Peripheral immune-derived matrix metalloproteinase promotes stress susceptibility

Summary paragraph Psychosocial stress has profound effects on the body, including the peripheral immune system and the brain1,2. Although a large number of pre-clinical and clinical studies have linked peripheral immune system alterations to stress-related disorders such as major depressive disorder (MDD)3,4,5, the underlying mechanisms are not well understood. Here we show that a peripheral myeloid cell-specific proteinase, matrix metalloproteinase 8 (MMP8), is elevated in serum of subjects with MDD as well as in stress-susceptible (SUS) mice following chronic social defeat stress (CSDS). In mice, we show that this increase leads to alterations in extracellular space and neurophysiological changes in the nucleus accumbens (NAc), thereby altering social behaviour. Using a combination of mass cytometry and single-cell RNA-sequencing, we performed high-dimensional phenotyping of immune cells in circulation and brain and demonstrate that peripheral monocytes are strongly affected by stress. Both peripheral and brain-infiltrating monocytes of SUS mice showed increased Mmp8 expression following CSDS. We further demonstrate that peripheral MMP8 directly infiltrates the NAc parenchyma to control the ultrastructure of the extracellular space. Depleting MMP8 prevented stress-induced social avoidance behaviour and alterations in NAc neurophysiology and extracellular space. Collectively, these data establish a novel mechanism by which peripheral immune factors can affect central nervous system function and behaviour in the context of stress. Targeting specific peripheral immune cell-derived matrix metalloproteinases could constitute novel therapeutic targets for stress-related neuropsychiatric disorders.


Main text
Stress-related neuropsychiatric disorders such as major depressive disorder (MDD) have a high worldwide 51 prevalence and tremendous individual burden 6 . While there are many effective treatments for MDD, more 52 than a third of affected individuals do not achieve full remission by available antidepressant medications or 53 established psychotherapeutic treatments 7,8 . One of the most important risk factors for depression is chronic 54 psychosocial stress 9 . Therefore, elucidating the pathophysiological mechanisms underlying the effects of 55 psychosocial stress is crucial to advancing our understanding of disorders like MDD and ultimately develop 56 treatment options and prevention strategies. 57 Immune interactions between the central nervous system and peripheral organ systems are tightly 58 regulated 10 . Psychosocial stress can profoundly impact this bi-directional communication, and disrupted 59 neuroimmune interactions are increasingly recognized as important factors in the pathogenesis of stress 60 disorders 11 . Chronic stress activates the innate immune system resulting in mobilization of peripheral 61 myeloid cells (e.g., monocytes and neutrophils) and the production of pro-inflammatory cytokines, such as 62 interleukin-6 (IL-6) 12,13,14 . In humans, it is well established that a subset of patients with stress-related 63 neuropsychiatric disorders, such as MDD, display a state of chronic low-grade inflammation, characterized 64 by increased circulating pro-inflammatory cytokines and leukocytosis 4,14,15 . In addition to these peripheral 65 immune changes, stress disrupts the endothelial blood-brain barrier (BBB) in mice allowing greater entry 66 of circulating proteins directly into brain reward regions like the nucleus accumbens (NAc) 16,17 . While these 67 findings have provided important insights into the pathophysiology of stress and depression, we still know 68 relatively little about the mechanisms by which these stress-induced immune changes affect neuronal 69 function and ultimately behaviour. 70 In the brain, neurons and non-neuronal cells are separated by the extracellular space which contains 71 interstitial fluid and the extracellular matrix (ECM), a dense scaffold of proteins and glycans secreted by 72 neurons and glial cells 18,19 . ECM molecules have been shown to play an important role in homeostatic 73 processes of the brain, including synaptic function 20 . ECM degradation and remodeling is regulated by 74 various enzymes, such as Matrix metalloproteinases (MMPs) 19,21 . MMPs in circulation have been associated 75 with numerous inflammatory processes and disorders, such as cancer and myocardial infarction 22,23 . While Chronic stress and pro-inflammatory monocytes 81 To investigate the effects of psychosocial stress on the immune system and its impact on the brain, we used 82 the chronic social defeat stress (CSDS) paradigm 27,28 . Interpersonal conflicts and social bullying are 83 commonly experienced major psychological stressors that can precipitate a major depressive episode 29,30 . 84 The CSDS paradigm -one of the best validated mouse models of psychosocial stress -consists of the  Table   96 2). 97 To investigate if these pre-clinical mouse findings also translate to human stress disorders, we assessed 98 leukocyte subpopulations in blood from patients with MDD and healthy controls (HC) and found that 99 patients with MDD displayed leukocytosis driven by increased numbers of monocytes and neutrophils (Figs.  Table 3). We also observed a significant positive 101 correlation between the number of monocytes and neutrophils in circulation with perceived stress, using the 102 Perceived Stress Scale 33 , a clinically validated self-report measure of stress (Figs. 1e, g). 103 In whole brains of mice following CSDS, we observed a specific increase of infiltrating pro-inflammatory 104 Ly6C high monocytes in SUS, but not RES, mice compared to CON mice (Figs. 1h, i, Extended Data Figs. 105 1f-h). To prevent contamination with circulating leukocytes, brains were thoroughly perfused with PBS. 106 Notably, we did not observe differences in other infiltrating leukocytes or brain-resident immune cells, such 107 as microglia or border-associated macrophages (Extended Data Fig. 1i). 108 To investigate differences in stress-induced transcriptional changes in the major circulating leukocyte 109 subpopulations from CON, SUS and RES mice, we performed cell type-specific RNA-sequencing of 110 Ly6C high and Ly6C low monocytes, B cells and T cells (Fig. 1j, Extended Data Figs. 3a-d).  changes in gene expression were most pronounced in Ly6C high monocytes, with a total of 785 differentially 112 expressed genes in SUS vs. CON mice and 311 genes in RES vs. CON mice (adjusted p-value < 0.05 and 113 log2 fold change > |1|) (Figs. 1k, l, Extended Data Fig. 3e), with approximately 10 times fewer differentially 114 expressed genes in the other cell types (Extended Data Figs. 3f-q). We then performed gene ontology (GO) 115 enrichment analysis of biological processes, cellular components and molecular function. Genes 116 upregulated in SUS vs. CON mice were involved in GO biological processes such as innate immune 117 response (GO:0045087) and inflammatory response (GO:0006954) and cellular components like 118 extracellular space (GO:0005615) (Fig. 1m). Taken together, CSDS increased monocyte numbers in 119 circulation and in the brain and induced a pro-inflammatory transcriptional signature in SUS mice. These 120 findings, which emphasize the role of peripheral myeloid cells in stress-linked disorders such as MDD, are 121 in line with several pre-clinical studies 13,34,12,35 and studies in humans 36,37,38 . Therefore, we focused on 122 elucidating the mechanisms by which peripheral monocytes can affect neuronal function and behaviour.

124
Brain-infiltrating monocytes express increased matrix metalloproteinase 8 (Mmp8) 125 First, we sought to investigate the exact locations where infiltrating inflammatory Ly6C high monocytes 126 traffic in the brain. We performed detailed anatomic mapping of brain-infiltrating monocytes with the whole 127 brain tissue clearing method iDISCO+ 39 . Using the Ccr2 rfp reporter line, where monocytes express a red 128 fluorescent protein, we cleared brains from CON, SUS and RES mice, then performed lightsheet microscopy 129 and registered the samples to the Allen Brain Atlas using ClearMap 39 (Figs. 2a,  4a-e). We first analysed total infiltrating cells in whole brains and confirmed our CyTOF data showing 131 increased monocytes only in SUS mice (Figs. 1i and 2c). Next, we examined the correlation between 132 infiltrating monocytes and social avoidance behaviour. Cell counts in striatal regions such as the NAc, were 133 highly correlated with the social interaction (SI) ratio of stressed mice, with higher numbers of NAc-134 infiltrating monocytes correlating with greater social avoidance behaviour (Fig. 2d). The NAc is a stress-135 responsive brain region central to processing rewarding and aversive stimuli 40 , and is critical in mediating 136 depression symptomatology 41,42 . Confocal microscopy revealed that brain-infiltrating monocytes were 137 attached to the vasculature in NAc but did not infiltrate the brain parenchyma (Fig. 2e). 138 Next, we investigated how brain-infiltrating monocytes contribute to stress-induced social avoidance and 139 performed single-cell RNA-sequencing of brain-infiltrating monocytes after CSDS (Fig. 2a, Extended   140 Data Figs. 5a-e). Unsupervised clustering of Ccr2 rfp+ monocytes revealed four unique clusters based on 141 their transcriptional profiles (Fig. 2f). We found cluster 0 to be enriched in SUS mice relative to CON and 142 RES mice (Fig. 2g). To determine cluster-defining genes, we performed a differential gene expression 143 analysis by investigating differentially expressed genes between clusters and total genes (adjusted p-value 144 < 0.05) (Fig. 2h). Several genes known to be involved in inflammatory processes were found to be 145 upregulated in Cluster 0: e.g., genes coding for S100 proteins, including S100A6 and S100A11 43,44 , Lyz1 146 coding for lysozyme, an antimicrobial protein critical for host defense 45 , and the annexins Anxa1 and 147 Anxa2 46 . GO term analysis of upregulated genes of cluster 0 revealed involvement in oxidation-reduction 148 process (GO:0055114), extracellular space (GO:00056150) and extracellular matrix (GO:003102) (Fig. 2i). 149 One of the top genes enriched in these extracellular space/matrix pathways was Mmp8, which codes for 150 matrix metalloproteinase 8 (Figs. 2j, k). Of note, Mmp8 was also one of the top differentially expressed 151 genes in circulating in Ly6C high monocytes from SUS vs. CON mice, as was the GO term extracellular space 152 (Figs. 1l, m). We also performed single-cell RNA-sequencing of brain-resident immune cells in NAc 153 (Extended Data Figs. 6a-e), however, we did not observe any stress-induced changes in homeostatic or signature in microglia from other brain regions 47,48,49,50 , this is the first study investigating cell type-specific 157 gene expression signatures at the single-cell level in NAc microglia. This finding suggests that there are 158 brain-region-specific differences in microglial reactivity to stress, which is in line with recent studies 159 showing substantial heterogeneity in microglia across brain regions 51 . Our results are also consistent with 160 recent single-cell studies conducted in postmortem brain samples from human subjects with MDD, where 161 no evidence of pro-inflammatory microglia signatures was found 52,53 .

162
MMP8 belongs to the group of collagenases and is derived and secreted largely from neutrophils and 163 monocytes 21,54,55 . Our data and published studies suggest that, unlike many other MMPs, MMP8 is not 164 produced or secreted by any cells of the central nervous system, including brain-resident myeloid cells 54 165 (Fig. 2l). Interestingly, in one of the largest whole blood gene expression studies to date, MMP8 was one of 166 the top upregulated genes in patients with MDD compared to HC 56 . In addition, Song et al. reported a single-167 nucleotide polymorphism in the coding region of MMP8 to be associated with MDD 57 . However, the 168 underlying mechanisms linking MMP8 with MDD have not been explored. Activated MMP8 can cleave a 169 wide range of ECM components, such as collagens, fibronectins, tenascins and aggrecan, many of which 170 are components of the brain ECM 20,54 .

172
Peripheral MMP8 is associated with brain extracellular space abnormalities 173 To test whether monocyte-derived MMP8 can promote stress susceptibility, we first confirmed the stress-174 induced increase in MMP8 at the protein level in plasma after CSDS (Fig. 3a, Extended Data Figs. 7a, b), 175 and showed that MMP8 levels were negatively correlated with SI ratio (Fig. 3b). We further demonstrated 176 that both 10 days of CSDS and 21 days of chronic variable stress increased plasma levels of MMP8 in 177 female mice (Extended Data Figs. 7c-f). Next, we measured plasma levels of other MMPs such as MMP2, 178 3, 9 and 12 in the same mice as shown in Fig. 3a after 10 days of CSDS. While we did observe a modest 179 increase of MMP3, we observed similar changes in both SUS and RES mice compared to CON mice,180 confirming that only MMP8 is uniquely upregulated in SUS mice but not CON or RES mice (Extended 181 Data Figs. 7g-j). Finally, we validated the increased MMP8 in serum from patients with MDD compared 182 to HC (Fig. 3c, Supplementary Table 4) and found a positive correlation with self-reported perceived 183 stress (Fig. 3d). We then confirmed that MMP8 was increased in the NAc of SUS mice following CSDS 184 (Fig. 3e, Extended Data Figs. 8a, b). By retro-orbitally injecting biotinylated mouse recombinant MMP8 185 (rMMP8) into stress-susceptible mice (Fig. 3f), we showed that peripheral MMP8 can access the brain 186 parenchyma (Fig. 3g). 187 We then investigated whether CSDS affects the brain extracellular space using transmission electron 188 microscopy imaging of NAc tissue sections (Fig. 3h). SUS mice showed increased extracellular space  (Fig. 4a). Indeed, we found that the combination of rMMP8 and subthreshold stress led to a 201 lower SI ratio compared to unstressed mice (Fig. 4b). We then tested if rMMP8 also changed social 202 preference by testing mice with a non-threatening same-sex juvenile mouse using a social conditioned place 203 preference (sCPP) test (Fig. 4c). This paradigm has been used historically to assess social reward 59 . While 204 mice that received vehicle injections during three days of a subthreshold CSDS did form a preference for 205 the chamber that was previously paired with the juvenile mouse (Fig. 4d), the social preference was 206 abolished in mice that received rMMP8 (Fig. 4e). 207 To selectively deplete Mmp8 in peripheral leukocytes, we created chimeric mice that lack Mmp8 specifically 208 in peripheral leukocytes (Mmp8 -/-→WT) or wild-type controls (WT→WT) by bone marrow transplantation 209 (BMT) with hematopoietic stem cells from Mmp8 -/or Mmp8 +/+ (WT) donor mice. These chimeric mice 210 were then exposed to CSDS and underwent behavioural testing and assessment of the extracellular space in 211 NAc (Fig. 4f). We first validated the efficiency of the BMT experiment and found a complete depletion of 212 MMP8 in the blood of Mmp8 -/-→WT mice not only validating the efficiency of the BMT but confirming 213 that the source of MMP8 is indeed from peripheral leukocytes (Fig. 4g). We also found high chimerism 214 (~85-90%) and no differences in frequencies of peripheral monocytes, neutrophils, cytokines or chemokines 215 between the wild-type and knockout BMT mice (Extended Data Figs. 10a-i), suggesting that peripheral 216 depletion of MMP8 does lead to major changes in the peripheral immune system. Behaviourally, mice that 217 were transplanted with hematopoietic stem cells from Mmp8 -/mice showed less social avoidance from a 218 CD-1 mouse compared to stressed WT→WT mice, as measured by social interaction ratio and time the 219 experimental mouse spent in the corner (Figs. 4h-j). Similar effects were observed when mice were tested 220 for SI with a same-sex juvenile mouse (Fig. 4k). Interestingly, we did not observe any effects of MMP8 221 depletion on changes in other non-social stress-related behaviours such as sucrose preference test, splash 222 test or elevated plus maze (Extended Data Figs. 10j-l). We also did not observe differences in sickness-  Lastly, we sought to assess whether MMP8 is linked to changes in extracellular space and NAc 228 neurophysiology. Again, using transmission electron microscopy we assessed the volume of extracellular 229 space in stressed mice and found that Mmp8 -/-→WT mice had reduced extracellular space compared to 230 WT→WT mice (Fig. 4l). Given that ECM changes in general and MMP-mediated reorganization of the 231 ECM in particular have been previously associated with altered neuronal physiology 24,55,60,61 , we performed 232 ex vivo whole cell patch clamp recordings of medium spiny neurons (MSNs) from the NAc of Mmp8 -/and 233 WT mice (Fig. 4m). Previous studies have shown that CSDS leads to neurophysiological changes in NAc

239
In the current study, we have shown that a stress-induced increase in peripheral MMP8 leads to alterations 240 in the extracellular space of the NAc associated with altered NAc neurophysiology and social avoidance 241 (Extended Data Fig. 12). This study provides evidence for a novel mechanism by which the peripheral    Heat map of top ten cluster-defining protein coding genes. i, Gene ontology (GO) terms of significantly 500 (adjusted p-value < 0.05) upregulated genes of Cluster 0. j, Genes of GO terms "extracellular space" and 501 "extracellular matrix" compared between SUS vs. CON, RES vs. CON and SUS vs. RES mice. Feature 502 plots of normalized gene expression of Mmp8 in k, brain-infiltrating Ccr2 rfp+ monocytes and l, brain-resident 503 immune cells in the NAc. ** p < 0.01. Data are shown as means ± s.e.m. ANOVAs followed by Tukey's post hoc testing (for panels b, d, g, h, i, k, n, p) or two-sided Student's t-test 543 (for panel l). * p < 0.05, ** p < 0.01, *** p < 0.001. Data are shown as means ± s.e.m.   Subthreshold stress. Subthreshold stress is a variation of the CSDS paradigm that is used to unravel pro-575 susceptible factors 40 without eliciting behavioural alterations in unmanipulated mice. Experimental mice 576 were exposed to three 5 min periods of direct physical interactions with an aggressive CD-1 mouse with a 577 15 min interval between defeats. 24 h after the last defeat bout, SI test was conducted as described below.

578
Social interaction (SI) test. SI test was performed 24 h after the last defeat session under red light 579 conditions. After a 1 h habituation period to the behavioural suite, mice were placed into a Plexiglas arena 580 (42 cm × 42 cm × 42 cm, Nationwide Plastics) with a small meshed enclosure on one end. For the first 2.5 581 min, the experimental mouse freely explored the arena. The mouse was then removed from the arena which 582 was subsequently cleaned with 70 % ethanol, then, a novel social target (CD-1 for males and ERα-Cre for 583 female CSDS) was placed into the enclosure and the experimental mouse was placed back into the arena for 584 another 2.5 min. Locomotor activity was tracked and recorded using a Noldus Ethovision System (Noldus 585 Information Technology Inc, Version 11.0, Leesburg, VA). SI ratio was calculated as the ratio between the 586 time the experimental mouse spent in the vicinity of the enclosure (SI zone) when a target mouse was present 587 vs. absent. Mice with an SI ratio of ≥ 1 show a behavioural profile similar to unstressed control mice and 588 were termed resilient, while mice with an SI ratio < 1 were termed susceptible. To test social avoidance 589 behaviour towards a juvenile mouse, SI test was performed as described above with a 4-6 week-old male 590 juvenile mouse as a social target. Additional parameters that were measured were total locomotion and time 591 spent in corners, calculated as the sum between the two corners opposite the wire enclosure.

614
Social conditioned place preference (sCPP). sCPP was performed as previously described 70 . The 615 experiment was done under red light conditions after mice were habituated to the CPP room for 1 h. The 616 CPP chamber (Med Associates) consisted of 3 different compartments: a neutral middle part, and two 617 adjacent chambers, each with distinct floors (grid pattern) and walls. On the pre-test day, mice were allowed 618 to explore all three chambers for 20 min and the time spent in each chamber was recorded. Based on these 619 durations, mice were balanced to account for pre-test preferences. During the four consecutive conditioning 620 days, mice were conditioned twice per day: In the morning, mice were placed in one chamber for 15 min 621 with a novel, same-sex juvenile (4-5 week-old) C57BL/6J mouse (paired chamber). In the afternoon, the 622 experimental mouse was put in the empty opposite chamber for the same amount of time (unpaired 623 chamber). On the testing day, mice were again allowed to freely explore all chambers for 20 min and the 624 time spent in each chamber was automatically recorded (Med Associates).

625
Sucrose preference test. Sucrose preference test was performed to assess hedonic behaviour towards a 626 sweet gustatory stimulus 16 . Mice were given access to two water bottles (50 mL conical tubes with sipper 627 tops) for 24 h for habituation. Then, one water bottle was exchanged with a bottle containing 1% sucrose 628 (Sigma, #S0389) in drinking water. After 24 h, the bottle positions were swapped to prevent position bias.

631
Splash test. The splash test, a test performed to assess self-care behaviour, was conducted under red-light 632 conditions as described previously 16 . Briefly, after 1 h of habituation to the testing room, a 10% sucrose 633 solution was gently sprayed onto the lower back of the mouse. Behaviour was recorded for 5 min, and time 634 spent grooming was scored.

635
Elevated plus maze test (EPM). The EPM was conducted to assess anxiety-like behaviours 16 . After 1 h of 636 habituation to the testing room, mice were placed on an elevated cross-shaped maze for 5 min under red 637 light conditions. The four arms (two arms without and two arms with walls, each arm of the maze measuring Aldrich, #A9647) and 2 mM EDTA (Invitrogen, #AM9260G). Cells were pelleted and red blood cells 644 (RBCs) were lysed using RBC lysis buffers (BD, #555899). Immune cells of the brain were isolated as 645 previously described 31 . Briefly, mice were anaesthetized with 10% chloral hydrate and transcardially 646 perfused with ice-cold PBS (0.1 M). Brains were then cut into small pieces using scissors in a total of 3 mL 647 digestion buffer (RPMI (Thermo Fisher Scientific, #11875093) with 2% fetal bovine serum (Thermo Fisher   Table 5). PCR was performed as described above. cDNA was purified with GRCm38 using STAR (version 2.5) 78 . Reads were assigned to cells and unique molecular identifiers 737 (UMIs) 79 . The expression matrix for single cell data was processed using the package Seurat v3.1.5 in R 80 .

738
Features for which fewer than 3 cells were detected were removed, effectively excluding unexpressed 739 features. Cells having at least 1,000 and at most 4,000 features were retained. Cells with more than 5% of 740 reads mapping to mitochondrial genes were discarded. The NormalizeData function was used to log-741 normalize the dataset with a scale factor of 10,000. The top 2,000 most variable features across cells were 742 found using the function FindVariableFeatures. The ScaleData function was applied to scale the dataset.

743
The variable features were used to carry out dimensional reduction using PCA. The optimal number of 744 principal components to be used for dimensional reduction using Uniform Manifold Approximation and 745 Projection (UMAP) was determined using ElbowPlot. FindNeighbors and FindClusters functions were 746 utilized to construct a nearest neighbour graph and cluster cells in the dataset. UMAP was generated using 747 the function DimPlot. The FindAllMarkers function was applied to determine markers for clusters in the 748 UMAP plot. The FindMarkers function was used to carry out differential expression analysis for the three 749 experimental groups. Alexa Fluor™ 647, Thermo Fisher Scientific, #A-31573, 1:1,000) were incubated with the brains for 7 days 757 each at 37 °C. A LaVision lightsheet microscope with zoom body was used for sagittal half brain scanning 758 with dynamic focus and a step size of 4 µm. Brain images were processed as previously described using 759 ClearMap 39 . RFP + cells were quantified using the cell detection module, with cell detection parameters 760 optimized and validated based on the intensity and shape parameters of the signal. The autofluorescence 761 channel was aligned to the Allen Institute's Common Coordinate Framework using the Elastix toolbox.

762
Brain areas were collapsed into their parent regions prior to analyses. RFP cell counts from each region 763 were then correlated (Pearson correlation) with the SI score of defeated mice. Perceived Stress Scale 33 , a 10-item self-rating scale, was used to determine perceived stress levels. All 800 participants underwent biochemistry and hematological laboratory testing, urine toxicology and pregnancy 801 (if applicable) testing. At the time of enrollment, all participants were free of medications known to affect 802 the immune system for at least two weeks. Participants were free of active infections or systemic illness.

803
Subjects with concomitant unstable medical illnesses were excluded. Participants were free of current 804 substances of abuse. On the day of blood draw, patients were fasted for at least 6 h. Blood was drawn into 805 Vacutainer Gold Top 5 mL Silica Gel tubes (BD, #365968) for serum isolation and EDTA tubes (BD, 806 #365975) to assess complete blood count and differential count (Sysmex XN-9100™ Automated 807 Hematology System). For serum, blood was allowed to clot for > 30 min, then centrifuged at 1300 g for 15 808 min at 4 °C, then aliquoted and stored at -80 °C.  Microscopy Techniques, Woburn, MA). Images were then imported into Adobe Photoshop (Adobe, 2022) 836 and the extracellular space was manually scored using a computer tablet. Scoring was done by two 837 independent investigators blinded to experimental conditions. Images were then imported into ImageJ 838 (v1.53f51) 86 and the percentage of marked area/total area was calculated.  Fisher Scientific, #A39256). Biotinylated rMMP8 was separated from unbound biotin using Pierce™ C18 856 Spin Columns, 7K MWCO, (Thermo Fisher Scientific, #89870), which recovers proteins and 857 macromolecules larger than 7 kDa. Biotinylated rMMP8 was injected retro-orbitally into anaesthetized 858 mice. After 2 h of circulation, mice were euthanized and perfused with ice-cold PBS followed by 4% PFA.

859
Brain tissue processing and imaging was performed as described in the Immunohistochemistry and confocal  Table 2. Unless described otherwise, statistical analyses were performed with GraphPad Prism software 866 (Version 9, GraphPad Software Inc.) or SPSS version 24 (IBM Corp., SPSS Inc., Chicago IL, USA).

867
Outliers were identified using the Grubbs's test and excluded from statistical analyses. Level of statistical 868 significance was set at p < 0.05.