing population progressively atrophy and die.
Aging as a model of neurocognitive disorders
In normal aging, compared with the adult, cholinergic cells lose TrkA density, possibly at a variable rate. In FI aged rats this process is accelerated or more pronounced, whereas in MI aged rats this process is slower or less pronounced.
This process is reminiscent of data published for human post-mortem studies of Down's syndrome, MCI and AD brains versus ‘normal’ matched brains, where an almost linear relationship has been shown between TrkA expression and cognitive decline (Salehi et al. 1996; Mufson et al. 1997, 2000; Counts et al. 2004).
Hence, the cholinergic cells in these other cognitive pathologies have to contend with two simultaneous problems. One is reduced neurotrophic support affecting primarily the cholinergic system and the other is a more widespread damage involving other injuries, such as oxidative stress, amyloid deposition and axonal transport deficits, or other genetic, molecular or environmental factors.
TrkA as a cognitive marker
Are there any reliable biomarkers predictive of cognitive decline or recovery? Our data support the notion that TrkA may be a better marker of cognition than cholinergic markers.
The MI aged rat group is impaired cognitively compared with the UI aged rat group but a cholinergic profile of MI aged rats would not anticipate deficits. There are no significant changes in ChAT, vesicular acetylcholine transporter (data not shown, also see Bruno et al. 2004) or the overall cellular size in the MI versus the UI aged rat groups. In contrast, an analysis of TrkA expression and density seems to be more accurate at predicting cognitive decline.
As only a fraction of cholinergic cells express TrkA, the data suggest that it is the TrkA+ cholinergic cells that are involved in cognition, rather than all cholinergic cells. If this was confirmed, it might validate TrkA as a biomarker of the cognitive state and the availability of labeled small molecule ligands that image TrkA expression in vivo (LeSauteur et al. 1996; Bruno et al. 2004) may be useful as a rapid and objective diagnostic tool or as a predictor of treatment outcome in cognitive disorders.
E. Tsang and A. Caron provided technical assistance. I am grateful to Dr E. Mufson and Dr S. Counts (Rush University Medical Center) and Dr A.C. Cuello, Dr R. Quirion and M. Bruno (McGill University) for help with rat behavioral tasks, discussions and revisions. This work was supported by the Canadian Institutes of Health Research Grant MT-13265.