There were significant cognitive group-specific differences in MMSE (NC, CI > AD), global ARCS (NC > CI > AD), and scaled memory domain scores (NC > CI, AD). Mean baseline UPSIT scores (UP_20) did not differ between NC and CI, but both were significantly greater than that for the AD group (Table ​(Table1).1). By contrast, mean AE differed between NC and both CI and AD groups, but there were no group-specific differences in AE between CI and AD. Similar findings were present for left hippocampal volume, adjusted for ICV (LHCV), which was greater in the NC than in either CI or AD groups.

AE ranged from +5 to −6. Figure ​Figure11 depicts the distribution of AE by cognitive category. AE < 0 was present in 31% NC, 92% with CI and 86% with AD. Figure ​Figure22 shows the scatter plot of AE against scaled memory score. Figure ​Figure33 shows the scatter plot of AE and LHCV.

The univariate associations between AE, baseline UPSIT_20, LHCV, APOE genotype and scaled cognitive domain scores are depicted in Table ​Table2.2. Correlations within the entire sample and in the non-demented sub-sample showed highly significant associations between AE, scaled memory domain score and LHCV. Baseline UPSIT_20 score was significantly associated with memory, but not with either LHCV or APOE genotype. Within the dementia sample, LHCV was significantly associated with overall cognition, memory, and visuospatial function.

We conducted a series of linear regression analyses within the non-demented sample with memory domain score as the outcome. (See Table ​Table3).3). In hierarchical regression analyses, age, gender and education were entered together, then baseline UPSIT_20 score and then either LHCV (Model 1) or AE (Model 2) separately. AE explained more variance (24%) in memory scores than did LHCV (15%). Next, we entered the same 4 initial variables, then AE followed by LHCV (Model 3). In this model, LHCV explained only 2% additional variance in memory, after accounting for that explained by AE. By contrast, in Model 4 in which AE was entered after LHCV, AE accounted for an additional 15% in the variance: equal to that explained by LHCV in the same model. Analytic models run without baseline UPSIT_20 score gave essentially the same results. Finally, in a linear regression model predicting memory domain score with age, gender, education, LHCV and AE entered simultaneously, AE (t = 3.3, p = 0.003) but not LHCV (t = 1.2, p = 0.24) retained significance.

In previous, larger studies of elderly individuals the results of conventional olfactory testing (i.e. comparable to our baseline testing during the OST) have been shown to correlate only modestly with cognitive performance and hippocampal volume. Among 1092 non-demented elderly participants in a recent community-based study [7], UPSIT scores correlated with delayed recall (r = 0.28) and, in a subsample of 571, hippocampal volume (r = 0.16), comparable to the estimates we obtained between baseline UPSIT and memory (r = 0.11) and hippocampal volume (r = 0.12) among non-demented individuals in the current study but substantially less than the correlations between AE and memory (r = 0.52) and hippocampal volume (r = 0.49) within the same sample. Other studies have examined the value of conventional olfactory testing for predicting subsequent cognitive decline [5,6,28,29]. Of these, several have demonstrated an interaction between olfaction and APOE genotype status indicating that the predictive value of olfactory testing may be restricted largely to individuals who are APOE ϵ4 positive [6,28]. Presumably, this reflects the greater probability that any olfactory decline is due to underlying AD (rather than to other non-specific local nasal pathology) in those at increased genetic risk for this condition, relative to the situation in APOE ϵ4 negative individuals. By contrast, correlations between AE of the OST and cognitive measures and hippocampal volume in the current study were at least as strong within the APOE ϵ4-negative sample relative to ϵ4-positive individuals. Together, the above results suggest that the OST (AE) may be more sensitive and specific for underlying AD pathology than is conventional olfactory testing.

Clinicopathologic studies of non-demented individuals have shown associations between antemortem cognition, particularly episodic memory, and the ‘burden’ of AD pathology at autopsy [1,30]. In one study, the strongest association was between limbic ‘diffuse senile plaques’ and logical memory (r = −0.58) [1]. Mortimer et al. [30] found associations between neurofibrillary counts and Braak stage (as indices of AD pathology) [2] and delayed memory one year prior to death. When hippocampal volume was added to the linear regression models, it alone remained a significant predictor of memory, suggesting that the effect of the neuropathology was mediated through hippocampal atrophy [30]. Similar conclusions have been reached from studies using Pittsburgh Compound B (PiB) Positron Emission Tomography (PET) imaging to detect insoluble amyloid deposits in the brain. In one such study, the PiB index of amyloid deposition within a combined sample of normal controls and PiB + MCI subjects was significantly associated with both episodic memory and hippocampal volume (HCV) [31]. However, when PiB index and HCV were both entered into a regression model predicting episodic memory, only HCV was significant. Our findings stand in direct contrast to these results. AE but not LHCV remained as the significant predictor of episodic memory in analyses, similar to those described above, in which both variables were included. This suggests that AE is a proxy for a process that subsumes hippocampal atrophy in the evolution of AD. Biological plausibility for the relevance of AE with respect to AD pathology is further supported by the strength of the relationship between AE < 0 and the APOE genotype. The odds ratios we obtained were very similar to the increased risk of AD due to APOE ϵ4 that has been estimated from clinical and pathological studies [32,33].

Structural and functional mechanisms warrant consideration in relation to the findings we report. The OB receives massive cholinergic input from the basal forebrain [34]. Neuropathological studies have consistently demonstrated a profound loss of cholinergic neurons in the Nucleus of Meynert (Ch4) in the presence of other pathological features of AD [35,36]. Similar changes in the adjacent Ch3 nucleus, which provides the rich cholinergic innervation to OB, have also been documented [10,37]. Damage to cholinergic structures occurs early in the evolution of AD, although how early, and to what extent, remains contentious [35]. Physiologically relevant concentrations of β-amyloid specifically interfere with cholinergic neurons and neurotransmission [36]. In neuronal culture/in vitro studies, β-amyloid has been shown to reduce high affinity choline uptake, decrease the rate of acetylcholine (ACh) synthesis, inhibit ACh release, and impair muscarinic receptor activation of G proteins [8,36]. Elegant studies by Bales and colleagues produced very direct and compelling evidence for the negative impact of β-amyloid on cholinergic function [38]. These workers measured ACh release within the hippocampus using an in vivo microdialysis technique in awake, moving mice. Measurements were made in PDAPP transgenic (a well-characterized model of AD in which β-amyloid is over-expressed) and wild type (WT) mice. Relative to WT, PDAPP mice had significantly lower basal production of ACh. When the experimental animals were injected with the pan-muscarinic receptor antagonist scopolamine, WT mice showed a seven-fold increase over basal rate in hippocampal ACh production, but the response in transgenic animals was very significantly blunted. Finally, when PDAPP mice were pre-treated with a monoclonal anti β‐amyloid antibody (m266), the deficient, blunted ACh response to scopolamine was normalised [38]. The results implicate soluble β-amyloid in the differential response of cholinergic neurons to an anticholinergic. Soluble β-amyloid within olfactory structures, perhaps the OB specifically, might account via similar mechanisms for the results we have obtained in the current study. Specifically, post-synaptic cholinergic blockade might be overcome by enhanced ACh release in normals but the absence of such a response, related to amyloid, could explain a reduction in olfactory performance from variable degrees of transient cholinergic transmission failure. Noteworthy in this context, Wesson et al. have recently shown in the Tg2576 (APP over-expressing) mouse model that soluble amyloid appears earlier in the OB than in any other brain region, and is associated with olfactory deficits [9]. The soluble form of β-amyloid, including oligomers of dimers and trimers, rather than the insoluble fibrillary form, such as is detected by PiB imaging, may be most toxic [39,40], is present before the development of amyloid plaques [41], and has been shown to correlate better with cognition than does fibrillary amyloid [42].

There are both uncertainties and important limitations in relation to our study and it would be premature and inappropriate at this point to apply the OST in a clinical context to aid in diagnosis or prognostication. Based on its modest molecular weight, lipophilic properties, and the fact that it is known to cross the blood brain barrier, we hypothesized that atropine sulphate would concentrate in the OB when delivered intranasally, as other small drugs have been shown to do [15]. However, we have no direct evidence either from the literature or from our own study to support that. It is possible, for example, that the atropine effect we have observed reflects the consequences of systemically-absorbed atropine operating more diffusely on cholinergic pathways within the olfactory sytem. We did not administer a control, pharmacologically inactive, nasal spray to any of our participants and it could be questioned whether our results could have arisen due to some non-specific effect of a nasal instillation. In the absence of data, we cannot absolutely refute such a criticism. However, based on the coherence of the findings, all predicted a priori, of associations between atropine effect and each of memory, hippocampal volume and APOE genotype, it seems implausible that these could be accounted for by some chance effect not under pinned by relevant biology. In future studies, we will additionally administer a non-active spray to a subset of study participants to address this concern directly. The current sample was relatively small. Inter- and intra-rater reliability of hippocampal volume estimation was lower than has sometimes been reported in the literature [7] and it is possible that measurement error may have diminished the apparent association between hippocampal volume and other measures. In particular, this may have inflated the contribution of AE, relative to hippocampal volume, in our regression analyses predicting memory score. However, the correlations we obtained between hippocampal volume and baseline UPSIT score, and between hippocampal volume and memory score, were very similar to those obtained in other mixed samples of cognitively normal and impaired individuals [7,43]. The current study does not speak to the specificity of the OST in terms of aetiology because we deliberately excluded individuals in whom conditions other than AD might have contributed to cognitive decline. We are undertaking further studies with larger samples, more diagnostic heterogeneity and longitudinal followup to better characterize the properties of the OST. The results of such studies would need to be carefully evaluated before consideration could be given to the use of the OST in clinical practice.