A new role for matrix metalloproteinase-3 in the NGF metabolic pathway: Proteolysis of mature NGF and sex-specific differences in the continuum of Alzheimer's pathology

Matrix metalloproteinase-3 (MMP-3) has been associated with risk of Alzheimer's disease (AD). In this study we introduce a novel role for MMP-3 in degrading nerve growth factor (NGF) in vivo and examine its mRNA and protein expression across the continuum of AD pathology. We provide evidence that MMP-3 participates in the degradation of mature NGF in vitro and in vivo and that it is secreted from the rat cerebral cortex in an activity-dependent manner. We show that cortical MMP-3 is upregulated in the McGill-R-Thy1-APP transgenic rat model of AD-like amyloidosis. A similar upregulation was found in AD and MCI brains as well as in cognitively normal individuals with elevated amyloid deposition. We also observed that frontal cortex MMP-3 protein levels are higher in males. MMP-3 protein correlated with more AD neuropathology, markers of NGF metabolism, and lower cognitive scores in males but not in females. These results suggest that MMP-3 upregulation in AD might contribute to NGF dysmetabolism, and therefore to cholinergic atrophy and cognitive deficits, in a sex-specific manner. MMP-3 should be further investigated as a biomarker candidate or as a therapeutic target in AD.


Introduction
Matrix metalloproteinase-3 (MMP-3) is a multifunctional brain protease with key roles in the developing and adult nervous system. Polymorphisms in the coding region of MMP-3 have been linked to increased risk of Alzheimer's disease (AD) (Helbecque et al., 2007;Saarela et al., 2004), while MMP-3 in cerebrospinal fluid (CSF) and in plasma has been demonstrated to be altered in AD and to correlate with cognitive impairment and core AD biomarkers (Hanzel et al., 2014;Horstmann et al., 2010;Iulita et al., 2019;Peng et al., 2015;Stomrud et al., 2010a). A recent longitudinal study demonstrated that MMP-3 levels are elevated early in AD plasma and confer risk of AD-related cognitive decline in females, but not males (Iulita et al., 2019).
derived supply of nerve growth factor (NGF) from target cells for the maintenance of their cholinergic function and phenotype (Cuello, 1993;Garofalo et al., 1992;Hefti, 1986). Mature NGF (mNGF) is produced from its precursor (proNGF) by the plasmin activating system and is subsequently degraded by matrix metallo-proteinase-9 (MMP-9) (Bruno and Cuello, 2006). The enzymes and zymogens involved in the NGF metabolic pathway are secreted at cortical/hippocampal cholinergic terminal regions in response to stimulation to produce a controlled trophic response (Bruno and Cuello, 2006).
Beginning early in the AD pathological process, the production of mNGF is insufficient and the degradation of mNGF is excessive (Bruno et al., 2009a;Bruno et al., 2009b;Pentz et al., 2020). This interruption of trophic support explains the progressive atrophy of the BFC projection system in AD (Pearson et al., 1983;Whitehouse et al., 1982), a neurotransmitter system that is critical for the maintenance of higher CNS functions such as learning, attention, and memory (Bartus et al., 1982;Drachman and Leavitt, 1974;Everitt and Robbins, 1997;Hasselmo and Sarter, 2011;Mesulam, 2004).
We were interested in the potential role of MMP-3 in mNGF proteolysis for the following reasons: MMP-3 has been observed to be secreted in response to cholinergic stimulation in peripheral tissues (De Couto et al., 2009;Reina et al., 2011), MMP-3 consensus target sequences overlap with MMP-9 consensus target sequences (Eckhard et al., 2016), MMP-3 can activate MMP-9, the main NGF-degrading protease (Ogata et al., 1992), MMP-3 is upregulated in AD plasma and CSF (Hanzel et al., 2014), and MMP-3 has also been reported to differentially modulate ADrelated cognitive decline by sex (Iulita et al., 2019). Therefore, we were interested in whether MMP-3 could play a direct or indirect role in mNGF proteolysis and, furthermore, whether its expression might be altered sex-specifically throughout the continuum of AD.
In this report, we have demonstrated for the first time that MMP-3 can contribute to mNGF degradation in vitro and in vivo and that it is released from cortical cells in response to cholinergic stimulation (or non-specific neuronal depolarization), like other NGF metabolic pathway proteins. We have demonstrated increased MMP-3 levels in males throughout the AD continuum, beginning prior to cognitive decline in human prefrontal cortex. A similar MMP-3 up-regulation was observed in a rat model of AD-like amyloidosis. Furthermore, we have found that human MMP-3 levels in AD consistently correlate to cognitive decline and measures of AD pathology in males, but not in females. The results of these studies indicate that MMP-3 should be considered as a novel member of the NGF metabolic pathway, which is altered in AD in a sex-specific manner.

Methods
All procedures were approved by the Animal Care Committee of McGill University, in accordance with the guidelines of the Canadian Council on Animal Care. Efforts were made to minimize the number and suffering of the animals used. McGill-R-Thy1-APP transgenic rats express the Swedish double and Indiana genetic mutations in the human amyloid precursor protein gene, hAβPP, leading to progressive amyloidosis and cognitive deficits (Iulita et al., 2014a;Leon et al., 2010). Animals were housed with controlled humidity and temperature under 12-h light cycle and had access to food and water ad libitum.

In vitro NGF degradation assay
Ten or twenty ng of 14 kDa recombinant human mNGF produced in E. coli (N-245; Alomone labs, Jerusalem, Israel) was diluted in 50 μl PBS and incubated with 0, 3.7, 7.4, or 14.8 mM of active recombinant human MMP-3 (ab96555; abcam, Cambridge, UK). Solutions were incubated overnight at 37 • C with agitation, then mixed with denaturing loading buffer 1:4 for Western blotting, the protocols for which are described later in the methods.

In vivo NGF degradation assays
Anaesthesia was induced in six 8-month-old wild-type Wistar rats (3 males, 3 females) with 5% isofluorane in oxygen in an induction chamber and maintained with 2% isofluorane in oxygen through a face mask. Rats were stereotaxically injected with 2.0 μl of 1.0 μg/μl human recombinant MMP-3 (ab96555; Abcam, Cambridge, UK) in saline or with 2.0 μl of 0.5 μM of the MMP-3 inhibitor, UK356618 (5.9 nM IC 50 , ~150× selectivity vs. other MMPs; Santa Cruz, CA, USA) in DMSO, with vehicle injected contralaterally. The side receiving the vehicle was randomized for each injection. Injections were performed with a 5.0 μl Hamilton microsyringe fitted with a 32 gauge needle (Hamilton, USA) in the dorsomedial hippocampus according to the Paxinos and Watson rat brain atlas (Paxinos and Watson, 2006) at the following coordinates: − 3.6 antero-posterior, +/− 2.0 mediolateral (+/− giving left/right, which was randomized to contralateral/ipsilateral), and − 2.8 dorsoventral from Bregma.
After 12-16 h, rats were deeply and irreversibly anaesthetized by 100 mg/100 g equithesin administered intraperitoneally and perfused transcardially with cold perfusion buffer, following (Côté et al., 1993) for composition. Each dorsal hippocampus was macrodissected and divided into anterior, medial, and posterior segments. Each segment was then homogenized both manually and through sonication in RIPA buffer with CompleteMini protease inhibitors (Sigma-Aldrich, St. Louis, USA), then prepared for Western blotting as described below.

Western blotting
Standard reducing loading buffer was added to samples at a ratio of 1:4. Each sample (8 μl of 3.0 μg/μl rat/human cortex samples, 12 μl in vitro assay samples, and 20 μl of superfusate fractions) was added to an SDS-polyacrylamide gel (10-15%) and run for 90 min at 120 V. The protein contents of each gel were transferred to nitrocellulose membranes (GE Healthcare, Chicago, USA) at 0.3 A for 80 min. Membranes were blocked for one hour with a solution of 5% milk in TBST, then incubated overnight in appropriate antibody solutions: ab52915 anti-MMP-3 (Abcam, Cambridge, UK) was used to reveal MMP-3 immunoreactivity at 45 kDa and proMMP-3 at 55 kDa, ANT-005 (Alomone labs, Jerusalem, Israel) was used to reveal 27 kDa (unmodified) proNGF, H-20 anti-NGF (Santa-Cruz, Dallas, Texas) was used to reveal all NGF immunoreactivity in in-vitro studies, and the NGF 9795 (Abcam, Cambridge, UK) was used to reveal mature NGF immunoreactivity at 14 kDa in human/rat tissue, following the protocol of Locke et al. (Locke, 2020) as adapted by (Pentz et al., 2020). In the latter case, a chloroform/ methanol protein extraction was required to detect mNGF at the low abundance at which it exists in the human cortex, following recently published protocols (Pentz et al., 2020). The identity of the quantified band at 45 kDa was established by comparison to a positive control (recombinant MMP-3 as above or lung tissue). Peroxidase-conjugated secondary antibodies (Jackson Immunoresearch, West Grove, USA) were subsequently added at 1:10000 in TBST for one hour at room temperature. The membranes were detected using Western Lightning Electrochemiluminscence Solution (PerkinElmer Inc., Waltham, USA) and exposed to Highblot chemiluminescence-detecting films (Denville, Canada). These films were developed using a KODAK X-OMAT 2000A (Eastman Kodak Company, Rochester, USA), and quantified using CLIQS software (Totallab, Newcastle, UK). The resulting integrated optical density (IOD) value for each sample was normalized to the IOD of GAPDH (AB2303; Millipore, Billerica, USA) on the same membrane and to an internal control to ensure consistency between membranes; each sample was measured in triplicate. IODs are represented as a fold change from the average of the control group. Average GAPDH IODs did not differ between groups.

Ex vivo superfusion-stimulation of rat cerebral cortex tissue
Superfusion experiments were done following the protocol of (Bruno and Cuello, 2006) to determine whether MMP-3 is released in response to cholinergic stimulation, like NGF metabolic pathway proteins. Threemonth-old type Wistar adult rats (n = 4; 2 male, 2 female) were euthanized by rapid decapitation and their brains were rapidly removed and macrodissected. The cerebral cortices were diced with a McIlwain sectioning instrument and were placed on a semi-permeable 8.0 μm pore membrane (Corning Life Sciences, New York, United States), submerged in Hanks buffer +5% O 2 and 1% CO 2, and superfused at a rate of 0.25 ml/min. The temperature was maintained at 37 • C ±, 2. After a 45-min adaptation period, a baseline sample was taken. A 100 nM solution of the cholinergic agonist carbachol (or 0.05 M KCl) was then added to the preparation and allowed to incubate (pumps off) for 1 min. Further samples were taken at 5, 10, 15, and 30 min after the initial aliquot was drawn. The preparation was then allowed to re-equilibrate for 15 min, and the procedure was repeated. To assess the intracellular versus extracellular Ca 2+ -dependency of activity-dependent release of MMP-3, incubations were made in the presence or absence of 10 μM of either the cell-permeant calcium chelator BAPTA-AM or cell-impermeant BAPTA (Sigma-Aldrich, St. Louis, Missouri, USA), added at the beginning of the adaptation period.
McGill-R-Thy1-APP rats express human Amyloid Precursor Protein (APP) with the double Swedish and the Indiana mutations under the Thy1 promoter, and recapitulate key features of amyloid pathology in AD (Leon et al., 2010). Amyloid-beta accumulates early in hippocampal and cortical neurons (Iulita et al., 2014b;Leon et al., 2010). As the amyloid pathology progresses with age, extracellular plaques start to develop progressively. The first plaques are observed in the subiculum at 6-9 months of age in homozygotes (Leon et al., 2010) and at 10-12 months of age in heterozygotes, and later spread through anatomically connected regions (Heggland et al., 2015). Cognitive deficits are apparent at 3 months of age in homozygotes and at 6 months of age in heterozygotes, and worsen as the pathology progresses (Iulita et al., 2014b;Leon et al., 2010).

Neurochemical analysis of human frontal cortex samples
We acquired dorsolateral/medial prefrontal cortex tissue (Brodmann's areas 9 and 46; areas previously shown to demonstrate significant proNGF accumulation/mNGF depletion (Iulita et al., 2014c;Pentz et al., 2020)) from 33 aged individuals with no cognitive impairment (NCI) and minimal cortical Aβ deposition (2 times the Standard Error of the Mean (SEM) under the mean of the MCI/AD groups; referred as Low-Amyloid-NCI; LA-NCI), 21 individuals with NCI but AD/MCI-like Aβ deposition (within 2 SEM of the mean of the MCI/AD groups; referred as High-Amyloid-No Cognitive Impairment; HA-NCI), 19 persons with a diagnosis of Mild Cognitive Impairment (MCI), and 24 patients with a clinical diagnosis of Alzheimer's dementia from the ROS (Rush Alzheimer's Disease Center, Chicago, IL). All participants enroll without known dementia and agree to annual detailed clinical evaluation and brain donation. The study was approved by an Institutional Review Board of Rush University Medical Center. All participants signed an informed consent, Anatomic Gift Act, and a repository consent to allow their biospecimens and data to be repurposed. ROS resources can be requested at https://www.radc.rush.edu. Diagnoses of Alzheimer's dementia were made according to the National Institute of Neurologic and Communicative Disorders and Stroke/AD and Related Disorders Association (NINCDS/ADRDA) criteria (McKhann et al., 1984). As part of the Religious Orders Study (ROS) (Bennett et al., 2012), these individuals had undergone extensive cognitive testing including the Modified Mental State Exam (MMSE) and a series of 21 cognitive tests (episodic memory, semantic memory, working memory, perceptual orientation, and processing speed) recorded before death, expressed as z-scores and summarized as the Global Cognitive Score (GCS).
The ROS also entails a uniform post-mortem evaluation of brain's neuropathology. A measure of Aβ deposition was taken by the average percent area of 6E10 Aβ immunoreactivity in the CA1, subiculum, angular gyrus, entorhinal cortex, superior frontal cortex, dorsolateral prefrontal cortex, inferior temporal cortex, anterior cingulate cortex and calcarine cortex by image analysis (Bennett et al., 2006). A measure of tau tangle pathology was calculated by the stereological assessment of AT8 immunoreactivity, averaged across the entorhinal cortex, CA1, superior frontal cortex, mid frontal cortex, inferior temporal cortex, angular gyrus, cingulate gyrus, and calcarine cortex (Bennett et al., 2006). Patients were diagnosed with MCI if determined by a blinded neuropsychologist to have a cognitive impairment while deemed by a clinician to not meet criteria for dementia. The demographic characteristics of the study groups are illustrated in Table 1.

qPCR for MMP-3 and TIMP1 in human and rat brain tissue
RNA was isolated from cortex tissue samples matching those described above using the RNeasy RNA Isolation Kit (QIAGEN, Hilen, Germany). The integrity of the mRNA samples was analyzed using the NANODROP to ensure no contamination by DNA or protein. The samples were then reverse transcribed to rtDNA using iScript Reverse Transcription Supermix (Bio-Rad, Hercules, California) in a T3 Thermocycler (Biometra, Gottingen, Germany). Samples were combined with forward and reverse primers for MMP-3 (human forward: 5 ′ -GTAGAAGGCACAATATGG-3 ′ , human reverse: 5 ′ -ACTCTATGTGA-CAAGGTG-3 ′ ; rat forward: 5 ′ -GTGCTCATCCTACCCATTGCAT-3 ′ , rat reverse: 5 ′ -TTTGACAACAGGGCTACTGTCCTT-3 ′ ; Integrated DNA Technologies, Coralville, USA), as well as SsoAdvanced CyberGreen Supermix (Bio-Rad). The samples were then analyzed in triplicate using CFX Connect Real Time PCR Machine (Bio-Rad, Hercules, USA), and the average 2 -ΔΔCq values of each group were compared. The Cq values for each sample were normalized by the Cq value for GAPDH (5 ′ -CCCCACTTGATTTTGGAGGGA-3 ′ forward, 5 ′ -AGGGCTGCTTT-TAACTCTGGT-3 ′ reverse) in each sample. Results are presented as fold change from the mean of the WT group. Average GAPDH Cq values did not differ significantly between groups.

Statistics
D'Agostino-Pearson Omnibus Normality Tests were used to determine the normality of groups, and group means were compared with one-way ANOVAs and Bonferroni post-hoc tests or Kruskal-Wallis tests and Dunn's post-hoc corrections. A two-way ANOVA with Bonferroni post-hoc tests was used to assess main effects of sex and interactions of sex and diagnosis. Differences in demographic characteristics were examined using ANOVA or Chi-square tests as appropriate (see Table 1). GraphPad Prism 5 was used for all statistics described above (GRAPH-PAD Software, San Diego, California, USA). All bar graphs indicate the mean ± Standard Error of the Mean (SEM).
Correlations were examined in R (R Project for Statistical Computing, Vienna, Austria) using multiple linear regression, with age and PMI included as cofactors and partial r 2 and p values reported independently within each sex. Illustrative scatterplots were prepared with GraphPad PRISM 5. All values were normalized to the mean of the LA-NCI group and represented as a fold-change from that mean.

MMP-3 as a mature NGF-degrading protease both in vitro and in vivo
Past work demonstrating that MMP-9 and MMP-3 target sequences overlap (Eckhard et al., 2016), that MMP-3 can be released in response to cholinergic stimulation (De Couto et al., 2009;Reina et al., 2011), and that MMP-3 was elevated in AD CSF where it correlated to cognitive decline (Hanzel et al., 2014) led us to consider the possibility that MMP-3 could be an mNGF-degrading protease like MMP-9 in the context of the proposed NGF metabolic pathway (Bruno and Cuello, 2006). To test this possibility, we co-incubated increasing doses of MMP-3 with different mNGF concentrations in vitro. The addition of recombinant MMP-3 overnight at 37C led to a dose-dependent decrease in mNGF immunoreactivity (n = 4; Fig. 1A). Indeed, 14.8 mM MMP-3 was enough to almost completely abrogate mNGF immunoreactivity in that time (n = 4, Fig. 1B).
Given these results, we next investigated if such mNGF proteolytic activity is observed in the mature CNS in vivo. For this, we resorted to inject stereotaxically in the medial dorsal hippocampus, in separate experimental animals, either MMP-3 or UK356618, a specific and wellvalidated inhibitor of MMP-3 (Artham et al., 2019), with the vehicle solution injected on the contralateral side in each condition. After 12-16 h we trisected each hippocampus for comparisons of injected and non-injected segments (see Fig. 2A for a schematic of injections).
Interestingly, MMP-3 injected into the medial dorsal hippocampus led to a near-total depletion of mNGF immunoreactivity in this segment in comparison both to the non-injected anterior and posterior segments, as well as each contralateral segment (One-way ANOVA, overall F = Data is presented as mean ± standard deviation. LA-NCI: low amyloid-no cognitive impairment, HA-NCI: high amyloid-no cognitive impairment, MCI: mild cognitive impairment, AD: Alzheimer's disease, NFT: neurofibrillary tangles, MMP-3: Matrix Metalloproteinase-3. The "p value" column gives the result of a one-way ANOVA or Chi-square test comparing the means of all four pathological group while the "Association" gives the results of regression/correlation between MMP-3 protein levels and demographic/pathological characteristics; the APOE4 association is the exception, showing the results of a two-way ANOVA between diagnosis and e4 carrier status in all samples. No such effect was observed for the injection of DMSO on the contralateral side (post-hoc tests for contralateral medial vs contralateral anterior/posterior, all t < 0.80, all p > 0.05; Fig. 2C). In contrast, no effect of the application MMP-3 or UK355618 was observed on hippocampal proNGF levels (Supplementary Fig. 1A and B respectively).

MMP-3 is released from the rat cerebral cortex cells in response to cholinergic stimulation
The NGF metabolic pathway is co-released from cholinergic target neurons in response to cholinergic stimulation, in a manner dependent on intracellular (but not extracellular) calcium (Bruno and Cuello, 2006). Having shown that MMP-3 has a role in mNGF degradation, we proceeded to investigate whether MMP-3 might be released in this same way, compatible with co-release as a part of the NGF metabolic pathway. Following (Bruno and Cuello, 2006), diced rat cortical tissue was superfused and stimulated with 100 nM carbachol or 50 mM KCl. Under both conditions, an activity-dependent release of MMP-3 was observed that peaked after 5 min of carbachol stimulation and returned to baseline by 30 min in superfusate samples (Fig. 3A-B; average values of three replicates). The non-specific depolarizing agent KCl (50 mM) acted equivalently to carbachol (Supplementary Fig. 2A and B), indicating that cholinergic stimulation suffices to elicit the release of MMP-3. The release of MMP-3 protein, as it is the case for other members of the NGF metabolic pathway (Bruno and Cuello, 2006), was shown to depend on intracellular but not extracellular calcium, as the cell permanent calcium chelator BAPTA-AM, but not the cell-impermeant BAPTA, abrogated MMP-3 release in response to carbachol (Fig. 3A vs. 3B) or KCl ( Supplementary Fig. 1A vs. 1B).

MMP-3 levels are increased in a transgenic rat model of Alzheimer'slike amyloid pathology
To examine whether MMP-3 levels are progressively altered in the presence of AD pathology, we assessed the transcription of MMP-3 mRNA at three time-points throughout the lifetime of the McGill-R-Thy1-APP transgenic rat model of Alzheimer's-like amyloid pathology.

Alterations of MMP-3 levels in the human cerebral cortex through the continuum of AD pathology
In order to understand how MMP-3 levels might change in the context of progressive AD pathology, we assessed the expression and levels of MMP-3 and proMMP-3 by qPCR and Western blot in dorsomedial prefrontal cortex samples from individuals with no cognitive impairment and minimal levels of amyloid deposition (LA-NCI; n = 33), in individuals with no cognitive impairment and MCI/AD-like high amyloid deposition (HA-NCI; n = 21), in Mild Cognitive Impairment (MCI; n = 19) and in Alzheimer's disease (AD; n = 24), cases. Demographic characteristics of the study cohort are displayed in Table 1; no association was found between MMP-3 levels and age, years of education, post-mortem interval, or APOE4 frequency.
We found that MMP-3 mRNA was elevated in HA-NCI, MCI, and AD cases as compared to LA-NCI individuals

Sex-specific differences in MMP-3 protein levels
Since we had previously observed higher expression of MMP-3 protein in the plasma of males compared to females (Iulita et al., 2019;Iulita, 2016), we investigated potential sex differences in CNS MMP-3 protein levels. A two-way, sex by diagnosis ANOVA demonstrated main effects (MEs) of diagnosis and sex, and while males expressed more MMP-3 in all groups, post-hoc tests revealed significant differences only in the LA-NCI group (Two-way ANOVA, F DIAGNOSIS = 25.74, p < 0.001, F SEX = 15. 19, p < 0.001, F INTERACTION = 0.04, p > 0.05; Bonferroni post hoc tests: sex within LA-NCI t = 2.72, p < 0.05; sex within HA-NCI t = 1.99, p > 0.05, sex within MCI t = 1.45, p > 0.05, sex within HA-NCI t =  MMP-3 mRNA was upregulated in homozygous transgenics compared to heterozygotes and wild-types, which were not different one from another. B/C: At both the six and the 13-15 month time-points, MMP-3 mRNA in both the heterozygous and homozygous transgenics were increased relative to the wild-types. D: Confirming that these transcriptional differences have relevance for protein levels, we showed that MMP-3 protein was also elevated in heterozygous and homozygous transgenics relative to the wild-types at the 13-15 months timepoint. E: Illustrative Western blot for MMP-3 in wildtype, hemizygous, and homozygous McGill-R-Thy1-APP rats, with GAPDH for reference. For all figures, stars indicate the significance of Bonferonni post-hoc tests vs. APP − /− in a 4 × 1 ANOVA: *p < 0.05, **p < 0.01, ***p < 0.001 Data represents Cq or Integrated Optic Densities, with bars representing mean values and error bars representing the SEM. MMP-3 = Matrix Metalloproteinase-3. 1.54, p > 0.05; Fig. 5D).

Sex-specific relationship between MMP-3 protein expression and AD neuropathology or cognitive scores
We were then interested in understanding whether MMP-3 might be associated with cognitive and neuropathological changes associated with AD. Considering only the persons with no overt cognitive impairment (NCI), we correlated levels of MMP-3 protein withtwo cognitive measures: the Mini-Mental State Exam (MMSE), a commonly used dementia screen, and the Global Cognitive Score (GCS), a multi-domain composite of cognitive tests used to broadly assess cognitive function. Additionally, we correlated MMP-3 protein levels to measures with Aß deposition and tau tangle density over multiple cortical regions. We found that higher MMP-3 protein was associated with lower scores on both the MMSE and the CGS, as well as higher burden of Aß and more tau tangles (Table 2).
Since we observed a sexually dimorphic expression of MMP-3 most pronounced at stages preceding cognitive impairment in this and previous studies (Iulita et al., 2019;Iulita, 2016) we were interested in whether sex would influence the associations between MMP-3 protein and the cognitive and pathological changes associated with AD at preclinical stages. Intriguingly, we observed significant associations between increased MMP-3 protein levels and lower scores on both the MMSE and GCS in males with NCI (n = 27) that were not found in females (n = 27; Table 2; Supplementary Fig. 3A and B). Similarly, MMP-3 protein correlated with Aß burden and tau tangle density in NCI males, while in females, only an association with Aß burden was observed (Table 2; Supplementary Fig. 3C and D).

Sex-specific relationship between MMP-3 protein expression, proNGF and TIMP1
Given that we propose a role for MMP-3 in the metabolism of mNGF, we were interested to observe whether MMP-3 protein levels would correlate to levels of mNGF, the NGF precursor proNGF, and the CNS

Table 2
Overall and Sex-stratified associations between MMP-3 and cognitive scores, neuropathological scores, and measures of NGF dysmetabolism in all cognitively normal aged individuals (NCI).

Covariate
Overall ( inhibitor of MMP-3, TIMP1 (given by mRNA expression levels). Interestingly, males exhibited a moderate negative correlation between MMP-3 protein levels and mNGF protein levels (Table 2; Supplementary  Fig. 4A), a strong positive correlation between MMP-3 levels and proNGF (Table 2; Supplementary Fig. 4B), and a weak correlation between TIMP1 mRNA levels and MMP-3 (Table 2; Supplementary Fig. 4C) in the dorsomedial prefrontal cortex, while none of these were observed in females (Table 2; Supplementary Fig. 4A-C). As before, while all these correlations were evident in the overall population (Table 2), they appear to have been driven by the males in the population.

Discussion
In this study, we demonstrate for the first time a role for MMP-3 in the proteolysis of mNGF and, furthermore, its involvement in the NGF dysmetabolism unleashed by AD pathology (Fig. 6), predominantly by amyloid accumulation. This proposition is supported by the fact that MMP-3 protein, its precursor, and mRNA transcripts were found elevated over the course of human AD, reaching near-maximum levels prior to the onset of dementia, in parallel with accumulating AD pathology. We also demonstrate that the expression of MMP-3 in the human frontal cortex exhibits dimorphism by sex, with males having consistently more MMP-3 immunoreactivity. Additionally, MMP-3 appears to correlate to measures of pathology in males in a way that is generally not apparent in females, though the underlying mechanism is unclear.

MMP-3 and AD: genetics and neuropathology
The involvement of MMP-3 in AD pathogenesis or symptomology is supported by limited genetic evidence. An interaction between APOE ε4 status and MMP-3-1171-5A/6A allele status (associated with promoter activity) that influences risk of AD was found in three cohorts (Helbecque et al., 2007;Saarela et al., 2004), though it was absent in a fourth (Reitz et al., 2008); an underlying mechanism was never elucidated, though several studies suggested that they might synergistically contribute to risk of vascular dysfunction and indeed, an interactions between MMP-3 and APOE ε4 status were found to contribute to pathology only in patients with heart disease in one study (Helbecque et al., 2007). The link between MMP-3 genetic variants and AD risk Fig. 6. Schematic of the NGF Metabolic Pathway in health in in the context of AD neuropathology. A: Under normal conditions, mNGF is derived from proNGF by plasmin (3), itself derived from plasminogen by tissue plasminogen activator (tPA) (2). tPA activity is controlled by neuroserpin, its endogenous inhibitor (1). MMP-9 and, we now show, MMP-3 degrade mNGF (7); these are regulated by the tissue inhibitor of metalloproteinases 1 (TIMP1) (8). Under normal conditions sufficient mNGF dimerizes and binds p75/TrkA receptor complexes (4). The activated receptors are internalized and retrogradely transported to the cell body where they ultimately exert their trophic effect (6). B: In the context of AD pathology, increased neuroserpin/decreased tPA (1) lead to reductions in local plasmin concentrations, reducing the maturation of proNGF to mNGF (2). Decreased TIMP1 (3) and increased MMP-3/MMP-9 lead to the excessive degradation of mNGF (4). As a result, trophic support to BFCNs is compromised (5). mNGF = mature Nerve Growth Factor, proNGF = precursor to NGF, tPA = tissue plasminogen activator, BFCN = basal forebrain cholinergic neuron, MMP-3/9 = matrix metalloproteinase 3/9, TIMP1 = tissue inhibitor of metalloproteinases 1.
would suggest a mechanistic role of this protease in AD pathology/ symptomology, which may in part explain the findings of our study.
While this is, to our knowledge, the first study to investigate brain MMP-3 levels in MCI and in cognitively normal people with AD-like neuropathology, it has been previously demonstrated that MMP-3 protein levels are increased in AD, particularly in white matter and in proximity to amyloid plaques (Yoshiyama et al., 2000); though this study did not assess differences by sex. That observation is in line with our finding of correlations between amyloid deposition and MMP-3 levels in cognitively normal aged persons as well as with previous studies showing elevated MMP-3 secretion in mixed hippocampal cultures after treatment with Aß peptides (Deb and Gottschall, 1996), suggesting a direct link between amyloid deposition and MMP-3 accumulation. While a role of MMP-3 in degrading Aß peptides has been proposed (White et al., 2006) it is unlikely that metallo-proteases have a significant effect in diminishing the amyloid pathology given that most studies have shown amyloid pathology and metalloproteases levels correlate positively, if at all (Baig et al., 2008).
While Aß was shown to induce MMP-3 expression, previous studies demonstrated that MMP-3 could in turn reduce tau oligomer aggregation by cleaving tau in core repeat sequences (Nübling et al., 2012). In the present investigation, we observed significant associations between MMP-3 and tau deposition in males, but not in females, providing partial support for this hypothesis, though it is still unclear how these mechanisms could be modulated by sex.
Aß pathology and increased MMP-3 levels are further linked by the present demonstration that MMP-3 transcription is upregulated in the McGill-R-Thy1-APP homozygotic transgenic rats modelling the AD-like amyloid pathology at 3 months and in heterozygotes at 6 months. It is worth noting that at 3 months of age, McGill-R-Thy1-APP homozygotes first display cognitive deficits (Leon et al., 2010) and start revealing proNGF accumulation (Iulita et al., 2017). Furthermore, no difference in MMP-3 levels were observed between McGill-R-Thy1-APP homozygotes and heterozygotes at the 13-15-month timepoint, when homozygotes display extensive plaque pathology and heterozygotes have relatively sparse plaques. This, as well as the observation of increased MMP-3 in heterozygotes at 6 months (prior to any plaques), suggests that the MMP-3 upregulation might be primarily driven by soluble Aß and therefore explain why MMP-3 and plaque deposition only correlated in the early disease state.

MMP-3 and AD: the bigger picture
In this study, we highlighted a novel role for MMP-3 as a mNGF protease, participating in the regulation of trophic support to BFCNs. However, other important roles have been proposed for MMP-3 in AD. MMP-3 is a critical mediator of inflammation, where it can serve to activate microglia (Kim et al., 2007;Kim et al., 2005;Lee et al., 2010;Walker and Rosenberg, 2009), facilitate peripheral immune cell invasion (Gurney et al., 2006), and cleave inactive precursors to produce active pro-inflammatory cytokines (Schönbeck et al., 1998;Woo et al., 2008); since MMP-3 is itself inducible by cytokine signalling (Borghaei et al., 2004;Sinfield et al., 2013;Tamai et al., 1995;Wang et al., 2014) and free oxygen radicals (Liu and Rosenberg, 2005), it may participate in a pathological positive feedback loop.

MMP-3 as a candidate AD biomarker
Previous studies have implicated MMP-3 as a potential early biomarker of AD. Several studies have shown that MMP-3 is increased in plasma (Horstmann et al., 2010;Iulita et al., 2019;Peng et al., 2015) and CSF (Hanzel et al., 2014;Horstmann et al., 2010) from patients with symptomatic AD, as well as in people with Down syndrome who manifest AD (Iulita, 2016). CSF MMP-3 has also been shown to be associated with core biomarkers of AD, showing positive correlations with CSF total-tau and phospho-tau (Hanzel et al., 2014;Stomrud et al., 2010b), and curiously, positive correlations with Aß42 were found in one study (Mlekusch and Humpel, 2009). MMP-3 activity in plasma (Peng et al., 2015) and levels in CSF (Hanzel et al., 2014) have been demonstrated to negatively correlate with MMSE scores. The current study, showing negative correlations between MMSE/GCS performance and brain MMP-3 levels demonstrates that in the previous studies, biofluid MMP-3 levels are reflecting CNS MMP-3.
While biofluid MMP-3 has yet to be systematically analyzed at the early stages of sporadic AD, plasma levels of MMP-3 were found elevated in patients with subjective memory impairments and mild cognitive impairment as compared to healthy controls (Iulita et al., 2019). Furthermore, healthy aged individuals with AD risk factors revealed higher levels of CSF MMP-3 than healthy aged individuals without AD risk factors (Stomrud et al., 2010b). In Down syndrome, a conditioned characterized by a lifelong and inexorably progressive amyloidosis that culminates in clinical AD (Lott and Head, 2019), plasma MMP-3 is elevated and increasing even prior to the onset of overt dementia (Iulita, 2016). An exploratory study also suggested an Aß-associated effect of MMP-3 on brain atrophy in cognitively normal aged individuals (Mattsson et al., 2014). Our present report provides new grounds supporting the above clinical observations, as we demonstrated an increase of MMP-3, associations with cognitive ability and associations with Aß deposition prior to the onset of cognitive symptoms in AD. Considering the newly described role of MMP-3 in mNGF proteolysis, MMP-3 could perhaps serve as part of a panel that could rarify AD diagnosis by adding information about NGF dysmetabolism and cholinergic decline, and could perhaps enhance precision medicine by providing sex-specific indications with regards to pathology and/or cognition.

A link between MMP-3 and NGF metabolism
The discovery of the NGF metabolic pathway, describing the maturation and subsequent degradation of NGF in the CNS (Bruno and Cuello, 2006), permitted a better understanding of trophic support to basal forebrain cholinergic neurons in health and disease (Bruno and Cuello, 2006;Cuello et al., 2010;Cuello et al., 2019;Iulita and Cuello, 2014). As these neurons are fully dependent on endogenous mNGF for the maintenance of their cholinergic phenotype and functions (Cuello, 1993;Debeir et al., 1998), we entertain the hypothesis that NGF dysmetabolism in AD best explains both cholinergic atrophy as well as the brain's build-up of proNGF together with the depletion of mNGF observable in this condition, even prior to the onset of overt dementia (Bruno et al., 2009b;Pentz et al., 2020). This neurotrophic failure results from impaired conversion of proNGF into mNGF and the exacerbated degradation of already diminished mNGF (see Fig. 6 for schematic overview).
The NGF metabolic pathway has been proposed as a source of new biomarkers and therapeutic targets for the cholinergic deficit in AD (Counts et al., 2016;Cuello et al., 2019;Iulita et al., 2019). In this report, we offer a fuller understanding of NGF dysmetabolism by demonstrating the in vitro and in vivo ability of an additional metallo-proteinase, MMP-3, to degrade mNGF. While it is difficult to estimate the exact correspondence of our experiments to physiological levels of MMP-3/ mNGF, it is likely that the excess of MMP-3 we used was appropriate, as mNGF is present at vanishingly low concentrations in the human brain (Fahnestock et al., 2001). The absence of proNGF cleavage is likely a product of the change in shape induced by the presence of the prodomain which may shield the vulnerable moiety (Trabjerg et al., 2017;Yan et al., 2019). We further demonstrated that MMP-3 is secreted in an activity-dependent manner, in response to nonspecific and cholinergic stimulation in a manner dependant on intracellular calcium, similarly to other members of the NGF metabolic pathway. We further validate the relevance of this finding by demonstrating that MMP-3 elevation occurs early in the pathogenesis of AD. These findings are in line with the thorough investigation of other molecular members of the NGF metabolic pathways across the continuum of AD (Pentz et al., 2020). Importantly, this report provides evidence that higher MMP-3 correlates to lower mNGF and higher proNGF in cognitively normal aged adults.

Sex differences in MMP-3 expression in AD
In this study we report that MMP-3 levels are lower in females, and that correlations between MMP-3 and the preclinical neuropathological and cognitive symptoms of AD are most-sometimes only-observable in males. Accordingly, previous studies have indicated higher biofluid levels of MMP-3 in males (Almodóvar et al., 2014;Arends et al., 2011;Iulita et al., 2019;Iulita, 2016;Yamanaka et al., 2000) and have shown differential response of MMP-3 to disease between sexes (Arends et al., 2011;Ribbens et al., 2002), supporting a differential biological regulation of MMP-3 in males and females. Interestingly, a previous retrospective cohort study had demonstrated that lower MMP-3 levels correlated with poorer performance on the Montreal Cognitive Assessment and Delated Recall tests in females with AD and MCI but not in males (Iulita et al., 2019). While a model integrating both of these studies-better correlations between brain MMP-3 and pathology in males and better correlations between plasma MMP-3 and cognitive decline in females-remains to be elucidated in the context of increased female risk of AD, it is possible that MMP-3 is responding to pathology in males in a way that is not paralleled in females; this could serve a protective function with regards to amyloidosis that unfortunately exacerbates NGF dysmetabolism. The lack of a correlation between MMP-3 and mNGF in females may also indicate that MMP-9 plays a more important role in females than in males in terms of NGF metabolism. This may also illuminate certain differences in cholinergic dysfunction and treatment in AD, with some data indicating a better response to cholinesterase inhibitors in males (Giacobini and Pepeu, 2018). These results also highlight the critical importance of considering the role of sex in AD pathology, and particularly the risks of generalizing findings in males, who have historically been disproportionately investigated, to the entire human population-as recently highlighted by the Women's Brain Project (Ferretti et al., 2018;Ferretti et al., 2020).

Conclusion
These studies would indicate that, independently of previously established roles, MMP-3 additionally plays a physiological role in regulating the availability of trophic support to BFCNs by participating in the degradation of mature NGF. We also demonstrate that MMP-3 is upregulated in the dorsomedial prefrontal cortex throughout preclinical and clinical stages of AD. We further show that these levels are higher in males, and that MMP-3 levels in cognitively normal aged males (but not females) consistently correlate to cognitive decline, AD neuropathology, and NGF dysmetabolism. These results support future preclinical studies investigating MMP-3 as an early biomarker and as a potential therapeutic target in AD with adequate consideration of sex as a factor.

Authors' contributions
RP, MFI and ACC conceived and designed the study. RP performed all experiments with experimental support from MM-C, save the in vitro degradation assays, which were performed by AD. MFI, AD, and ACC provided input into experimental and analytic methods. DAB provided brain tissue as well as the corresponding neuropathological and cognitive data. RP analyzed the data and prepared the figures. The manuscript was written by RP and ACC with input from all other authors. All authors have read, revised, and approved the submitted manuscript.

Declaration of Competing Interest
The authors have no conflict of interest to report.