Gamma frequency entrainment attenuates amyloid load and modifies microglia

Abstract

Changes in gamma oscillations (20-50 Hz) have been observed in several neurological disorders. However, the relationship between gamma oscillations and cellular pathologies is unclear. Here we show reduced, behaviourally driven gamma oscillations before the onset of plaque formation or cognitive decline in a mouse model of Alzheimer's disease. Optogenetically driving fast-spiking parvalbumin-positive (FS-PV)-interneurons at gamma (40 Hz), but not other frequencies, reduces levels of amyloid-β (Aβ)1-40 and Aβ 1-42 isoforms. Gene expression profiling revealed induction of genes associated with morphological transformation of microglia, and histological analysis confirmed increased microglia co-localization with Aβ. Subsequently, we designed a non-invasive 40 Hz light-flickering regime that reduced Aβ1-40 and Aβ1-42 levels in the visual cortex of pre-depositing mice and mitigated plaque load in aged, depositing mice. Our findings uncover a previously unappreciated function of gamma rhythms in recruiting both neuronal and glial responses to attenuate Alzheimer's-disease-associated pathology.

Figures

Extended Data Figure 1. 5XFAD mice have reduced power in gamma during hippocampal SWRs

b) Mouse in virtual reality environment. c) Local field potential recorded in CA1, above, filtered for theta (left) or sharp wave ripples (right), middle, and gamma, below. d) Mean and standard deviation of the normalized power spectrum during theta. Each animal's power spectral density was normalized to its peak (n=6 mice per group). e) Normalized power spectral densities during theta periods in 3-month-old 5XFAD (green, n=6 mice) and WT (black/grey, n=6 mice) mice. Each animal's power spectral density was normalized to its peak (in theta). f) Average SWR-triggered spectrograms for one WT and one 5XFAD mouse shows an increase in the gamma band during SWRs. This increase is lower in the 5XFAD mouse than in the WT mouse (n=370 and 514 SWRs in WT and 5XFAD, respectively; WT mouse shown here is the same as in Fig. 1a). g) Distributions for each recording (left) and the mean and standard error across sessions (right) of instantaneous gamma frequencies during SWRs in 5XFAD (green) and WT (black) mice show distributions around 40 Hz (n=820, 800, 679, 38, 1875, 57 gamma cycles per session in 6 5XFAD animals and 181, 1075, 919, 1622, 51, 1860, 1903 gamma cycles per session in 6 WT animals). h) Cumulative distribution of the Z-scored gamma power during the 100 ms around the peak of the SWR for WT (black) and 5XFAD animals (green) for each animal (left) and the mean and standard error (shaded) across animals (right) (n=514, 358, 430, 22, 805, 37 SWRs per session in 6 5XFAD animals and 82, 311, 370, 776, 18, 710, 818 SWRs per session in 6 WT animals) i) Fraction of spikes in CA1 during SWRs as a function of the phase of gamma in 5XFAD (green) and WT (black) mice for each animal (left) and the mean and standard error across animals (right, n=2475, 1060, 3092, 25, 6521, 123 spikes during SWRs per session in 6 5XFAD mice and 360, 4741, 1564, 2961, 88, 3058, 4270 spikes during SWRs per session in 6 WT mice). j) SWR rate per non-theta period in 5XFAD (green) and WT (black) mice for each animal (left) and all animals combined (right, ranksum test, p -10, n=117, 210, 151, 55, 100, 1 non-theta periods per session in 6 5XFAD mice and 80, 68, 115, 95, 15, 159, 218 non-theta periods per session in 6 WT mice). k) The cumulative distribution of gamma power during large SWRs (detection threshold greater than 6 standard deviations above the mean, Methods) shows significantly smaller increases in 5XFAD (green) than WT (black) mice (ranksum test, p-5, n=1000 SWRs in 6 5XFAD mice and 1467 SWRs in 6 WT mice). l) Fraction of spikes as a function of the phase of gamma during large SWRs (detection threshold greater than 6 standard deviations above the mean, Methods), mean ± SEM (left) and histogram of the depth of modulation of spiking (right) as a function of gamma phase in 3-month-old 5XFAD (green, n=6 mice) and WT (black, n=6 mice) mice (ranksum test, bootstrap resampling p -10, n=2500 5XFAD spike-gamma phase distributions and 3000 WT distributions). m) Power spectral density during 40 Hz stimulation (red, left), random stimulation (blue, center), or no stimulation (black, right) of FS-PV-interneurons in CA1 for each mouse (n=4 5XFAD mice with 169, 130, 240, 73 40 Hz, 143, 129, 150, 72 random, and 278, 380, 52, 215 no stimulation periods per animal and n=3 WT mice with 65, 93, 91 40 Hz, 64, 93, 90 random, and 187, 276, 270 no stimulation periods per animal). n) Above: Example raw LFP trace (above) and the trace filtered for spikes (300-6000 Hz, below), with spikes indicated with red stars after optogenetic stimulation (blue vertical lines). Below: histogram of spikes per pulse after the onset of the 1 ms laser pulse during 40 Hz stimulation (red), random stimulation (blue), or no stimulation (black, n=345762 40 Hz, 301559 random pulses, and 32350 randomly selected no stimulation times at least 500 ms apart from 552 40 Hz, 543 random, and 1681 no stimulation periods in 4 5XFAD and 3 WT mice). o) Histogram of the difference in firing rates between 40 Hz stimulation and random stimulation periods shows that both types of stimulation elicit similar amounts of spiking activity (Wilcoxon signed rank test for zero median, p>0.6, n=538 stimulation periods from 4 5XFAD and 3 WT mice, n.s. indicates not significant). p) Multiunit firing rates per 40 Hz stimulation (red), random stimulation (blue), and no stimulation (black) period for each animal. Box and whisker plots show median (white lines in box) and quartiles (top and bottom of box). In all animals firing rates between 40 Hz and random stimulation were not significantly different, showing that the random stimulation condition serves as a control for spiking activity (ranksum tests for each animal, 3 WT and 4 5XFAD mice, p's>0.09, n=87, 130, 8, 65, 93, 91, 73 40 Hz stimulation periods and 85, 129, 5, 64, 93, 90, 72 random stimulation periods per animal). We also examined whether 40 Hz stimulation caused neuronal hyperactivity relative to no stimulation, because according to a recent report, this could have negative effects on neural circuit function. In most animals the firing rates between 40 Hz or random stimulation and no stimulation were not significantly different (ranksum tests for each animal, 2 WT and 2 5XFAD, p's>0.25, n=8, 93, 91, 73 40 Hz stimulation periods and 15, 277, 270, 215 baseline periods per animal) or the firing rates during 40 Hz or random stimulation were lower than during no stimulation (ranksum tests for each animal, 1 WT and 1 5XFAD, p's-5, which is significant when corrected for performing multiple comparisons, n=130, 65 40 Hz stimulation periods and 379, 187 baseline periods per animal) indicating that 40 Hz stimulation did not cause neuronal hyperactivity. In one animal there was significantly more activity with 40 Hz or random stimulation than during baseline (ranksum test for 1 5XFAD, mouse, p<10-5, n=87 40 Hz stimulation periods and 251 baseline periods per animal). Therefore in six out of seven animals we see no evidence that the 40 Hz optogenetic stimulation of FS-PV-interneurons causes hyperactivity.

Extended Data Figure 2. ChR2 was expressed in FS-PV-interneurons for optogenetic stimulation

a) AAV-DIO-ChR2-EYFP or AAV-DIO-EYFP drives Cre-dependent expression of ChR2-EYFP or EYFP to produce celltype-specific targeting of ChR2 or EYFP, respectively. In the presence of Cre, ChR2-EYFP or EYFP is inverted into the sense direction and expressed from the EF-1α promoter in PV-positive cells. ITR, inverted terminal repeat; polyA; WPRE, woodchuck hepatitis B virus post-transcriptional element. b) ChR2-EYFP was strongly expressed in PV-positive interneurons in CA1 of 3-month-old 5XFAD/PV-Cre mice (scale bar = 100 μm). c) Immunohistochemistry with anti-EYFP and anti-PV antibodies in CA1 of 3-month-old 5XFAD/PV-Cre mice expressing AAV-DIO-ChR2-EYFP shows EYFP expression only in PV-positive cells (scale bar = 50 μm). d) Representative western blots showing levels of full-length APP (top left, CT695), APP CTFs (bottom left, CT695), APP NTFs (top right, A8967) and β-actin (bottom right, A5316, loading control) in CA1 in EYFP, random, and 40 Hz stimulation conditions, 1 mouse per lane, with 2 biological replicates of each condition. e) Immunohistochemistry with anti-A β (12F4, red) antibodies in CA1 of 5XFAD/PV-Cre mice expressing only EYFP or ChR2 with 40 Hz, and random stimulation conditions (scale bar = 50 μm). f) Bar graphs represent the relative immunoreactivity of A β normalized to EYFP (n=4 mice per group; * indicates p<0.05 and *** indicates p<0.001 by one-way ANOVA). Bar graphs show mean + SEM. g) Immunohistochemistry with anti-Rab5 antibody (ADI-KAP-GP006-E, green) in CA1 of 5XFAD/PV-Cre mice (scale bar = 50 μm). h) Relative Rab5 intensity levels normalized to EYFP controls (n=3 mice per group).

Extended Data Figure 3. Optogenetically driven 40 Hz oscillations in CA1 cause changes in gene regulation and immediate early gene expression

a) Table of 130 genes up-regulated by 40 Hz FS-PV-interneuron stimulation determined by whole transcriptome RNA-Seq of CA1 from 3-month-old 5XFAD/PV-Cre mice (p52 b) Table of 393 genes down-regulated by 40 Hz FS-PV-interneuron stimulation determined by whole transcriptome RNA-Seq of CA1 from 3-month-old 5XFAD/PV-Cre mice (p52 c) Box plot showing fragments per kilobase (FPKM) values of up- and down-regulated genes in EYFP and 40 Hz groups. Box shows median (black lines in box) and quartiles (top and bottom of box), whiskers show minimum and maximum values, and circles show outliers. d) GSEA statistics tables showing statistical significance of correlation between genes up- or down- regulated by 40 Hz stimulation and publicly available neuron, microglia, and macrophage specific RNA-Seq data under different chemical and genetic perturbations; the perturbation terms were ranked based on the FDR q-values for the up-regulated gene list, from the smallest to the largest (Methods). e) RT-qPCR verification of specific gene targets in the RNA-Seq data set. Bar graph shows relative RNA levels (fold change) from EYFP (black) and 40 Hz stimulation (red) conditions (* indicates p

Extended Data Figure 4. 40 Hz light flicker drives 40 Hz oscillations in VC, while random flickering does not

a) Power spectral densities of local field potentials in VC during 40 Hz light flicker (red, far left), random light flicker (blue, center left), dark (black, center right), or light (green, far right) in VC for each recording session for each mouse (n=5 recordings from 4 5XFAD mice with 47, 51, 64, 49, 16 40 Hz flicker, 47, 50, 64, 50, 16 random flicker, 279, 301, 382, 294, 93 dark and 47, 50, 64, 49, 15 light periods). Light flicker at other frequencies increased power in the flicker frequency, as others have found previously, (data not shown). b) Histogram of the difference in firing rates between 40 Hz light flicker and random light flicker (n=226 stimulation periods from 5 recording sessions in 4 5XFAD mice). c) Multiunit firing rates in VC during 40 Hz light flicker (red), random light flicker (blue), dark (black), or light (green) periods. Box plots show median (white lines in box) and quartiles (top and bottom of box). In all animals, firing rates between 40 Hz flicker and random flicker conditions were not significantly different showing that the random stimulation condition serves as a control for spiking activity (ranksum tests for each of 5 recording session from 4 5XFAD mice, p's>0.06, n=47, 51, 64, 49, 16 40 Hz flicker periods and 47, 50, 64, 50, 16 random flicker periods per recording). There were no significant differences in firing rates between 40 Hz flicker and light conditions indicating that 40 Hz light flicker generally did not cause neuronal hyperexcitability (ranksum tests for each of 5 recording session from 4 5XFAD mice, p's > 0.2 for 4 recording sessions, p

Extended Data Figure 5. 40 Hz light flicker does not affect Aβ levels in hippocampus or barrel cortex

a) Example local field potential trace in hippocampal CA1 before and during 40 Hz light flicker (above). Mean (solid line) and standard deviation (shaded area) of power spectral density during 40 Hz light flicker (red), random light flicker (blue), or dark (black) in CA1 (n=2 5XFAD and 3 WT mice). b) Histogram of the fraction of spikes in hippocampus as a function of time for 4 cycles of 40 Hz light flicker (left, red) or the equivalent period of time for random light flicker (right, blue, n=2 5XFAD and 3 WT mice, mean ± SEM across animals). Bar above indicates when light was on (yellow) or off (black). For random stimulation, spiking was aligned to the start of the light turning on, additional periods with light-on occurred at random intervals indicated by grey (Methods). c) Histogram of the difference in firing rates between 40 Hz light flicker and random light flicker (bottom n=168 stimulation periods from 5 recording sessions in 2 5XFAD and 3 WT mice). d) Power spectral densities of local field potentials in CA1 during 40 Hz light flicker (red, far left), random light flicker (blue, center left), dark (black, center right), or light (green, far right) for each recording session for each mouse (n=5 recordings from 2 5XFAD and 3 WT mice with 22, 54, 42, 71, 55, 40 Hz flicker, 12, 34, 32, 54, 36 random flicker, 115, 240, 224, 342, 282 dark and 12, 33, 33, 54, 35 light periods). e) Multiunit firing rates in CA1 during 40 Hz light flicker (red), random light flicker (blue), dark (black), or light (green) periods. Box plots show median (white lines in box) and quartiles (top and bottom of box). In all animals firing rates between 40 Hz flicker and random flicker conditions were not significantly different showing that the random stimulation condition serves as a control for spiking activity (ranksum tests for each of 5 recordings from 2 5XFAD and 3 WT animals, p's>0.2, n=22, 54, 42, 71, 55 40 Hz flicker periods and 12, 34, 32, 54, 36 random flicker periods per recording). There were no significant differences in firing rates between 40 Hz flicker and light conditions indicating that 40 Hz light flicker generally did not cause neuronal hyperexcitability (ranksum tests for each of 5 recordings from 2 5XFAD and 3 WT animals, p's > 0.3, n=22, 54, 42, 71, 55 40 Hz periods and 12, 34, 33, 54, 35 light periods per recording). f) Bar graphs of relative A β1-40 levels in VC of 5XFAD mice in dark, 40 Hz flicker, and random flicker conditions, normalized to dark (n=4 mice per group; n.s. indicates not significant). Bar graphs represent mean + SEM. Circles superimposed on bars in bar graphs indicate individual data points in each group. g) Bar graphs of relative A β1-42 levels in VC of 5XFAD mice in dark, 40 Hz flicker, and random flicker conditions, normalized to dark (n=4 mice per group; n.s. indicates not significant). Bar graphs represent mean + SEM. Circles superimposed on bars in bar graphs indicate individual data points in each group. h) Bar graph of relative A β1-40 and A β1-42 levels in barrel cortex of 5XFAD mice in dark and 40 Hz flicker conditions, normalized to dark (n=3 mice per group; n.s. indicates not significant by Student's t-test).

Extended Data Figure 6. Acute reduction in A β after light flicker in APP/PS1 and WT mice and at various time points

a) Bar graph of relative A β1-40 and A β1-42 levels of APP/PS1 in VC in dark and 40 Hz flicker conditions, normalized to dark (n=5 mice per group for dark and n=4 mice per group for 40 Hz flicker conditions; n.s. indicates not significant and * indicates p<0.05, by Student's t-test). All bar graphs show mean + SEM throughout this figure. Circles superimposed on bars in bar graphs indicate individual data points in each group. b) Bar graph of relative mouse A β1-40 and Aβ1-42 levels in VC of 9-month-old WT mice in dark and 40 Hz flicker conditions, normalized to dark (n=11 mice per group for dark and n=9 mice per group for 40 Hz flicker conditions; * indicates p<0.05, by Student's t-test).

Extended Data Figure 7. 40 Hz light flicker does not decrease synaptic density in VC

a) Schematic depicting isolation of microglia from VC. VC was dissected, then single cells were suspended and labeled with CD11b and CD45 antibodies. Subsequently, cells were sorted via fluorescence-activated cell sorting (FACS) and lysed. A β1-40 levels were analyzed by ELISA. b) Bar graph of A β1-40 levels in microglia purified using FACS (Methods) from VC of 3-month-old 5XFAD and WT mice (n=8 mice per group for 5XFAD and n=4 mice per group for WT mice; * indicates p<0.05 by Student's t-test). Circles superimposed on bars in bar graphs indicate individual data points in each group. c) Immunohistochemistry with SVP38 (red) antibodies to detect synaptophysin in VC of3-month-old 5XFAD mice in dark and 40 Hz flicker conditions (Images were taken with 40x objective; scale bar = 50 μm). Right: 100X rendering of dark and 40 Hz flicker conditions. d) Bar graph of relative SVP38 intensity levels in VC of 5XFAD mice after dark (black) and 40 Hz (red) flicker conditions, normalized to dark (n=4 mice per group; n.s. indicates not significant, by Student's t-test).

Extended Data Figure 8. Aβ levels in VC return to baseline 24 hr after 1 hr of 40 Hz light flicker

a) Bar graph of relative Aβ1-40 and Aβ1-42 levels in VC of 5XFAD mice 1, 4, 12, and 24 hours after 1 hour of dark or 40 Hz flicker treatment, normalized to dark (n=4 mice per group for 4 and 12 hr wait, n=6 for 1 and 24 hr wait, n=12 for dark; n.s. indicates not significant, * indicates p<0.05 and ** indicates p<0.01, by one-way ANOVA).

Extended Data Figure 9. Driving 40 Hz oscillations in VC via light flicker reduces phosphorylated tau in a tauopathy mouse model

a) Immunohistochemistry with anti-pTau (S202, green) and anti-MAP2 (red) antibodies in VC of 4-month-old P301S mice after 7 days of 1 hr/day dark or 40 Hz flicker conditions (Images were taken with 40× objective; scale bar = 50 μm). Insets: 100X rendering of representative cell body in dark and 40 Hz flicker conditions. No changes were observed by western blot (Data not shown). b) Bar graph of relative pTau (S202) intensity levels in P301S mice after 7 days of 1 hr/day dark (black) and 40 Hz flicker (red) conditions (n=8 mice per group; * indicates p

Figure 1. 5XFAD mice have reduced power in gamma during hippocampal SWRs

a) Average SWR-triggered spectrograms for one mouse (left) showing gamma (yellow arrow) during SWRs (red arrow);right: frequencies below 80 Hz enlarged (n=370 SWRs). b) Histogram of instantaneous gamma frequencies during SWRs for mouse in d. a) Above: Z-scored gamma power around SWR peak for one WT and one 5XFAD mouse (mean ± SEM). Below: Cumulative distribution of gamma power during SWRs (ranksum test, n=2166 and 3085 SWRs in 6 5XFAD and WT mice, respectively). c) Above: Fraction of spikes during SWRs as a function of gamma phase (mean ± SEM). Below: Depth of gamma spiking modulation during SWRs. (ranksum test, bootstrap resampling, n=2500 5XFAD and 3000 WT phase distributions). d) Above: Local field potential trace before and during 40 Hz optogenetic stimulation.Below: Mean and standard deviation of power spectral density (n=4 5XFAD and 3 WT mice). e) Relative A β1-40 levels in CA1 of 5XFAD/PV-Cre mice in each stimulation condition normalized to EYFP controls (n=8 EYFP, n=7 40 Hz, n=4 8 Hz n=6 random mice). f) As in i for A β1-42(n=4 EYFP, n=4 40 Hz, n=3 8 Hz n=3 random mice). g) Relative A β1-40 levels in CA1 of 5XFAD/αCamKII-Cre mice in each stimulation condition normalized to EYFP controls (n=6 40 Hz, n=3 8 Hz n=3 random mice). h) As in k for A β1-42(n=3 mice per group). n.s. not significant, * p<0.05, ** p<0.01, *** p<0.001 by one-way ANOVA; circles indicate n, mean+SEM in bar graphs.

Figure 2. Driving 40 Hz oscillations optogenetically in hippocampus reduces A β in 5XFAD mice

a) Representative western blot showing levels of APP (CT695), APP NTFs (A8967), APP CTFs (CT695), and β-Actin (A5316, loading control) in CA1 of 5XFAD/PV-Cre mice expressing only EYFP or ChR2 with 40 Hz, or random stimulation conditions. 1 mouse per lane, 2 biological replicates. b) Relative immunoreactivity of full-length APP normalized to actin (for b-d, n=6 mice per group). c) Relative immunoreactivity of APP NTF normalized to actin. d) Relative immunoreactivity of APP CTFs normalized to actin. e) Immunohistochemistry with anti-A β (D54D2, green) and anti-EEA1 (610457, red) antibodies in CA1 of 5XFAD/PV-Cre mice (scale bar = 50 μm). f) Relative immunoreactivity of A β normalized to EYFP controls(for f, g, n=3 mice per group). g) Relative immunoreactivity of EEA1 normalized to EYFP controls. n.s. not significant, * p<0.05, ** p<0.01, by one-way ANOVA; mean + SEM in bar graphs.

Figure 3. Driving 40 Hz oscillations optogenetically in hippocampus causes a distinct morphological transformation of microglia in 5XFAD mice

a) Heat map of differentially expressed genes determined by whole-transcriptome RNA-Seq of CA1 from 5XFAD/PV-Cre mice expressing only EYFP or ChR2 with 40 Hz stimulation. Normalized z-score values (high: red, low: blue) were calculated for each differentially expressed gene (row). b) Cell-type-specific expression patterns of up-regulated genes following 40 Hz stimulation (MO: myelinating oligodendrocyte, OPC: oligodendrocyte progenitor cell, NFO: newly formed oligodendrocyte). c) RT-qPCR of specific up-regulated genes:relative RNA levels (fold change) in CA1 of 5XFAD/PV-Cre expressing only EYFP or ChR2 with 40 Hz stimulation, normalized to EYFP controls (Student's t-test; n=6 mice per group). d) Immunohistochemistry with anti-Iba1 (019-19741, green) to identify microglia and anti-A β(12F4, red) antibodies in CA1 of 5XFAD/PV-Cre mice expressing only EYFP or ChR2 with 40 Hz, and random stimulation (40× objective; scale bar = 50 μm). e) Number of Iba1-positive microglia (for f-I,one-way ANOVA; n=4 mice per group). f) Diameter of Iba1-positive microglia cell bodies. g) Average length of Iba1-positive microglia primary processes. h) Percent of Iba1-positive microglia cell bodies that are also A β-positive. n.s. not significant, * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001; mean + SEM in bar graphs.

Figure 4. Driving 40 Hz oscillations in VC via light flicker reduces A β and amyloid plaques in 5XFAD mice

a) Local field potential trace in VC before and during 40 Hz light flicker (above). Power spectral density mean and standard deviation (below, n=4 5XFAD mice, 5 recording sessions). b) Fraction of spikes in VC over 4 cycles of 40 Hz flicker (left) or the equivalent time for random flicker (right, n=4 5XFAD mice from 5 recording sessions, mean ± SEM across animals). For random stimulation, spiking was aligned to light turning on, grey indicates additional light-on flickers occurring randomly (Methods). c) Relative A β1-40 (left) and A β1-42 (right) levels normalized to dark, in VC of 5XFAD mice exposed to dark, light, 40 Hz, 20 Hz, 80 Hz, 40 Hz with picrotoxin (PTX), and random conditions (n=12 dark; n=6 light, 40 Hz, 20 Hz, 80 Hz flicker and PTX; n=4 random mice; one-way ANOVA). d) Immunohistochemistry with anti-Iba1 (019-19741, green) and anti-A β (12F4, red)antibodies in VC of 5XFAD mice exposed to dark or 40 Hz flicker. Right: 120X zoom; arrows indicate +Iba1/+Aβ signal in cell body(scale bar=50 μm). e) Number of Iba1-positive microglia(for e-h Student's t-test unpaired, n=4 mice per group) f) Diameter of Iba1-positive microglia cell bodies. g) Average length of Iba1-positive microglia primary processes. h) Percent of Iba1-positive microglia cell bodies that are also A β-positive. i) Relative A β1-40 levels in VC of 6-month-old 5XFAD mice after 7 days of 1 hr/day dark or 40 Hz flicker (Student's t-test unpaired; n=13 mice per group). j) As in i forA β1-42. k) Immunohistochemistry with anti-A β(D5452, green) antibody in 6-month-old VC of 5XFAD mice after 7 days of 1 hr/day dark or 40 Hz flicker showing plaques (white arrows; scale bar=50 μm). i) Number of A β-positive plaques; (for l,m Student's t-test unpaired, n=8 mice per group). m) Area of A β-positive plaques. n.s. not significant, * p<0.05, ** p<0.01, *** p<0.001; circles indicate n, mean + SEM in bar graphs.