Introduction

The neurodegenerative processes characteristic of Alzheimer disease (AD) predate the development of clinical dementia by decades (Bateman et al. 2012; Katzman et al. 1988). These changes include neuronal loss and the development of amyloid plaques, neurofibrillary tangles and associated gliosis (Scheltens et al. 2016). Persons in this presymptomatic stage provide the opportunity to prevent the symptomatic disease and the development of therapeutics that target this early neurodegeneration requires biomarkers that accurately predict the development and progression of clinical dementia. A number of biomarkers have been proposed to behave in a predictable fashion in AD including beta-amyloid (Aβ) peptide and hyperphosphorylated tau protein levels in the cerebrospinal fluid (CSF), a specific pattern of hypometabolism on FDG-PET, identification of amyloid pathology using amyloid PET and hippocampal atrophy on structural MRI (Jack et al. 2010, 2011, 2013). These biomarker changes have also been identified in ADAD (Bateman et al. 2012; Ringman et al. 2008). These biomarkers, however, have limitations including accessibility in clinical practice, standardization, and the definition of cut-off thresholds for reliable identification of AD pathology.

1H magnetic resonance spectroscopy (MRS) provides a way to study brain biochemical processes in vivo. In patients with dementia due to AD, 1H MRS has consistently shown decreases in N-acetylaspartate (NAA), which is produced in neuronal mitochondria and found predominantly in neurons, as well as increased levels of myo-inositol (mI), a glial cell marker (Wang et al. 2015; Zhang et al. 2014). Alterations in the levels of these metabolites have been shown to correlate with the neuropathological findings in AD (Kejal Kantarci et al. 2008; Murray et al. 2014) and to also correlate with performance on tests of cognition (Frederick et al. 2004). Changes in NAA/Cr in the occipital cortex (Modrego et al. 2005, 2011), posteromedial parietal lobe (including the posterior cingulate gyrus and precuneus) (Modrego et al. 2011), and left paratrigonal white matter (Metastasio et al. 2006) as well as mI/Cr in the right parietal lobe (Targosz-gajniak et al. 2013) also appear to predict cognitive decline and conversion to dementia in patients with amnestic mild cognitive impairment (MCI). Results of 1H MRS studies of other metabolites including Cho (representing phosphatidyl choline and other membrane precursors, a marker of membrane turnover), Cr (representing creatine and phosphocreatine) and Glx (representing glutamate and glutamine) have been less consistent (Zhang et al. 2014). Prior 1H MRS studies in AD have been limited in the preclinical stage of the disease due to the difficulty of identifying appropriate subjects.

Autosomal dominant AD is caused by mutations in the PSEN1, APP, and PSEN2 genes with symptoms typically starting at a relatively young age. This population represents an opportunity to study metabolite levels in preclinical AD since carriers of the mutation develop AD with essentially 100% penetrance. A single study of 7 such ADAD mutation carriers (MCs) previously showed lower ratios of NAA/Cr and NAA/mI compared to non-carriers (NCs) in a single midline voxel containing the posterior cingulate (Godbolt et al. 2006). In the present study we attempted to identify if these changes are seen more broadly and at what stage in ADAD.

Methods

Subjects were participants in a comprehensive study of preclinical and manifest ADAD at a tertiary dementia referral center. The study was approved by the University of California, Los Angeles (UCLA) Human Subjects Committee and informed consent was obtained from all individual participants included in the study. All subjects were first-degree relatives of someone known to carry a pathogenic mutation in the PSEN1, APP, or PSEN2 genes, placing them at 50% risk for inheriting such a mutation and therefore developing ADAD. Genetic testing for the relevant mutation was performed as part of this study but subjects were not informed of the results. Revealing clinical genetic testing was offered outside of the research protocol for interested subjects. Extraction of DNA and genotyping of apolipoprotein E (APOE) were performed using standard techniques. APOE SNP genotyping was carried out by real-time PCR on an Applied Biosystems 7900HT Real Time PCR machine (Applied Biosystems, Foster City, CA), using Taqman SNP Genotyping Assays (#C3084793_20 and C904973_10 for rs429358 and rs7412, respectively). The SDS version 2.3 software was used to analyze the raw data and to call the genotype. The presence or absence of the specific mutation each subject was known to be at-risk for were assessed using standard Sanger sequencing, according to published protocols and primers.

The Clinical Dementia Rating Scale (CDR) was performed with an unrelated informant (Morris 1997). The CDR is a structured interview with input from both the subject and an informant who knows the subject well. In the CDR, asymptomatic persons are rated 0, and persons with questionable cognitive impairment are rated 0.5. Scores of 1, 2, and 3 represent mild, moderate, and severe stages of dementia, respectively. Subjects with clinical dementia (CDR of 1 or higher) were excluded from the analysis.

Each subject’s age in relation to their estimated age of dementia diagnosis was calculated. As the age of onset of symptoms is fairly consistent within a family and mutation but more variable between families (Ryman et al. 2014), an “adjusted age” can be calculated that estimates how many years from disease manifestation a given subject is (Ryman et al. 2014). In our experience, we have found that the age of clinical diagnosis of dementia is a more reproducible measure and therefore we calculated an adjusted age for each subject as his or her chronological age minus the median age of dementia diagnosis in his or her family.

Whole-brain structural MRI and water-suppressed 1H MRS were acquired at 1.5 T on a Siemens Sonata scanner. MRS was acquired in 2D magnetic resonance spectroscopic imaging (MRSI) configuration with a point-resolved spectroscopy sequence (PRESS) with repetition-time (TR) of 1500 ms, echo-time (TE) of 30 ms, and 8 excitations. Two 4 × 4 MRSI arrays or “slabs” (i.e., PRESS excitation volumes within 16 × 16 phase-encoding grids) were acquired, each with in-plane voxel resolution of 11 × 11 mm2. The first slab16 was 9 mm thick and axial-oblique, oriented parallel to the genu-splenium plane. It was positioned dorsoventrally symmetrically about the genu of the corpus callosum and left-right symmetrically astride the longitudinal fissure with the posterior slab edge abutting the genual rostrum. Posterior voxels within the slab sampled left and right pregenual anterior cingulate cortex (LpAC and RpAC, respectively) (Fig. 1). The second slab (O’Neill et al. 2015) was 12 mm thick and sagitally oriented. It was positioned left-right symmetrically astride the longitudinal fissure. The rear edge of the slab was aligned parallel to the parieto-occipital sulcus; the inferior edge was parallel or tangent to the dorsal margin of the isthmus or splenium of the corpus callosum. Thus, the slab sampled midline posterior cingulate (mPCG; inferior voxels) and midline precuneus (mPC; superior voxels) (Fig. 2). MR spectra were fit automatically with the LCModel commercial software package (Provencher 2001) yielding levels of NAA, Glx, Cr, Cho, and mI, referenced to unsuppressed water and expressed in institutional units (IU). Water-reference was preferred to referencing metabolite ratios to Cr due to certain weaknesses of the latter (Alger 2011). Cr measurement can be limited by overlap with neighboring signals (especially Cho). The Cr signal is a much smaller than the water signal and is thus more sensitive to noise. Also, accumulating evidence that Cr varies across the brain, with age, and with disease has gradually undermined earlier faith in Cr as a constant standard. After segregation of the whole brain MRI into gray matter, white matter, and CSF binary masks (Shattuck et al. 2001) the MRSI Voxel Picker (MVP) package (Seese et al. 2011) was used to compute the volume percent of gray matter, white matter, and CSF in each voxel and correct the LCModel-derived levels of each metabolite for voxel CSF content. Spectra not meeting quality control criteria (linewidth ≤0.1 ppm and signal-to-noise ratio ≥ 3) were rejected. Additionally, within spectra, individual metabolite peaks were rejected that were not considered reliable by LCModel (standard deviation of metabolite signal >20%).

Fig. 1
figure 1

Sagittal T1-weighted MRI of the human brain showing prescription of midline parietal 1H MRSI slab (yellow box). Usable spectra are obtained from the voxels (green squares) within the PRESS excitation volume (white box)

Full size image
Fig. 2
figure 2

T2-weighted axial-oblique (parallel to genu-splenium line) MRI of the human brain showing prescription of anterior cingulate 1H MRSI slab (yellow box). Usable spectra are obtained from the voxels (green squares) within the PRESS excitation volume (white box). Posterior mesial voxels within the slab sample left and right pregenual anterior cingulate cortex (LpAC, RpAC), anterior mesial voxels sample mesial superior frontal cortex, lateral voxels sample prefrontal white matter (anterior corona radiata)

Full size image

Characteristics of the study population were compared using two-sided Fisher exact tests for nominal variables and Student’s t-test for quantitative variables. For comparison between groups and regions of interest, a multivariate analysis of covariance (MANCOVA) with age as a covariate was performed for each region of interest across five metabolites (NAA, Glx, Cr, Cho, and mI). If the overall MANCOVA for a region showed a significant effect of diagnosis, protected post-hoc ANCOVA testing was performed for individual metabolites.

Within the MC group, we determined whether there was a correlation between metabolite levels and time until expected onset of dementia using a Spearman correlation. To correct for multiple comparisons a Holm-Bonferroni test was applied. A p value <0.05 was considered statistically significant. All statistical analyses were conducted in R version 3.3.1.

Results

The study population was comprised of 16 preclinical (CDR < 1) MCs and 11 NC controls with MRS data acquisition as part of the above study protocol. Data were missing from 3 MCs for the LpAC and RpAC, and 3 different MCs for mPCG and mPC; data were missing from 1 NC for LpAC, 2 NCs for RpAC, and 3 NCs for mPCG and mPC. These subjects were excluded from the analysis of the relevant voxels. Demographic information for the study participants is found in Table 1. The majority of MCs carried a mutation in PSEN1 (n = 13) while the remaining 3 had mutations in APP. Among MCs, 11 had CDR of 0 and 5 had CDR of 0.5. Among NCs, 2 subjects were blindly rated as having a CDR score of 0.5 but considering their ages (37 and 39), the etiology was considered unlikely to represent early AD. There was a significant difference in mean age between MCs (33, range 19–43) and NCs (40, range 28–59) (p = 0.03). Study participants were predominantly female (81 and 73% in MCs and NCs, respectively; not significant). Most participants were APOE ε3/ε3 or ε2/ε3 genotype, with 2 in each group being ε3/ε4 (13 and 18%, respectively; not significant). The mean time until expected onset of dementia in MCs was 12.3 years.

Table 1 Patient characteristics
Full size table

Results of MANCOVA analysis for the five metabolites of interest by voxel with age as a covariate showed a significant effect of mutation status in the LpAC (F = 3.067, p = 0.040) and mPC (F = 4.400, p = 0.013) (Tables 2 and 3). There was no effect of diagnosis in the RpAC or mPCG. Post-hoc protected ANCOVA with age as a covariate in the LpAC showed that MCs had lower NAA than NCs (7.61 vs 8.38, p = 0.04) and lower Glx (12.84 vs 13.55, p = 0.007). In the mPC, MCs had higher mI compared to NCs (5.3 vs 4.2, p = 0.0004), higher Cho (1.23 vs 1.20, p = 0.03) and lower NAA (8.86 vs 9.66, p = 0.04) (Fig. 3).

Table 2 Results of multivariate analysis of covariance by voxel
Full size table
Table 3 1H-MRS mean metabolite values by voxel in MCs and NCs
Full size table
Fig. 3
figure 3

Metabolite levels by carrier status and CDR for voxels/metabolite combinations for which there was a significant effect by carrier status. Horizontal bars represent mean metabolite values for the entire group (including individuals with both CDR 0 and 0.5)

Full size image

Metabolite levels for MCs and NCs are listed in Table 3; while we did not use Cr-normalized values for this analysis, we do report them for reference in Tables 4. Among MCs, the level of mI in the LpAC (ρ = −0.88, p = 0.00007) and mPC (ρ = −0.84, p = 0.0003) was higher in subjects with age closer to the estimated age of dementia diagnosis (Fig. 4, Table 5). Levels of Cr, Cho and mI all increased in the mPCG as onset of dementia approached, as did levels of Glx in the mPC however these findings did not remain significant after correcting for multiple comparisons. There were no other significant correlations between metabolite levels and time until expected onset of dementia in MCs (Table 5).

Table 4 Cr-normalized 1H-MRS mean metabolite values by voxel in MCs and NCs
Full size table
Fig. 4
figure 4

Correlation of PC and LpAC mI with years until expected diagnosis for MCs for asymptomatic (CDR = 0) and questionably symptomatic (CDR = 0.5) individuals

Full size image
Table 5 Correlation between metabolite levels and expected time to onset of AD
Full size table

Discussion

In this study we found that preclinical ADAD MCs had higher levels of mI in the mPC than NCs, and that levels of mI in the mPC as well as the LpAC increased in association with progressive neurodegeneration. mI levels have been consistently shown to be above-normal in 1H MRS studies of AD and elevated levels of mI appear to predate the development of dementia or alterations in other metabolites (K. Kantarci et al. 2000). The mean time until expected onset of dementia for MCs in this study was more than 12 years, confirming that mI is highly sensitive to early pathologic changes in AD.

mI is found primarily in glial cells, particularly astrocytes, where it appears to function as a cellular osmolyte and in cell signaling (Fisher et al. 2002). Elevated levels are thought to represent glial proliferation and/or greater cell size (Soares and Law 2009). In a postmortem study examining pathologic correlates of 1H MRS metabolites in AD, mI/Cr has been shown to correlate with Aβ burden (Murray et al. 2014) rather than gliosis or neurofibrillary tangles. Since early pathologic changes in AD are predominantly related to amyloid deposition, this supports the role of mI as an early marker of AD pathology.

Studies of late-onset AD which have consistently shown diffusely below-normal levels of NAA (K. Kantarci et al. 2000; Zhang et al. 2014), and reductions in NAA/Cr levels have been shown to correlate with reduced synaptic integrity and increased hyperphosphorylated tau pathology (Murray et al. 2014). The NAA peak reflects mainly N-acetylaspartate with a smaller contribution from N-acetylaspartylglutamate (NAAG). NAA is synthesized by neuronal mitochondria and is predominantly found in neurons, although it is eventually transported out and degraded on the oligodendrocyte cell membrane; the exact function is unknown although it is thought to possibly represent an osmolyte (Hajek and Dezortova 2008; Soares and Law 2009). NAAG is synthesized from NAA and glutamate in neurons and appears to be involved in signaling local energy balance (Baslow 2016). Depressed levels of NAA are therefore thought to reflect either neuronal loss, impaired neuronal functioning or energy metabolism. We found that levels of NAA in the mPC and LpAC were lower in MCs than in NCs. The Glx peak, which represents glutamate and glutamine, has been shown to be below-normal in the anterior cingulate in AD (Huang et al. 2016); we also found MCs to have lower levels of Glx in the LpAC compared to NCs.

MCs also had elevated mPC Cho compared to NCs. The Cho peak includes phosphocholine and other precursors needed for synthesis of membranes; increased levels are thought to reflect membrane disruption, cellular proliferation or inflammation (Hajek and Dezortova 2008). Choline is also a precursor of the neurotransmitter acetylcholine, the deficit of which is an important part of the pathophysiology of AD. In late-onset AD, the role of Cho has been less clear with some studies supporting elevated levels in AD and others finding no effect (K. Kantarci et al. 2000; Wang et al. 2015); in addition, studies of neuropathological correlates of 1H MRS metabolites in AD have not shown a clear correlation between Cho/Cr and AD pathology (Kejal Kantarci et al. 2008; Murray et al. 2014).

To our knowledge only one prior study has looked at 1H MRS metabolites in ADAD. This single-voxel study examined metabolic changes in the posterior cingulate and found that presymptomatic MCs had depressed NAA/Cr and NAA/mI in the region of interest. In the present study, the mean level of mPCG NAA was lower in MCs but the difference was not statistically significant. This discrepancy may be partially explained by a longer time until expected diagnosis in our subjects (12.3 years vs 9.8); as mentioned above, NAA has been shown to correlate with hyperphosphorylated tau pathology and reduced synaptic integrity, which are thought to occur later in AD than amyloid deposition (Jack et al. 2010); thus it is possible that the difference between MCs and NCs would continue to diverge with time until the difference reached statistical significance. Other factors that may have contributed to our not fully replicating the previous finding include technical factors and choice of voxel placement. We did find that both mI and Cho in this region increased as age of expected diagnosis approached; although this finding was not significant after correction for multiple comparisons, it does suggest that this area is affected relatively early in the disease as is known from studies of late-onset AD.

Our study was limited by the relatively small number of subjects enrolled (n = 27). This is a common problem in studies of ADAD due to its relative rarity as a cause of AD. However, despite the small number of subjects we were able to show clear differences between MCs and NCs. MCs were also younger on average than NCs, likely in part because we excluded persons with clinical dementia, and the typical age of onset of clinical dementia in ADAD is frequently in the 5th decade. However since there is a decrease in NAA and increase in mI in normal aging (Minati et al. 2007) the increased age in NC controls would have acted to decrease between group differences, and thus would not impact the validity of our results. Finally, two NCs were blindly rated 0.5 on the CDR; given their young ages (37 and 39), the etiology was considered unlikely to represent early AD and more likely due to heightened anxiety and potentially depression associated with living at-risk for ADAD. CDR rating is dependent on both the report of informants and subjects and anticipation of the onset of symptoms may have led them to emphasize subtle symptoms not related to incipient neurodegenerative disease.

In summary, levels of brain metabolites are altered in MCs for ADAD compared to NCs many years prior to expected onset of clinical dementia. mI is most sensitive as an early marker of AD pathology. For studies of disease-modifying interventions targeting prevention of neuronal death, MRS measures may provide an important surrogate outcome measure.